r600g: add support for subsampled rgb formats
[mesa.git] / src / gallium / drivers / r600 / r600_texture.c
index 1c6f39adee8686e2ec3db57fc66e71b26085756a..ce66117ed47eaf5bf4ce5acaac8769ae4bbccdb7 100644 (file)
  *      Jerome Glisse
  *      Corbin Simpson
  */
-#include <errno.h>
-#include <pipe/p_screen.h>
-#include <util/u_format.h>
-#include <util/u_format_s3tc.h>
-#include <util/u_math.h>
-#include <util/u_inlines.h>
-#include <util/u_memory.h>
-#include "pipebuffer/pb_buffer.h"
-#include "r600_pipe.h"
-#include "r600_resource.h"
-#include "r600d.h"
 #include "r600_formats.h"
+#include "r600d.h"
+
+#include <errno.h>
+#include "util/u_format_s3tc.h"
+#include "util/u_memory.h"
 
 /* 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)
@@ -43,7 +37,7 @@ static void r600_copy_to_staging_texture(struct pipe_context *ctx, struct r600_t
        struct pipe_transfer *transfer = (struct pipe_transfer*)rtransfer;
        struct pipe_resource *texture = transfer->resource;
 
-       ctx->resource_copy_region(ctx, rtransfer->staging_texture,
+       ctx->resource_copy_region(ctx, &rtransfer->staging->b.b.b,
                                0, 0, 0, 0, texture, transfer->level,
                                &transfer->box);
 }
@@ -63,10 +57,8 @@ static void r600_copy_from_staging_texture(struct pipe_context *ctx, struct r600
        sbox.depth = 1;
        ctx->resource_copy_region(ctx, texture, transfer->level,
                                  transfer->box.x, transfer->box.y, transfer->box.z,
-                                 rtransfer->staging_texture,
+                                 &rtransfer->staging->b.b.b,
                                  0, &sbox);
-
-       r600_flush(ctx, NULL, RADEON_FLUSH_ASYNC);
 }
 
 unsigned r600_texture_get_offset(struct r600_resource_texture *rtex,
@@ -93,19 +85,19 @@ static unsigned r600_get_block_alignment(struct pipe_screen *screen,
        switch(array_mode) {
        case V_038000_ARRAY_1D_TILED_THIN1:
                p_align = MAX2(8,
-                              ((rscreen->tiling_info->group_bytes / 8 / pixsize)));
+                              ((rscreen->tiling_info.group_bytes / 8 / pixsize)));
                break;
        case V_038000_ARRAY_2D_TILED_THIN1:
-               p_align = MAX2(rscreen->tiling_info->num_banks,
-                              (((rscreen->tiling_info->group_bytes / 8 / pixsize)) *
-                               rscreen->tiling_info->num_banks)) * 8;
+               p_align = MAX2(rscreen->tiling_info.num_banks,
+                              (((rscreen->tiling_info.group_bytes / 8 / pixsize)) *
+                               rscreen->tiling_info.num_banks)) * 8;
                break;
        case V_038000_ARRAY_LINEAR_ALIGNED:
-               p_align = MAX2(64, rscreen->tiling_info->group_bytes / pixsize);
+               p_align = MAX2(64, rscreen->tiling_info.group_bytes / pixsize);
                break;
        case V_038000_ARRAY_LINEAR_GENERAL:
        default:
-               p_align = rscreen->tiling_info->group_bytes / pixsize;
+               p_align = rscreen->tiling_info.group_bytes / pixsize;
                break;
        }
        return p_align;
@@ -119,7 +111,7 @@ static unsigned r600_get_height_alignment(struct pipe_screen *screen,
 
        switch (array_mode) {
        case V_038000_ARRAY_2D_TILED_THIN1:
-               h_align = rscreen->tiling_info->num_channels * 8;
+               h_align = rscreen->tiling_info.num_channels * 8;
                break;
        case V_038000_ARRAY_1D_TILED_THIN1:
        case V_038000_ARRAY_LINEAR_ALIGNED:
@@ -145,14 +137,14 @@ static unsigned r600_get_base_alignment(struct pipe_screen *screen,
 
        switch (array_mode) {
        case V_038000_ARRAY_2D_TILED_THIN1:
-               b_align = MAX2(rscreen->tiling_info->num_banks * rscreen->tiling_info->num_channels * 8 * 8 * pixsize,
+               b_align = MAX2(rscreen->tiling_info.num_banks * rscreen->tiling_info.num_channels * 8 * 8 * pixsize,
                               p_align * pixsize * h_align);
                break;
        case V_038000_ARRAY_1D_TILED_THIN1:
        case V_038000_ARRAY_LINEAR_ALIGNED:
        case V_038000_ARRAY_LINEAR_GENERAL:
        default:
-               b_align = rscreen->tiling_info->group_bytes;
+               b_align = rscreen->tiling_info.group_bytes;
                break;
        }
        return b_align;
@@ -198,6 +190,16 @@ static unsigned r600_texture_get_nblocksy(struct pipe_screen *screen,
        height = util_format_get_nblocksy(rtex->real_format, height);
        tile_height = r600_get_height_alignment(screen,
                                                rtex->array_mode[level]);
+
+       /* XXX Hack around an alignment issue. Less tests fail with this.
+        *
+        * The thing is depth-stencil buffers should be tiled, i.e.
+        * the alignment should be >=8. If I make them tiled, stencil starts
+        * working because it no longer overlaps with the depth buffer
+        * in memory, but texturing like drawpix-stencil breaks. */
+       if (util_format_is_depth_or_stencil(rtex->real_format) && tile_height < 8)
+               tile_height = 8;
+
        height = align(height, tile_height);
        return height;
 }
@@ -233,13 +235,137 @@ static void r600_texture_set_array_mode(struct pipe_screen *screen,
        }
 }
 
+static int r600_init_surface(struct radeon_surface *surface,
+                            const struct pipe_resource *ptex,
+                            unsigned array_mode)
+{
+       surface->npix_x = ptex->width0;
+       surface->npix_y = ptex->height0;
+       surface->npix_z = ptex->depth0;
+       surface->blk_w = util_format_get_blockwidth(ptex->format);
+       surface->blk_h = util_format_get_blockheight(ptex->format);
+       surface->blk_d = 1;
+       surface->array_size = 1;
+       surface->last_level = ptex->last_level;
+       surface->bpe = util_format_get_blocksize(ptex->format);
+       /* align byte per element on dword */
+       if (surface->bpe == 3) {
+               surface->bpe = 4;
+       }
+       surface->nsamples = 1;
+       surface->flags = 0;
+       switch (array_mode) {
+       case V_038000_ARRAY_1D_TILED_THIN1:
+               surface->flags |= RADEON_SURF_SET(RADEON_SURF_MODE_1D, MODE);
+               break;
+       case V_038000_ARRAY_2D_TILED_THIN1:
+               surface->flags |= RADEON_SURF_SET(RADEON_SURF_MODE_2D, MODE);
+               break;
+       case V_038000_ARRAY_LINEAR_ALIGNED:
+               surface->flags |= RADEON_SURF_SET(RADEON_SURF_MODE_LINEAR_ALIGNED, MODE);
+               break;
+       case V_038000_ARRAY_LINEAR_GENERAL:
+       default:
+               surface->flags |= RADEON_SURF_SET(RADEON_SURF_MODE_LINEAR, MODE);
+               break;
+       }
+       switch (ptex->target) {
+       case PIPE_TEXTURE_1D:
+               surface->flags |= RADEON_SURF_SET(RADEON_SURF_TYPE_1D, TYPE);
+               break;
+       case PIPE_TEXTURE_RECT:
+       case PIPE_TEXTURE_2D:
+               surface->flags |= RADEON_SURF_SET(RADEON_SURF_TYPE_2D, TYPE);
+               break;
+       case PIPE_TEXTURE_3D:
+               surface->flags |= RADEON_SURF_SET(RADEON_SURF_TYPE_3D, TYPE);
+               break;
+       case PIPE_TEXTURE_1D_ARRAY:
+               surface->flags |= RADEON_SURF_SET(RADEON_SURF_TYPE_1D_ARRAY, TYPE);
+               surface->array_size = ptex->array_size;
+               break;
+       case PIPE_TEXTURE_2D_ARRAY:
+               surface->flags |= RADEON_SURF_SET(RADEON_SURF_TYPE_2D_ARRAY, TYPE);
+               surface->array_size = ptex->array_size;
+               break;
+       case PIPE_TEXTURE_CUBE:
+               surface->flags |= RADEON_SURF_SET(RADEON_SURF_TYPE_CUBEMAP, TYPE);
+               break;
+       case PIPE_BUFFER:
+       default:
+               return -EINVAL;
+       }
+       if (ptex->bind & PIPE_BIND_SCANOUT) {
+               surface->flags |= RADEON_SURF_SCANOUT;
+       }
+       if (util_format_is_depth_and_stencil(ptex->format)) {
+               surface->flags |= RADEON_SURF_ZBUFFER;
+               surface->flags |= RADEON_SURF_SBUFFER;
+       }
+
+       return 0;
+}
+
+static int r600_setup_surface(struct pipe_screen *screen,
+                             struct r600_resource_texture *rtex,
+                             unsigned array_mode,
+                             unsigned pitch_in_bytes_override)
+{
+       struct pipe_resource *ptex = &rtex->resource.b.b.b;
+       struct r600_screen *rscreen = (struct r600_screen*)screen;
+       unsigned i;
+       int r;
+
+       if (util_format_is_depth_or_stencil(rtex->real_format)) {
+               rtex->surface.flags |= RADEON_SURF_ZBUFFER;
+               rtex->surface.flags |= RADEON_SURF_SBUFFER;
+       }
+
+       r = rscreen->ws->surface_init(rscreen->ws, &rtex->surface);
+       if (r) {
+               return r;
+       }
+       rtex->size = rtex->surface.bo_size;
+       if (pitch_in_bytes_override && pitch_in_bytes_override != rtex->surface.level[0].pitch_bytes) {
+               /* old ddx on evergreen over estimate alignment for 1d, only 1 level
+                * for those
+                */
+               rtex->surface.level[0].nblk_x = pitch_in_bytes_override / rtex->surface.bpe;
+               rtex->surface.level[0].pitch_bytes = pitch_in_bytes_override;
+               rtex->surface.level[0].slice_size = pitch_in_bytes_override * rtex->surface.level[0].nblk_y;
+               if (rtex->surface.flags & RADEON_SURF_SBUFFER) {
+                       rtex->surface.stencil_offset = rtex->surface.level[0].slice_size;
+               }
+       }
+       for (i = 0; i <= ptex->last_level; i++) {
+               rtex->offset[i] = rtex->surface.level[i].offset;
+               rtex->layer_size[i] = rtex->surface.level[i].slice_size;
+               rtex->pitch_in_bytes[i] = rtex->surface.level[i].pitch_bytes;
+               switch (rtex->surface.level[i].mode) {
+               case RADEON_SURF_MODE_LINEAR_ALIGNED:
+                       rtex->array_mode[i] = V_038000_ARRAY_LINEAR_ALIGNED;
+                       break;
+               case RADEON_SURF_MODE_1D:
+                       rtex->array_mode[i] = V_038000_ARRAY_1D_TILED_THIN1;
+                       break;
+               case RADEON_SURF_MODE_2D:
+                       rtex->array_mode[i] = V_038000_ARRAY_2D_TILED_THIN1;
+                       break;
+               default:
+               case RADEON_SURF_MODE_LINEAR:
+                       rtex->array_mode[i] = 0;
+                       break;
+               }
+       }
+       return 0;
+}
+
 static void r600_setup_miptree(struct pipe_screen *screen,
                               struct r600_resource_texture *rtex,
                               unsigned array_mode)
 {
        struct pipe_resource *ptex = &rtex->resource.b.b.b;
-       struct radeon *radeon = ((struct r600_screen*)screen)->radeon;
-       enum chip_class chipc = r600_get_family_class(radeon);
+       enum chip_class chipc = ((struct r600_screen*)screen)->chip_class;
        unsigned size, layer_size, i, offset;
        unsigned nblocksx, nblocksy;
 
@@ -252,7 +378,11 @@ static void r600_setup_miptree(struct pipe_screen *screen,
                nblocksx = r600_texture_get_nblocksx(screen, rtex, i);
                nblocksy = r600_texture_get_nblocksy(screen, rtex, i);
 
-               layer_size = nblocksx * nblocksy * blocksize;
+               if (chipc >= EVERGREEN && array_mode == V_038000_ARRAY_LINEAR_GENERAL)
+                       layer_size = align(nblocksx, 64) * nblocksy * blocksize;
+               else
+                       layer_size = nblocksx * nblocksy * blocksize;
+
                if (ptex->target == PIPE_TEXTURE_CUBE) {
                        if (chipc >= R700)
                                size = layer_size * 8;
@@ -293,7 +423,7 @@ static boolean permit_hardware_blit(struct pipe_screen *screen,
        /* hackaround for S3TC */
        if (util_format_is_compressed(res->format))
                return TRUE;
-           
+
        if (!screen->is_format_supported(screen,
                                res->format,
                                res->target,
@@ -308,14 +438,7 @@ static boolean permit_hardware_blit(struct pipe_screen *screen,
                                 PIPE_BIND_SAMPLER_VIEW))
                return FALSE;
 
-       switch (res->usage) {
-       case PIPE_USAGE_STREAM:
-       case PIPE_USAGE_STAGING:
-               return FALSE;
-
-       default:
-               return TRUE;
-       }
+       return TRUE;
 }
 
 static boolean r600_texture_get_handle(struct pipe_screen* screen,
@@ -324,10 +447,10 @@ static boolean r600_texture_get_handle(struct pipe_screen* screen,
 {
        struct r600_resource_texture *rtex = (struct r600_resource_texture*)ptex;
        struct r600_resource *resource = &rtex->resource;
-       struct radeon *radeon = ((struct r600_screen*)screen)->radeon;
+       struct r600_screen *rscreen = (struct r600_screen*)screen;
 
-       return r600_bo_get_winsys_handle(radeon, resource->bo,
-                       rtex->pitch_in_bytes[0], whandle);
+       return rscreen->ws->buffer_get_handle(resource->buf,
+                                             rtex->pitch_in_bytes[0], whandle);
 }
 
 static void r600_texture_destroy(struct pipe_screen *screen,
@@ -339,9 +462,10 @@ static void r600_texture_destroy(struct pipe_screen *screen,
        if (rtex->flushed_depth_texture)
                pipe_resource_reference((struct pipe_resource **)&rtex->flushed_depth_texture, NULL);
 
-       if (resource->bo) {
-               r600_bo_reference(&resource->bo, NULL);
-       }
+       if (rtex->stencil)
+               pipe_resource_reference((struct pipe_resource **)&rtex->stencil, NULL);
+
+       pb_reference(&resource->buf, NULL);
        FREE(rtex);
 }
 
@@ -352,9 +476,9 @@ static const struct u_resource_vtbl r600_texture_vtbl =
        r600_texture_get_transfer,      /* get_transfer */
        r600_texture_transfer_destroy,  /* transfer_destroy */
        r600_texture_transfer_map,      /* transfer_map */
-       u_default_transfer_flush_region,/* transfer_flush_region */
+       NULL,                           /* transfer_flush_region */
        r600_texture_transfer_unmap,    /* transfer_unmap */
-       u_default_transfer_inline_write /* transfer_inline_write */
+       NULL                            /* transfer_inline_write */
 };
 
 static struct r600_resource_texture *
@@ -363,11 +487,14 @@ r600_texture_create_object(struct pipe_screen *screen,
                           unsigned array_mode,
                           unsigned pitch_in_bytes_override,
                           unsigned max_buffer_size,
-                          struct r600_bo *bo)
+                          struct pb_buffer *buf,
+                          boolean alloc_bo,
+                          struct radeon_surface *surface)
 {
        struct r600_resource_texture *rtex;
        struct r600_resource *resource;
-       struct radeon *radeon = ((struct r600_screen*)screen)->radeon;
+       struct r600_screen *rscreen = (struct r600_screen*)screen;
+       int r;
 
        rtex = CALLOC_STRUCT(r600_resource_texture);
        if (rtex == NULL)
@@ -378,24 +505,25 @@ r600_texture_create_object(struct pipe_screen *screen,
        resource->b.b.vtbl = &r600_texture_vtbl;
        pipe_reference_init(&resource->b.b.b.reference, 1);
        resource->b.b.b.screen = screen;
-       resource->bo = bo;
        rtex->pitch_override = pitch_in_bytes_override;
        rtex->real_format = base->format;
 
        /* We must split depth and stencil into two separate buffers on Evergreen. */
-       if (r600_get_family_class(((struct r600_screen*)screen)->radeon) >= EVERGREEN &&
-           util_format_is_depth_and_stencil(base->format)) {
+       if (!(base->flags & R600_RESOURCE_FLAG_TRANSFER) &&
+           ((struct r600_screen*)screen)->chip_class >= EVERGREEN &&
+           util_format_is_depth_and_stencil(base->format) &&
+           !rscreen->use_surface_alloc) {
                struct pipe_resource stencil;
                unsigned stencil_pitch_override = 0;
 
                switch (base->format) {
-               case PIPE_FORMAT_Z24_UNORM_S8_USCALED:
+               case PIPE_FORMAT_Z24_UNORM_S8_UINT:
                        rtex->real_format = PIPE_FORMAT_Z24X8_UNORM;
                        break;
-               case PIPE_FORMAT_S8_USCALED_Z24_UNORM:
+               case PIPE_FORMAT_S8_UINT_Z24_UNORM:
                        rtex->real_format = PIPE_FORMAT_X8Z24_UNORM;
                        break;
-               case PIPE_FORMAT_Z32_FLOAT_S8X24_USCALED:
+               case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
                        rtex->real_format = PIPE_FORMAT_Z32_FLOAT;
                        break;
                default:
@@ -406,16 +534,17 @@ r600_texture_create_object(struct pipe_screen *screen,
 
                /* Divide the pitch in bytes by 4 for stencil, because it has a smaller pixel size. */
                if (pitch_in_bytes_override) {
-                       assert(base->format == PIPE_FORMAT_Z24_UNORM_S8_USCALED ||
-                              base->format == PIPE_FORMAT_S8_USCALED_Z24_UNORM);
+                       assert(base->format == PIPE_FORMAT_Z24_UNORM_S8_UINT ||
+                              base->format == PIPE_FORMAT_S8_UINT_Z24_UNORM);
                        stencil_pitch_override = pitch_in_bytes_override / 4;
                }
 
                /* Allocate the stencil buffer. */
                stencil = *base;
-               stencil.format = PIPE_FORMAT_S8_USCALED;
+               stencil.format = PIPE_FORMAT_S8_UINT;
                rtex->stencil = r600_texture_create_object(screen, &stencil, array_mode,
-                                                          stencil_pitch_override, max_buffer_size, bo);
+                                                          stencil_pitch_override,
+                                                          max_buffer_size, NULL, FALSE, surface);
                if (!rtex->stencil) {
                        FREE(rtex);
                        return NULL;
@@ -425,15 +554,20 @@ r600_texture_create_object(struct pipe_screen *screen,
 
        /* only mark depth textures the HW can hit as depth textures */
        if (util_format_is_depth_or_stencil(rtex->real_format) && permit_hardware_blit(screen, base))
-               rtex->depth = 1;
+               rtex->is_depth = true;
 
        r600_setup_miptree(screen, rtex, array_mode);
+       if (rscreen->use_surface_alloc) {
+               rtex->surface = *surface;
+               r = r600_setup_surface(screen, rtex, array_mode, pitch_in_bytes_override);
+               if (r) {
+                       FREE(rtex);
+                       return NULL;
+               }
+       }
 
-       resource->size = rtex->size;
-
-       /* If bo is not NULL, in which case depth and stencil must share the same buffer,
-        * and we initialized separate stencil for Evergreen. place it after depth. */
-       if (bo && rtex->stencil) {
+       /* If we initialized separate stencil for Evergreen. place it after depth. */
+       if (rtex->stencil) {
                unsigned stencil_align, stencil_offset;
 
                stencil_align = r600_get_base_alignment(screen, rtex->stencil->real_format, array_mode);
@@ -441,18 +575,36 @@ r600_texture_create_object(struct pipe_screen *screen,
 
                for (unsigned i = 0; i <= rtex->stencil->resource.b.b.b.last_level; i++)
                        rtex->stencil->offset[i] += stencil_offset;
+
+               rtex->size = stencil_offset + rtex->stencil->size;
        }
 
-       if (!resource->bo) {
+       /* Now create the backing buffer. */
+       if (!buf && alloc_bo) {
                struct pipe_resource *ptex = &rtex->resource.b.b.b;
                unsigned base_align = r600_get_base_alignment(screen, ptex->format, array_mode);
 
-               resource->bo = r600_bo(radeon, rtex->size, base_align, base->bind, base->usage);
-               if (!resource->bo) {
+               if (rscreen->use_surface_alloc) {
+                       base_align = rtex->surface.bo_alignment;
+               } else if (util_format_is_depth_or_stencil(rtex->real_format)) {
+                       /* ugly work around depth buffer need stencil room at end of bo */
+                       rtex->size += ptex->width0 * ptex->height0;
+               }
+               if (!r600_init_resource(rscreen, resource, rtex->size, base_align, base->bind, base->usage)) {
                        pipe_resource_reference((struct pipe_resource**)&rtex->stencil, NULL);
                        FREE(rtex);
                        return NULL;
                }
+       } else if (buf) {
+               resource->buf = buf;
+               resource->cs_buf = rscreen->ws->buffer_get_cs_handle(buf);
+               resource->domains = RADEON_DOMAIN_GTT | RADEON_DOMAIN_VRAM;
+       }
+
+       if (rtex->stencil) {
+               pb_reference(&rtex->stencil->resource.buf, rtex->resource.buf);
+               rtex->stencil->resource.cs_buf = rtex->resource.cs_buf;
+               rtex->stencil->resource.domains = rtex->resource.domains;
        }
        return rtex;
 }
@@ -460,34 +612,33 @@ r600_texture_create_object(struct pipe_screen *screen,
 struct pipe_resource *r600_texture_create(struct pipe_screen *screen,
                                                const struct pipe_resource *templ)
 {
+       struct r600_screen *rscreen = (struct r600_screen*)screen;
+       struct radeon_surface surface;
        unsigned array_mode = 0;
-       static int force_tiling = -1;
-
-       /* Would like some magic "get_bool_option_once" routine.
-        */
-       if (force_tiling == -1) {
-#if 0
-               /* reenable when 2D tiling is fixed better */
-               struct r600_screen *rscreen = (struct r600_screen *)screen;
-               if (r600_get_minor_version(rscreen->radeon) >= 9)
-                       force_tiling = debug_get_bool_option("R600_TILING", TRUE);
-#endif
-               force_tiling = debug_get_bool_option("R600_TILING", FALSE);
-       }
+       int r;
 
-       if (force_tiling && permit_hardware_blit(screen, templ)) {
-               if (!(templ->flags & R600_RESOURCE_FLAG_TRANSFER) &&
-                   !(templ->bind & PIPE_BIND_SCANOUT)) {
+       if (!(templ->flags & R600_RESOURCE_FLAG_TRANSFER)) {
+               if (rscreen->use_surface_alloc &&
+                   !(templ->bind & PIPE_BIND_SCANOUT) &&
+                   templ->usage != PIPE_USAGE_STAGING &&
+                   templ->usage != PIPE_USAGE_STREAM &&
+                   permit_hardware_blit(screen, templ)) {
                        array_mode = V_038000_ARRAY_2D_TILED_THIN1;
+               } else if (util_format_is_compressed(templ->format)) {
+                       array_mode = V_038000_ARRAY_1D_TILED_THIN1;
                }
        }
 
-       if (!(templ->flags & R600_RESOURCE_FLAG_TRANSFER) &&
-           util_format_is_compressed(templ->format))
-               array_mode = V_038000_ARRAY_1D_TILED_THIN1;
-
+       r = r600_init_surface(&surface, templ, array_mode);
+       if (r) {
+               return NULL;
+       }
+       r = rscreen->ws->surface_best(rscreen->ws, &surface);
+       if (r) {
+               return NULL;
+       }
        return (struct pipe_resource *)r600_texture_create_object(screen, templ, array_mode,
-                                                                 0, 0, NULL);
+                                                                 0, 0, NULL, TRUE, &surface);
 }
 
 static struct pipe_surface *r600_create_surface(struct pipe_context *pipe,
@@ -501,8 +652,6 @@ static struct pipe_surface *r600_create_surface(struct pipe_context *pipe,
        assert(surf_tmpl->u.tex.first_layer == surf_tmpl->u.tex.last_layer);
        if (surface == NULL)
                return NULL;
-       /* XXX no offset */
-/*     offset = r600_texture_get_offset(rtex, level, surf_tmpl->u.tex.first_layer);*/
        pipe_reference_init(&surface->base.reference, 1);
        pipe_resource_reference(&surface->base.texture, texture);
        surface->base.context = pipe;
@@ -527,28 +676,46 @@ static void r600_surface_destroy(struct pipe_context *pipe,
        FREE(surface);
 }
 
-
 struct pipe_resource *r600_texture_from_handle(struct pipe_screen *screen,
                                               const struct pipe_resource *templ,
                                               struct winsys_handle *whandle)
 {
-       struct radeon *rw = ((struct r600_screen*)screen)->radeon;
-       struct r600_bo *bo = NULL;
+       struct r600_screen *rscreen = (struct r600_screen*)screen;
+       struct pb_buffer *buf = NULL;
        unsigned stride = 0;
        unsigned array_mode = 0;
+       enum radeon_bo_layout micro, macro;
+       struct radeon_surface surface;
+       int r;
 
        /* Support only 2D textures without mipmaps */
        if ((templ->target != PIPE_TEXTURE_2D && templ->target != PIPE_TEXTURE_RECT) ||
              templ->depth0 != 1 || templ->last_level != 0)
                return NULL;
 
-       bo = r600_bo_handle(rw, whandle, &stride, &array_mode);
-       if (bo == NULL) {
+       buf = rscreen->ws->buffer_from_handle(rscreen->ws, whandle, &stride);
+       if (!buf)
                return NULL;
-       }
 
+       rscreen->ws->buffer_get_tiling(buf, &micro, &macro,
+                                      &surface.bankw, &surface.bankh,
+                                      &surface.tile_split,
+                                      &surface.stencil_tile_split,
+                                      &surface.mtilea);
+
+       if (macro == RADEON_LAYOUT_TILED)
+               array_mode = V_0280A0_ARRAY_2D_TILED_THIN1;
+       else if (micro == RADEON_LAYOUT_TILED)
+               array_mode = V_0280A0_ARRAY_1D_TILED_THIN1;
+       else
+               array_mode = 0;
+
+       r = r600_init_surface(&surface, templ, array_mode);
+       if (r) {
+               return NULL;
+       }
        return (struct pipe_resource *)r600_texture_create_object(screen, templ, array_mode,
-                                                                 stride, 0, bo);
+                                                                 stride, 0, buf, FALSE, &surface);
 }
 
 int r600_texture_depth_flush(struct pipe_context *ctx,
@@ -560,19 +727,17 @@ int r600_texture_depth_flush(struct pipe_context *ctx,
        if (rtex->flushed_depth_texture)
                goto out;
 
-       resource.target = PIPE_TEXTURE_2D;
+       resource.target = texture->target;
        resource.format = texture->format;
        resource.width0 = texture->width0;
        resource.height0 = texture->height0;
-       resource.depth0 = 1;
-       resource.array_size = 1;
+       resource.depth0 = texture->depth0;
+       resource.array_size = texture->array_size;
        resource.last_level = texture->last_level;
-       resource.nr_samples = 0;
+       resource.nr_samples = texture->nr_samples;
        resource.usage = PIPE_USAGE_DYNAMIC;
-       resource.bind = 0;
-       resource.flags = R600_RESOURCE_FLAG_TRANSFER;
-
-       resource.bind |= PIPE_BIND_DEPTH_STENCIL;
+       resource.bind = texture->bind | PIPE_BIND_DEPTH_STENCIL;
+       resource.flags = R600_RESOURCE_FLAG_TRANSFER | texture->flags;
 
        rtex->flushed_depth_texture = (struct r600_resource_texture *)ctx->screen->resource_create(ctx->screen, &resource);
        if (rtex->flushed_depth_texture == NULL) {
@@ -604,6 +769,7 @@ struct pipe_transfer* r600_texture_get_transfer(struct pipe_context *ctx,
                                                unsigned usage,
                                                const struct pipe_box *box)
 {
+       struct r600_context *rctx = (struct r600_context*)ctx;
        struct r600_resource_texture *rtex = (struct r600_resource_texture*)texture;
        struct pipe_resource resource;
        struct r600_transfer *trans;
@@ -623,21 +789,19 @@ struct pipe_transfer* r600_texture_get_transfer(struct pipe_context *ctx,
        if ((usage & PIPE_TRANSFER_READ) && u_box_volume(box) > 1024)
                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 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 (!permit_hardware_blit(ctx->screen, texture) ||
                (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;
@@ -645,7 +809,7 @@ struct pipe_transfer* r600_texture_get_transfer(struct pipe_context *ctx,
        trans->transfer.level = level;
        trans->transfer.usage = usage;
        trans->transfer.box = *box;
-       if (rtex->depth) {
+       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
@@ -683,8 +847,8 @@ struct pipe_transfer* r600_texture_get_transfer(struct pipe_context *ctx,
                        resource.bind |= PIPE_BIND_SAMPLER_VIEW;
                }
                /* Create the temporary texture. */
-               trans->staging_texture = ctx->screen->resource_create(ctx->screen, &resource);
-               if (trans->staging_texture == NULL) {
+               trans->staging = (struct r600_resource*)ctx->screen->resource_create(ctx->screen, &resource);
+               if (trans->staging == NULL) {
                        R600_ERR("failed to create temporary texture to hold untiled copy\n");
                        pipe_resource_reference(&trans->transfer.resource, NULL);
                        FREE(trans);
@@ -692,7 +856,7 @@ struct pipe_transfer* r600_texture_get_transfer(struct pipe_context *ctx,
                }
 
                trans->transfer.stride =
-                       ((struct r600_resource_texture *)trans->staging_texture)->pitch_in_bytes[0];
+                       ((struct r600_resource_texture *)trans->staging)->pitch_in_bytes[0];
                if (usage & PIPE_TRANSFER_READ) {
                        r600_copy_to_staging_texture(ctx, trans);
                        /* Always referenced in the blit. */
@@ -713,14 +877,14 @@ void r600_texture_transfer_destroy(struct pipe_context *ctx,
        struct pipe_resource *texture = transfer->resource;
        struct r600_resource_texture *rtex = (struct r600_resource_texture*)texture;
 
-       if (rtransfer->staging_texture) {
+       if (rtransfer->staging) {
                if (transfer->usage & PIPE_TRANSFER_WRITE) {
                        r600_copy_from_staging_texture(ctx, rtransfer);
                }
-               pipe_resource_reference(&rtransfer->staging_texture, NULL);
+               pipe_resource_reference((struct pipe_resource**)&rtransfer->staging, NULL);
        }
 
-       if (rtex->depth && !rtex->is_flushing_texture) {
+       if (rtex->is_depth && !rtex->is_flushing_texture) {
                if ((transfer->usage & PIPE_TRANSFER_WRITE) && rtex->flushed_depth_texture)
                        r600_blit_push_depth(ctx, rtex);
        }
@@ -732,30 +896,29 @@ void r600_texture_transfer_destroy(struct pipe_context *ctx,
 void* r600_texture_transfer_map(struct pipe_context *ctx,
                                struct pipe_transfer* transfer)
 {
-       struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
+       struct r600_context *rctx = (struct r600_context *)ctx;
        struct r600_transfer *rtransfer = (struct r600_transfer*)transfer;
-       struct r600_bo *bo;
+       struct pb_buffer *buf;
        enum pipe_format format = transfer->resource->format;
-       struct radeon *radeon = rctx->screen->radeon;
        unsigned offset = 0;
        char *map;
 
-       if (rtransfer->staging_texture) {
-               bo = ((struct r600_resource *)rtransfer->staging_texture)->bo;
+       if (rtransfer->staging) {
+               buf = ((struct r600_resource *)rtransfer->staging)->buf;
        } else {
                struct r600_resource_texture *rtex = (struct r600_resource_texture*)transfer->resource;
 
                if (rtex->flushed_depth_texture)
-                       bo = ((struct r600_resource *)rtex->flushed_depth_texture)->bo;
+                       buf = ((struct r600_resource *)rtex->flushed_depth_texture)->buf;
                else
-                       bo = ((struct r600_resource *)transfer->resource)->bo;
+                       buf = ((struct r600_resource *)transfer->resource)->buf;
 
                offset = rtransfer->offset +
                        transfer->box.y / util_format_get_blockheight(format) * transfer->stride +
                        transfer->box.x / util_format_get_blockwidth(format) * util_format_get_blocksize(format);
        }
 
-       if (!(map = r600_bo_map(radeon, bo, rctx->ctx.cs, transfer->usage))) {
+       if (!(map = rctx->ws->buffer_map(buf, rctx->cs, transfer->usage))) {
                return NULL;
        }
 
@@ -766,24 +929,24 @@ void r600_texture_transfer_unmap(struct pipe_context *ctx,
                                 struct pipe_transfer* transfer)
 {
        struct r600_transfer *rtransfer = (struct r600_transfer*)transfer;
-       struct radeon *radeon = ((struct r600_screen*)ctx->screen)->radeon;
-       struct r600_bo *bo;
+       struct r600_context *rctx = (struct r600_context*)ctx;
+       struct pb_buffer *buf;
 
-       if (rtransfer->staging_texture) {
-               bo = ((struct r600_resource *)rtransfer->staging_texture)->bo;
+       if (rtransfer->staging) {
+               buf = ((struct r600_resource *)rtransfer->staging)->buf;
        } else {
                struct r600_resource_texture *rtex = (struct r600_resource_texture*)transfer->resource;
 
                if (rtex->flushed_depth_texture) {
-                       bo = ((struct r600_resource *)rtex->flushed_depth_texture)->bo;
+                       buf = ((struct r600_resource *)rtex->flushed_depth_texture)->buf;
                } else {
-                       bo = ((struct r600_resource *)transfer->resource)->bo;
+                       buf = ((struct r600_resource *)transfer->resource)->buf;
                }
        }
-       r600_bo_unmap(radeon, bo);
+       rctx->ws->buffer_unmap(buf);
 }
 
-void r600_init_surface_functions(struct r600_pipe_context *r600)
+void r600_init_surface_functions(struct r600_context *r600)
 {
        r600->context.create_surface = r600_create_surface;
        r600->context.surface_destroy = r600_surface_destroy;
@@ -843,6 +1006,7 @@ uint32_t r600_translate_texformat(struct pipe_screen *screen,
        const struct util_format_description *desc;
        boolean uniform = TRUE;
        static int r600_enable_s3tc = -1;
+       bool is_srgb_valid = FALSE;
 
        int i;
        const uint32_t sign_bit[4] = {
@@ -863,26 +1027,26 @@ uint32_t r600_translate_texformat(struct pipe_screen *screen,
                case PIPE_FORMAT_Z16_UNORM:
                        result = FMT_16;
                        goto out_word4;
-               case PIPE_FORMAT_X24S8_USCALED:
+               case PIPE_FORMAT_X24S8_UINT:
                        word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT);
                case PIPE_FORMAT_Z24X8_UNORM:
-               case PIPE_FORMAT_Z24_UNORM_S8_USCALED:
+               case PIPE_FORMAT_Z24_UNORM_S8_UINT:
                        result = FMT_8_24;
                        goto out_word4;
-               case PIPE_FORMAT_S8X24_USCALED:
+               case PIPE_FORMAT_S8X24_UINT:
                        word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT);
                case PIPE_FORMAT_X8Z24_UNORM:
-               case PIPE_FORMAT_S8_USCALED_Z24_UNORM:
+               case PIPE_FORMAT_S8_UINT_Z24_UNORM:
                        result = FMT_24_8;
                        goto out_word4;
-               case PIPE_FORMAT_S8_USCALED:
+               case PIPE_FORMAT_S8_UINT:
                        result = FMT_8;
                        word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT);
                        goto out_word4;
                case PIPE_FORMAT_Z32_FLOAT:
                        result = FMT_32_FLOAT;
                        goto out_word4;
-               case PIPE_FORMAT_Z32_FLOAT_S8X24_USCALED:
+               case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
                        result = FMT_X24_8_32_FLOAT;
                        goto out_word4;
                default:
@@ -897,7 +1061,7 @@ uint32_t r600_translate_texformat(struct pipe_screen *screen,
                default:
                        break;
                }
-               goto out_unknown; /* TODO */
+               goto out_unknown; /* XXX */
 
        case UTIL_FORMAT_COLORSPACE_SRGB:
                word4 |= S_038010_FORCE_DEGAMMA(1);
@@ -909,7 +1073,7 @@ uint32_t r600_translate_texformat(struct pipe_screen *screen,
 
        if (r600_enable_s3tc == -1) {
                struct r600_screen *rscreen = (struct r600_screen *)screen;
-               if (r600_get_minor_version(rscreen->radeon) >= 9)
+               if (rscreen->info.drm_minor >= 9)
                        r600_enable_s3tc = 1;
                else
                        r600_enable_s3tc = debug_get_bool_option("R600_ENABLE_S3TC", FALSE);
@@ -954,14 +1118,32 @@ uint32_t r600_translate_texformat(struct pipe_screen *screen,
                case PIPE_FORMAT_DXT1_SRGB:
                case PIPE_FORMAT_DXT1_SRGBA:
                        result = FMT_BC1;
+                       is_srgb_valid = TRUE;
                        goto out_word4;
                case PIPE_FORMAT_DXT3_RGBA:
                case PIPE_FORMAT_DXT3_SRGBA:
                        result = FMT_BC2;
+                       is_srgb_valid = TRUE;
                        goto out_word4;
                case PIPE_FORMAT_DXT5_RGBA:
                case PIPE_FORMAT_DXT5_SRGBA:
                        result = FMT_BC3;
+                       is_srgb_valid = TRUE;
+                       goto out_word4;
+               default:
+                       goto out_unknown;
+               }
+       }
+
+       if (desc->layout == UTIL_FORMAT_LAYOUT_SUBSAMPLED) {
+               switch (format) {
+               case PIPE_FORMAT_R8G8_B8G8_UNORM:
+               case PIPE_FORMAT_G8R8_B8R8_UNORM:
+                       result = FMT_GB_GR;
+                       goto out_word4;
+               case PIPE_FORMAT_G8R8_G8B8_UNORM:
+               case PIPE_FORMAT_R8G8_R8B8_UNORM:
+                       result = FMT_BG_RG;
                        goto out_word4;
                default:
                        goto out_unknown;
@@ -983,7 +1165,7 @@ uint32_t r600_translate_texformat(struct pipe_screen *screen,
                }
        }
 
-       /* R8G8Bx_SNORM - TODO CxV8U8 */
+       /* R8G8Bx_SNORM - XXX CxV8U8 */
 
        /* See whether the components are of the same size. */
        for (i = 1; i < desc->nr_channels; i++) {
@@ -992,6 +1174,9 @@ uint32_t r600_translate_texformat(struct pipe_screen *screen,
 
        /* Non-uniform formats. */
        if (!uniform) {
+               if (desc->colorspace != UTIL_FORMAT_COLORSPACE_SRGB &&
+                   desc->channel[0].pure_integer)
+                       word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT);
                switch(desc->nr_channels) {
                case 3:
                        if (desc->channel[0].size == 5 &&
@@ -1035,10 +1220,15 @@ uint32_t r600_translate_texformat(struct pipe_screen *screen,
        switch (desc->channel[i].type) {
        case UTIL_FORMAT_TYPE_UNSIGNED:
        case UTIL_FORMAT_TYPE_SIGNED:
+#if 0
                if (!desc->channel[i].normalized &&
                    desc->colorspace != UTIL_FORMAT_COLORSPACE_SRGB) {
                        goto out_unknown;
                }
+#endif
+               if (desc->colorspace != UTIL_FORMAT_COLORSPACE_SRGB &&
+                   desc->channel[i].pure_integer)
+                       word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT);
 
                switch (desc->channel[i].size) {
                case 4:
@@ -1061,6 +1251,7 @@ uint32_t r600_translate_texformat(struct pipe_screen *screen,
                                goto out_word4;
                        case 4:
                                result = FMT_8_8_8_8;
+                               is_srgb_valid = TRUE;
                                goto out_word4;
                        }
                        goto out_unknown;
@@ -1124,6 +1315,9 @@ uint32_t r600_translate_texformat(struct pipe_screen *screen,
        }
 
 out_word4:
+
+       if (desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB && !is_srgb_valid)
+               return ~0;
        if (word4_p)
                *word4_p = word4;
        if (yuv_format_p)