X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Fr600%2Fr600_texture.c;h=14422bbfe4169b38969535066ee880278423c3b5;hb=d0293290ad620084d490b51693d97731a8935094;hp=158ae227b1d78e51a0554339fad4d8b0af541ced;hpb=9a589961a281f77b938a4b5bf80409326743f147;p=mesa.git

diff --git a/src/gallium/drivers/r600/r600_texture.c b/src/gallium/drivers/r600/r600_texture.c
index 158ae227b1d..14422bbfe41 100644
--- a/src/gallium/drivers/r600/r600_texture.c
+++ b/src/gallium/drivers/r600/r600_texture.c
@@ -31,114 +31,359 @@
 #include <util/u_inlines.h>
 #include <util/u_memory.h>
 #include "state_tracker/drm_driver.h"
-#include "r600_screen.h"
-#include "r600_context.h"
+#include "pipebuffer/pb_buffer.h"
+#include "r600_pipe.h"
 #include "r600_resource.h"
 #include "r600_state_inlines.h"
 #include "r600d.h"
+#include "r600_formats.h"
 
 extern struct u_resource_vtbl r600_texture_vtbl;
 
-/* Copy from a tiled texture to a detiled one. */
-static void r600_copy_from_tiled_texture(struct pipe_context *ctx, struct r600_transfer *rtransfer)
+/* 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_subresource subdst;
-
-	subdst.face = 0;
-	subdst.level = 0;
-	ctx->resource_copy_region(ctx, rtransfer->linear_texture,
-				subdst, 0, 0, 0, texture, transfer->sr,
-				transfer->box.x, transfer->box.y, transfer->box.z,
-				transfer->box.width, transfer->box.height);
+
+	ctx->resource_copy_region(ctx, rtransfer->staging_texture,
+				0, 0, 0, 0, texture, 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_box sbox;
+
+	sbox.x = sbox.y = sbox.z = 0;
+	sbox.width = transfer->box.width;
+	sbox.height = transfer->box.height;
+	/* XXX that might be wrong */
+	sbox.depth = 1;
+	ctx->resource_copy_region(ctx, texture, transfer->level,
+				  transfer->box.x, transfer->box.y, transfer->box.z,
+				  rtransfer->staging_texture,
+				  0, &sbox);
+
+	ctx->flush(ctx, 0, NULL);
 }
 
-static unsigned long r600_texture_get_offset(struct r600_resource_texture *rtex,
-					unsigned level, unsigned zslice,
-					unsigned face)
+unsigned r600_texture_get_offset(struct r600_resource_texture *rtex,
+					unsigned level, unsigned layer)
 {
-	unsigned long offset = rtex->offset[level];
+	unsigned offset = rtex->offset[level];
 
 	switch (rtex->resource.base.b.target) {
 	case PIPE_TEXTURE_3D:
-		assert(face == 0);
-		return offset + zslice * rtex->layer_size[level];
 	case PIPE_TEXTURE_CUBE:
-		assert(zslice == 0);
-		return offset + face * rtex->layer_size[level];
+		return offset + layer * rtex->layer_size[level];
 	default:
-		assert(zslice == 0 && face == 0);
+		assert(layer == 0);
 		return offset;
 	}
 }
 
-static void r600_setup_miptree(struct r600_resource_texture *rtex)
+static unsigned r600_get_pixel_alignment(struct pipe_screen *screen,
+					 enum pipe_format format,
+					 unsigned array_mode)
+{
+	struct r600_screen* rscreen = (struct r600_screen *)screen;
+	unsigned pixsize = util_format_get_blocksize(format);
+	int p_align;
+
+	switch(array_mode) {
+	case V_038000_ARRAY_1D_TILED_THIN1:
+		p_align = MAX2(8,
+			       ((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;
+		break;
+	case V_038000_ARRAY_LINEAR_GENERAL:
+	default:
+		p_align = rscreen->tiling_info->group_bytes / pixsize;
+		break;
+	}
+	return p_align;
+}
+
+static unsigned r600_get_height_alignment(struct pipe_screen *screen,
+					  unsigned array_mode)
+{
+	struct r600_screen* rscreen = (struct r600_screen *)screen;
+	int h_align;
+
+	switch (array_mode) {
+	case V_038000_ARRAY_2D_TILED_THIN1:
+		h_align = rscreen->tiling_info->num_channels * 8;
+		break;
+	case V_038000_ARRAY_1D_TILED_THIN1:
+		h_align = 8;
+		break;
+	default:
+		h_align = 1;
+		break;
+	}
+	return h_align;
+}
+
+static unsigned r600_get_base_alignment(struct pipe_screen *screen,
+					enum pipe_format format,
+					unsigned array_mode)
+{
+	struct r600_screen* rscreen = (struct r600_screen *)screen;
+	unsigned pixsize = util_format_get_blocksize(format);
+	int p_align = r600_get_pixel_alignment(screen, format, array_mode);
+	int h_align = r600_get_height_alignment(screen, array_mode);
+	int b_align;
+
+	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,
+			       p_align * pixsize * h_align);
+		break;
+	case V_038000_ARRAY_1D_TILED_THIN1:
+	default:
+		b_align = rscreen->tiling_info->group_bytes;
+		break;
+	}
+	return b_align;
+}
+
+static unsigned mip_minify(unsigned size, unsigned level)
+{
+	unsigned val;
+	val = u_minify(size, level);
+	if (level > 0)
+		val = util_next_power_of_two(val);
+	return val;
+}
+
+static unsigned r600_texture_get_stride(struct pipe_screen *screen,
+					struct r600_resource_texture *rtex,
+					unsigned level)
+{
+	struct pipe_resource *ptex = &rtex->resource.base.b;
+	unsigned width, stride, tile_width;
+
+	if (rtex->pitch_override)
+		return rtex->pitch_override;
+
+	width = mip_minify(ptex->width0, level);
+	if (util_format_is_plain(ptex->format)) {
+		tile_width = r600_get_pixel_alignment(screen, ptex->format,
+						      rtex->array_mode[level]);
+		width = align(width, tile_width);
+	}
+	stride = util_format_get_stride(ptex->format, width);
+
+	return stride;
+}
+
+static unsigned r600_texture_get_nblocksy(struct pipe_screen *screen,
+					  struct r600_resource_texture *rtex,
+					  unsigned level)
 {
 	struct pipe_resource *ptex = &rtex->resource.base.b;
-	unsigned long w, h, pitch, size, layer_size, i, offset;
+	unsigned height, tile_height;
+
+	height = mip_minify(ptex->height0, level);
+	if (util_format_is_plain(ptex->format)) {
+		tile_height = r600_get_height_alignment(screen,
+							rtex->array_mode[level]);
+		height = align(height, tile_height);
+	}
+	return util_format_get_nblocksy(ptex->format, height);
+}
+
+/* Get a width in pixels from a stride in bytes. */
+static unsigned pitch_to_width(enum pipe_format format, unsigned pitch_in_bytes)
+{
+	return (pitch_in_bytes / util_format_get_blocksize(format)) *
+		util_format_get_blockwidth(format);
+}
+
+static void r600_texture_set_array_mode(struct pipe_screen *screen,
+					struct r600_resource_texture *rtex,
+					unsigned level, unsigned array_mode)
+{
+	struct pipe_resource *ptex = &rtex->resource.base.b;
+
+	switch (array_mode) {
+	case V_0280A0_ARRAY_LINEAR_GENERAL:
+	case V_0280A0_ARRAY_LINEAR_ALIGNED:
+	case V_0280A0_ARRAY_1D_TILED_THIN1:
+	default:
+		rtex->array_mode[level] = array_mode;
+		break;
+	case V_0280A0_ARRAY_2D_TILED_THIN1:
+	{
+		unsigned w, h, tile_height, tile_width;
+
+		tile_height = r600_get_height_alignment(screen, array_mode);
+		tile_width = r600_get_pixel_alignment(screen, ptex->format, array_mode);
+
+		w = mip_minify(ptex->width0, level);
+		h = mip_minify(ptex->height0, level);
+		if (w < tile_width || h < tile_height)
+			rtex->array_mode[level] = V_0280A0_ARRAY_1D_TILED_THIN1;
+		else
+			rtex->array_mode[level] = array_mode;
+	}
+	break;
+	}
+}
+
+static void r600_setup_miptree(struct pipe_screen *screen,
+			       struct r600_resource_texture *rtex,
+			       unsigned array_mode)
+{
+	struct pipe_resource *ptex = &rtex->resource.base.b;
+	struct radeon *radeon = (struct radeon *)screen->winsys;
+	enum chip_class chipc = r600_get_family_class(radeon);
+	unsigned pitch, size, layer_size, i, offset;
+	unsigned nblocksy;
 
-	rtex->bpt = util_format_get_blocksize(ptex->format);
 	for (i = 0, offset = 0; i <= ptex->last_level; i++) {
-		w = u_minify(ptex->width0, i);
-		h = u_minify(ptex->height0, i);
-		h = util_next_power_of_two(h);
-		pitch = util_format_get_stride(ptex->format, align(w, 64));
-		pitch = align(pitch, 256);
-		layer_size = pitch * h;
-		if (ptex->target == PIPE_TEXTURE_CUBE)
-			size = layer_size * 6;
+		r600_texture_set_array_mode(screen, rtex, i, array_mode);
+
+		pitch = r600_texture_get_stride(screen, rtex, i);
+		nblocksy = r600_texture_get_nblocksy(screen, rtex, i);
+
+		layer_size = pitch * nblocksy;
+
+		if (ptex->target == PIPE_TEXTURE_CUBE) {
+			if (chipc >= R700)
+				size = layer_size * 8;
+			else
+				size = layer_size * 6;
+		}
 		else
 			size = layer_size * u_minify(ptex->depth0, i);
+		/* align base image and start of miptree */
+		if ((i == 0) || (i == 1))
+			offset = align(offset, r600_get_base_alignment(screen, ptex->format, array_mode));
 		rtex->offset[i] = offset;
 		rtex->layer_size[i] = layer_size;
-		rtex->pitch[i] = pitch;
-		rtex->width[i] = w;
-		rtex->height[i] = h;
+		rtex->pitch_in_bytes[i] = pitch;
+		rtex->pitch_in_pixels[i] = pitch_to_width(ptex->format, pitch);
 		offset += size;
 	}
 	rtex->size = offset;
 }
 
-struct pipe_resource *r600_texture_create(struct pipe_screen *screen,
-						const struct pipe_resource *templ)
+static struct r600_resource_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 r600_bo *bo)
 {
 	struct r600_resource_texture *rtex;
 	struct r600_resource *resource;
 	struct radeon *radeon = (struct radeon *)screen->winsys;
 
 	rtex = CALLOC_STRUCT(r600_resource_texture);
-	if (!rtex) {
+	if (rtex == NULL)
 		return NULL;
-	}
+
 	resource = &rtex->resource;
-	resource->base.b = *templ;
+	resource->base.b = *base;
 	resource->base.vtbl = &r600_texture_vtbl;
 	pipe_reference_init(&resource->base.b.reference, 1);
 	resource->base.b.screen = screen;
-	r600_setup_miptree(rtex);
+	resource->bo = bo;
+	rtex->pitch_override = pitch_in_bytes_override;
 
-	/* FIXME alignment 4096 enought ? too much ? */
-	resource->domain = r600_domain_from_usage(resource->base.b.bind);
-	resource->bo = radeon_bo(radeon, 0, rtex->size, 4096, NULL);
-	if (resource->bo == NULL) {
-		FREE(rtex);
-		return NULL;
+	if (util_format_is_depth_or_stencil(base->format))
+		rtex->depth = 1;
+
+	if (array_mode)
+		rtex->tiled = 1;
+	r600_setup_miptree(screen, rtex, array_mode);
+
+	resource->size = rtex->size;
+
+	if (!resource->bo) {
+		struct pipe_resource *ptex = &rtex->resource.base.b;
+		int 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) {
+			FREE(rtex);
+			return NULL;
+		}
 	}
-	return &resource->base.b;
+	return rtex;
 }
 
-static void r600_texture_destroy_state(struct pipe_resource *ptexture)
+/* Figure out whether u_blitter will fallback to a transfer operation.
+ * If so, don't use a staging resource.
+ */
+static boolean permit_hardware_blit(struct pipe_screen *screen,
+					const struct pipe_resource *res)
 {
-	struct r600_resource_texture *rtexture = (struct r600_resource_texture*)ptexture;
+	unsigned bind;
+
+	if (util_format_is_depth_or_stencil(res->format))
+		bind = PIPE_BIND_DEPTH_STENCIL;
+	else
+		bind = PIPE_BIND_RENDER_TARGET;
+
+	/* See r600_resource_copy_region: there is something wrong
+	 * with depth resource copies at the moment so avoid them for
+	 * now.
+	 */
+	if (util_format_get_component_bits(res->format,
+				UTIL_FORMAT_COLORSPACE_ZS,
+				0) != 0)
+		return FALSE;
+
+	if (!screen->is_format_supported(screen,
+				res->format,
+				res->target,
+				res->nr_samples,
+				bind, 0))
+		return FALSE;
+
+	if (!screen->is_format_supported(screen,
+				res->format,
+				res->target,
+				res->nr_samples,
+				PIPE_BIND_SAMPLER_VIEW, 0))
+		return FALSE;
+
+	return TRUE;
+}
 
-	for (int i = 0; i < PIPE_MAX_TEXTURE_LEVELS; i++) {
-		radeon_state_fini(&rtexture->scissor[i]);
-		radeon_state_fini(&rtexture->db[i]);
-		for (int j = 0; j < 8; j++) {
-			radeon_state_fini(&rtexture->cb[j][i]);
+struct pipe_resource *r600_texture_create(struct pipe_screen *screen,
+						const struct pipe_resource *templ)
+{
+	unsigned array_mode = 0;
+	static int force_tiling = -1;
+
+	/* Would like some magic "get_bool_option_once" routine.
+	 */
+	if (force_tiling == -1)
+		force_tiling = debug_get_bool_option("R600_FORCE_TILING", FALSE);
+
+	if (force_tiling && permit_hardware_blit(screen, templ)) {
+		if (!(templ->flags & R600_RESOURCE_FLAG_TRANSFER) &&
+		    !(templ->bind & PIPE_BIND_SCANOUT)) {
+			array_mode = V_038000_ARRAY_2D_TILED_THIN1;
 		}
 	}
+
+	return (struct pipe_resource *)r600_texture_create_object(screen, templ, array_mode,
+								  0, 0, NULL);
+
 }
 
 static void r600_texture_destroy(struct pipe_screen *screen,
@@ -148,128 +393,214 @@ static void r600_texture_destroy(struct pipe_screen *screen,
 	struct r600_resource *resource = &rtex->resource;
 	struct radeon *radeon = (struct radeon *)screen->winsys;
 
+	if (rtex->flushed_depth_texture)
+		pipe_resource_reference((struct pipe_resource **)&rtex->flushed_depth_texture, NULL);
+
 	if (resource->bo) {
-		radeon_bo_decref(radeon, resource->bo);
+		r600_bo_reference(radeon, &resource->bo, NULL);
 	}
-	if (rtex->uncompressed) {
-		radeon_bo_decref(radeon, rtex->uncompressed);
-	}
-	r600_texture_destroy_state(ptex);
 	FREE(rtex);
 }
 
-static struct pipe_surface *r600_get_tex_surface(struct pipe_screen *screen,
+static boolean r600_texture_get_handle(struct pipe_screen* screen,
+					struct pipe_resource *ptex,
+					struct winsys_handle *whandle)
+{
+	struct r600_resource_texture *rtex = (struct r600_resource_texture*)ptex;
+	struct r600_resource *resource = &rtex->resource;
+	struct radeon *radeon = (struct radeon *)screen->winsys;
+
+	return r600_bo_get_winsys_handle(radeon, resource->bo,
+			rtex->pitch_in_bytes[0], whandle);
+}
+
+static struct pipe_surface *r600_create_surface(struct pipe_context *pipe,
 						struct pipe_resource *texture,
-						unsigned face, unsigned level,
-						unsigned zslice, unsigned flags)
+						const struct pipe_surface *surf_tmpl)
 {
 	struct r600_resource_texture *rtex = (struct r600_resource_texture*)texture;
-	struct pipe_surface *surface = CALLOC_STRUCT(pipe_surface);
-	unsigned long offset;
+	struct r600_surface *surface = CALLOC_STRUCT(r600_surface);
+	unsigned tile_height;
+	unsigned level = surf_tmpl->u.tex.level;
 
+	assert(surf_tmpl->u.tex.first_layer == surf_tmpl->u.tex.last_layer);
 	if (surface == NULL)
 		return NULL;
-	offset = r600_texture_get_offset(rtex, level, zslice, face);
-	pipe_reference_init(&surface->reference, 1);
-	pipe_resource_reference(&surface->texture, texture);
-	surface->format = texture->format;
-	surface->width = u_minify(texture->width0, level);
-	surface->height = u_minify(texture->height0, level);
-	surface->offset = offset;
-	surface->usage = flags;
-	surface->zslice = zslice;
-	surface->texture = texture;
-	surface->face = face;
-	surface->level = level;
-	return surface;
+	/* 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;
+	surface->base.format = surf_tmpl->format;
+	surface->base.width = mip_minify(texture->width0, level);
+	surface->base.height = mip_minify(texture->height0, level);
+	surface->base.usage = surf_tmpl->usage;
+	surface->base.texture = texture;
+	surface->base.u.tex.first_layer = surf_tmpl->u.tex.first_layer;
+	surface->base.u.tex.last_layer = surf_tmpl->u.tex.last_layer;
+	surface->base.u.tex.level = level;
+
+	tile_height = r600_get_height_alignment(pipe->screen, rtex->array_mode[level]);
+	surface->aligned_height = align(surface->base.height, tile_height);
+	return &surface->base;
 }
 
-static void r600_tex_surface_destroy(struct pipe_surface *surface)
+static void r600_surface_destroy(struct pipe_context *pipe,
+				 struct pipe_surface *surface)
 {
 	pipe_resource_reference(&surface->texture, NULL);
 	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 radeon*)screen->winsys;
-	struct r600_resource_texture *rtex;
-	struct r600_resource *resource;
-	struct radeon_bo *bo = NULL;
+	struct r600_bo *bo = NULL;
+	unsigned array_mode = 0;
 
 	/* 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;
 
-	rtex = CALLOC_STRUCT(r600_resource_texture);
-	if (rtex == NULL)
-		return NULL;
-
-	bo = radeon_bo(rw, whandle->handle, 0, 0, NULL);
+	bo = r600_bo_handle(rw, whandle->handle, &array_mode);
 	if (bo == NULL) {
-		FREE(rtex);
 		return NULL;
 	}
 
-	resource = &rtex->resource;
-	resource->base.b = *templ;
-	resource->base.vtbl = &r600_texture_vtbl;
-	pipe_reference_init(&resource->base.b.reference, 1);
-	resource->base.b.screen = screen;
-	resource->bo = bo;
-	rtex->depth = 0;
-	rtex->pitch_override = whandle->stride;
-	rtex->bpt = util_format_get_blocksize(templ->format);
-	rtex->pitch[0] = whandle->stride;
-	rtex->width[0] = templ->width0;
-	rtex->height[0] = templ->height0;
-	rtex->offset[0] = 0;
-	rtex->size = align(rtex->pitch[0] * templ->height0, 64);
-
-	return &resource->base.b;
+	return (struct pipe_resource *)r600_texture_create_object(screen, templ, array_mode,
+								  whandle->stride,
+								  0,
+								  bo);
 }
 
 static unsigned int r600_texture_is_referenced(struct pipe_context *context,
 						struct pipe_resource *texture,
-						unsigned face, unsigned level)
+						unsigned level, int layer)
 {
 	/* FIXME */
 	return PIPE_REFERENCED_FOR_READ | PIPE_REFERENCED_FOR_WRITE;
 }
 
+int r600_texture_depth_flush(struct pipe_context *ctx,
+			     struct pipe_resource *texture)
+{
+	struct r600_resource_texture *rtex = (struct r600_resource_texture*)texture;
+	struct pipe_resource resource;
+
+	if (rtex->flushed_depth_texture)
+		goto out;
+
+	resource.target = PIPE_TEXTURE_2D;
+	resource.format = texture->format;
+	resource.width0 = texture->width0;
+	resource.height0 = texture->height0;
+	resource.depth0 = 1;
+	resource.last_level = 0;
+	resource.nr_samples = 0;
+	resource.usage = PIPE_USAGE_DYNAMIC;
+	resource.bind = 0;
+	resource.flags = R600_RESOURCE_FLAG_TRANSFER;
+
+	resource.bind |= PIPE_BIND_DEPTH_STENCIL;
+
+	rtex->flushed_depth_texture = (struct r600_resource_texture *)ctx->screen->resource_create(ctx->screen, &resource);
+	if (rtex->flushed_depth_texture == NULL) {
+		R600_ERR("failed to create temporary texture to hold untiled copy\n");
+		return -ENOMEM;
+	}
+
+out:
+	/* XXX: only do this if the depth texture has actually changed:
+	 */
+	r600_blit_uncompress_depth(ctx, rtex);
+	return 0;
+}
+
+/* Needs adjustment for pixelformat:
+ */
+static INLINE unsigned u_box_volume( const struct pipe_box *box )
+{
+	return box->width * box->depth * box->height;
+};
+
 struct pipe_transfer* r600_texture_get_transfer(struct pipe_context *ctx,
 						struct pipe_resource *texture,
-						struct pipe_subresource sr,
+						unsigned level,
 						unsigned usage,
 						const struct pipe_box *box)
 {
 	struct r600_resource_texture *rtex = (struct r600_resource_texture*)texture;
 	struct pipe_resource resource;
 	struct r600_transfer *trans;
+	int r;
+	boolean use_staging_texture = FALSE;
+
+	/* 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->tiled)
+		use_staging_texture = TRUE;
+
+	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_staging_texture = TRUE;
+
+	if (!permit_hardware_blit(ctx->screen, texture) ||
+		(texture->flags & R600_RESOURCE_FLAG_TRANSFER))
+		use_staging_texture = FALSE;
 
 	trans = CALLOC_STRUCT(r600_transfer);
 	if (trans == NULL)
 		return NULL;
 	pipe_resource_reference(&trans->transfer.resource, texture);
-	trans->transfer.sr = sr;
+	trans->transfer.level = level;
 	trans->transfer.usage = usage;
 	trans->transfer.box = *box;
-	trans->transfer.stride = rtex->pitch[sr.level];
-	trans->offset = r600_texture_get_offset(rtex, sr.level, box->z, sr.face);
-	if (rtex->tilled && !rtex->depth) {
+	if (rtex->depth && rtex->tile_type == 1) {
+		/* XXX: only readback the rectangle which is being mapped?
+		*/
+		/* XXX: when discard is true, no need to read back from depth texture
+		*/
+		r = r600_texture_depth_flush(ctx, texture);
+		if (r < 0) {
+			R600_ERR("failed to create temporary texture to hold untiled copy\n");
+			pipe_resource_reference(&trans->transfer.resource, NULL);
+			FREE(trans);
+			return NULL;
+		}
+		trans->transfer.stride = rtex->flushed_depth_texture->pitch_in_bytes[level];
+		trans->offset = r600_texture_get_offset(rtex->flushed_depth_texture, level, box->z);
+		return &trans->transfer;
+	} else if (use_staging_texture) {
 		resource.target = PIPE_TEXTURE_2D;
 		resource.format = texture->format;
 		resource.width0 = box->width;
 		resource.height0 = box->height;
-		resource.depth0 = 0;
+		resource.depth0 = 1;
+		resource.array_size = 1;
 		resource.last_level = 0;
 		resource.nr_samples = 0;
-		resource.usage = PIPE_USAGE_DYNAMIC;
+		resource.usage = PIPE_USAGE_STAGING;
 		resource.bind = 0;
-		resource.flags = 0;
+		resource.flags = R600_RESOURCE_FLAG_TRANSFER;
 		/* For texture reading, the temporary (detiled) texture is used as
 		 * a render target when blitting from a tiled texture. */
 		if (usage & PIPE_TRANSFER_READ) {
@@ -281,21 +612,25 @@ struct pipe_transfer* r600_texture_get_transfer(struct pipe_context *ctx,
 			resource.bind |= PIPE_BIND_SAMPLER_VIEW;
 		}
 		/* Create the temporary texture. */
-		trans->linear_texture = ctx->screen->resource_create(ctx->screen, &resource);
-		if (trans->linear_texture == NULL) {
+		trans->staging_texture = ctx->screen->resource_create(ctx->screen, &resource);
+		if (trans->staging_texture == 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->transfer.stride =
+			((struct r600_resource_texture *)trans->staging_texture)->pitch_in_bytes[0];
 		if (usage & PIPE_TRANSFER_READ) {
-			/* We cannot map a tiled texture directly because the data is
-			 * in a different order, therefore we do detiling using a blit. */
-			r600_copy_from_tiled_texture(ctx, trans);
+			r600_copy_to_staging_texture(ctx, trans);
 			/* Always referenced in the blit. */
 			ctx->flush(ctx, 0, NULL);
 		}
+		return &trans->transfer;
 	}
+	trans->transfer.stride = rtex->pitch_in_bytes[level];
+	trans->offset = r600_texture_get_offset(rtex, level, box->z);
 	return &trans->transfer;
 }
 
@@ -304,8 +639,11 @@ void r600_texture_transfer_destroy(struct pipe_context *ctx,
 {
 	struct r600_transfer *rtransfer = (struct r600_transfer*)transfer;
 
-	if (rtransfer->linear_texture) {
-		pipe_resource_reference(&rtransfer->linear_texture, NULL);
+	if (rtransfer->staging_texture) {
+		if (transfer->usage & PIPE_TRANSFER_WRITE) {
+			r600_copy_from_staging_texture(ctx, rtransfer);
+		}
+		pipe_resource_reference(&rtransfer->staging_texture, NULL);
 	}
 	pipe_resource_reference(&transfer->resource, NULL);
 	FREE(transfer);
@@ -315,39 +653,55 @@ void* r600_texture_transfer_map(struct pipe_context *ctx,
 				struct pipe_transfer* transfer)
 {
 	struct r600_transfer *rtransfer = (struct r600_transfer*)transfer;
-	struct radeon_bo *bo;
+	struct r600_bo *bo;
 	enum pipe_format format = transfer->resource->format;
 	struct radeon *radeon = (struct radeon *)ctx->screen->winsys;
-	struct r600_resource_texture *rtex;
-	unsigned long offset = 0;
+	unsigned offset = 0;
+	unsigned usage = 0;
 	char *map;
-	int r;
 
-	r600_flush(ctx, 0, NULL);
-	if (rtransfer->linear_texture) {
-		bo = ((struct r600_resource *)rtransfer->linear_texture)->bo;
+	if (rtransfer->staging_texture) {
+		bo = ((struct r600_resource *)rtransfer->staging_texture)->bo;
 	} else {
-		rtex = (struct r600_resource_texture*)transfer->resource;
-		if (rtex->depth) {
-			r = r600_texture_from_depth(ctx, rtex, transfer->sr.level);
-			if (r) {
-				return NULL;
-			}
-			r600_flush(ctx, 0, NULL);
-			bo = rtex->uncompressed;
-		} 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;
+		else
 			bo = ((struct r600_resource *)transfer->resource)->bo;
-		}
+
 		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 (radeon_bo_map(radeon, bo)) {
+
+	if (transfer->usage & PIPE_TRANSFER_WRITE) {
+		usage |= PB_USAGE_CPU_WRITE;
+
+		if (transfer->usage & PIPE_TRANSFER_DISCARD) {
+		}
+
+		if (transfer->usage & PIPE_TRANSFER_FLUSH_EXPLICIT) {
+		}
+	}
+
+	if (transfer->usage & PIPE_TRANSFER_READ) {
+		usage |= PB_USAGE_CPU_READ;
+	}
+
+	if (transfer->usage & PIPE_TRANSFER_DONTBLOCK) {
+		usage |= PB_USAGE_DONTBLOCK;
+	}
+
+	if (transfer->usage & PIPE_TRANSFER_UNSYNCHRONIZED) {
+		usage |= PB_USAGE_UNSYNCHRONIZED;
+	}
+
+	map = r600_bo_map(radeon, bo, usage, ctx);
+	if (!map) {
 		return NULL;
 	}
-	radeon_bo_wait(radeon, bo);
 
-	map = bo->data;
 	return map + offset;
 }
 
@@ -356,25 +710,25 @@ void r600_texture_transfer_unmap(struct pipe_context *ctx,
 {
 	struct r600_transfer *rtransfer = (struct r600_transfer*)transfer;
 	struct radeon *radeon = (struct radeon *)ctx->screen->winsys;
-	struct r600_resource_texture *rtex;
-	struct radeon_bo *bo;
+	struct r600_bo *bo;
 
-	if (rtransfer->linear_texture) {
-		bo = ((struct r600_resource *)rtransfer->linear_texture)->bo;
+	if (rtransfer->staging_texture) {
+		bo = ((struct r600_resource *)rtransfer->staging_texture)->bo;
 	} else {
-		rtex = (struct r600_resource_texture*)transfer->resource;
-		if (rtex->depth) {
-			bo = rtex->uncompressed;
+		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;
 		} else {
 			bo = ((struct r600_resource *)transfer->resource)->bo;
 		}
 	}
-	radeon_bo_unmap(radeon, bo);
+	r600_bo_unmap(radeon, bo);
 }
 
 struct u_resource_vtbl r600_texture_vtbl =
 {
-	u_default_resource_get_handle,	/* get_handle */
+	r600_texture_get_handle,	/* get_handle */
 	r600_texture_destroy,		/* resource_destroy */
 	r600_texture_is_referenced,	/* is_resource_referenced */
 	r600_texture_get_transfer,	/* get_transfer */
@@ -385,10 +739,10 @@ struct u_resource_vtbl r600_texture_vtbl =
 	u_default_transfer_inline_write	/* transfer_inline_write */
 };
 
-void r600_init_screen_texture_functions(struct pipe_screen *screen)
+void r600_init_surface_functions(struct r600_pipe_context *r600)
 {
-	screen->get_tex_surface = r600_get_tex_surface;
-	screen->tex_surface_destroy = r600_tex_surface_destroy;
+	r600->context.create_surface = r600_create_surface;
+	r600->context.surface_destroy = r600_surface_destroy;
 }
 
 static unsigned r600_get_swizzle_combined(const unsigned char *swizzle_format,
@@ -441,7 +795,7 @@ static unsigned r600_get_swizzle_combined(const unsigned char *swizzle_format,
 
 /* texture format translate */
 uint32_t r600_translate_texformat(enum pipe_format format,
-				  const unsigned char *swizzle_view, 
+				  const unsigned char *swizzle_view,
 				  uint32_t *word4_p, uint32_t *yuv_format_p)
 {
 	uint32_t result = 0, word4 = 0, yuv_format = 0;
@@ -464,13 +818,23 @@ uint32_t r600_translate_texformat(enum pipe_format format,
 	case UTIL_FORMAT_COLORSPACE_ZS:
 		switch (format) {
 		case PIPE_FORMAT_Z16_UNORM:
-			result = V_0280A0_COLOR_16;
+			result = FMT_16;
 			goto out_word4;
+		case PIPE_FORMAT_X24S8_USCALED:
+			word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT);
 		case PIPE_FORMAT_Z24X8_UNORM:
-			result = V_0280A0_COLOR_8_24;
-			goto out_word4;
 		case PIPE_FORMAT_Z24_UNORM_S8_USCALED:
-			result = V_0280A0_COLOR_8_24;
+			result = FMT_8_24;
+			goto out_word4;
+		case PIPE_FORMAT_S8X24_USCALED:
+			word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT);
+		case PIPE_FORMAT_X8Z24_UNORM:
+		case PIPE_FORMAT_S8_USCALED_Z24_UNORM:
+			result = FMT_24_8;
+			goto out_word4;
+		case PIPE_FORMAT_S8_USCALED:
+			result = FMT_8;
+			word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT);
 			goto out_word4;
 		default:
 			goto out_unknown;
@@ -479,13 +843,13 @@ uint32_t r600_translate_texformat(enum pipe_format format,
 	case UTIL_FORMAT_COLORSPACE_YUV:
 		yuv_format |= (1 << 30);
 		switch (format) {
-                case PIPE_FORMAT_UYVY:
-                case PIPE_FORMAT_YUYV:
+		case PIPE_FORMAT_UYVY:
+		case PIPE_FORMAT_YUYV:
 		default:
 			break;
 		}
 		goto out_unknown; /* TODO */
-		
+
 	case UTIL_FORMAT_COLORSPACE_SRGB:
 		word4 |= S_038010_FORCE_DEGAMMA(1);
 		if (format == PIPE_FORMAT_L8A8_SRGB || format == PIPE_FORMAT_L8_SRGB)
@@ -498,11 +862,33 @@ uint32_t r600_translate_texformat(enum pipe_format format,
 
 	/* S3TC formats. TODO */
 	if (desc->layout == UTIL_FORMAT_LAYOUT_S3TC) {
-		goto out_unknown;
+		static int r600_enable_s3tc = -1;
+
+		if (r600_enable_s3tc == -1)
+			r600_enable_s3tc =
+				debug_get_bool_option("R600_ENABLE_S3TC", FALSE);
+
+		if (!r600_enable_s3tc)
+			goto out_unknown;
+
+		switch (format) {
+		case PIPE_FORMAT_DXT1_RGB:
+		case PIPE_FORMAT_DXT1_RGBA:
+			result = FMT_BC1;
+			goto out_word4;
+		case PIPE_FORMAT_DXT3_RGBA:
+			result = FMT_BC2;
+			goto out_word4;
+		case PIPE_FORMAT_DXT5_RGBA:
+			result = FMT_BC3;
+			goto out_word4;
+		default:
+			goto out_unknown;
+		}
 	}
 
 
-	for (i = 0; i < desc->nr_channels; i++) {	
+	for (i = 0; i < desc->nr_channels; i++) {
 		if (desc->channel[i].type == UTIL_FORMAT_TYPE_SIGNED) {
 			word4 |= sign_bit[i];
 		}
@@ -516,7 +902,7 @@ uint32_t r600_translate_texformat(enum pipe_format format,
 	for (i = 1; i < desc->nr_channels; i++) {
 		uniform = uniform && desc->channel[0].size == desc->channel[i].size;
 	}
-	
+
 	/* Non-uniform formats. */
 	if (!uniform) {
 		switch(desc->nr_channels) {
@@ -524,7 +910,7 @@ uint32_t r600_translate_texformat(enum pipe_format format,
 			if (desc->channel[0].size == 5 &&
 			    desc->channel[1].size == 6 &&
 			    desc->channel[2].size == 5) {
-				result = V_0280A0_COLOR_5_6_5;
+				result = FMT_5_6_5;
 				goto out_word4;
 			}
 			goto out_unknown;
@@ -533,14 +919,14 @@ uint32_t r600_translate_texformat(enum pipe_format format,
 			    desc->channel[1].size == 5 &&
 			    desc->channel[2].size == 5 &&
 			    desc->channel[3].size == 1) {
-				result = V_0280A0_COLOR_1_5_5_5;
+				result = FMT_1_5_5_5;
 				goto out_word4;
 			}
 			if (desc->channel[0].size == 10 &&
 			    desc->channel[1].size == 10 &&
 			    desc->channel[2].size == 10 &&
 			    desc->channel[3].size == 2) {
-				result = V_0280A0_COLOR_10_10_10_2;
+				result = FMT_2_10_10_10;
 				goto out_word4;
 			}
 			goto out_unknown;
@@ -548,83 +934,93 @@ uint32_t r600_translate_texformat(enum pipe_format format,
 		goto out_unknown;
 	}
 
+	/* Find the first non-VOID channel. */
+	for (i = 0; i < 4; i++) {
+		if (desc->channel[i].type != UTIL_FORMAT_TYPE_VOID) {
+			break;
+		}
+	}
+
+	if (i == 4)
+		goto out_unknown;
+
 	/* uniform formats */
-	switch (desc->channel[0].type) {
+	switch (desc->channel[i].type) {
 	case UTIL_FORMAT_TYPE_UNSIGNED:
 	case UTIL_FORMAT_TYPE_SIGNED:
-		if (!desc->channel[0].normalized &&
+		if (!desc->channel[i].normalized &&
 		    desc->colorspace != UTIL_FORMAT_COLORSPACE_SRGB) {
 			goto out_unknown;
 		}
 
-		switch (desc->channel[0].size) {
+		switch (desc->channel[i].size) {
 		case 4:
 			switch (desc->nr_channels) {
 			case 2:
-				result = V_0280A0_COLOR_4_4;
+				result = FMT_4_4;
 				goto out_word4;
 			case 4:
-				result = V_0280A0_COLOR_4_4_4_4;
+				result = FMT_4_4_4_4;
 				goto out_word4;
 			}
 			goto out_unknown;
 		case 8:
 			switch (desc->nr_channels) {
 			case 1:
-				result = V_0280A0_COLOR_8;
+				result = FMT_8;
 				goto out_word4;
 			case 2:
-				result = V_0280A0_COLOR_8_8;
+				result = FMT_8_8;
 				goto out_word4;
 			case 4:
-				result = V_0280A0_COLOR_8_8_8_8;
+				result = FMT_8_8_8_8;
 				goto out_word4;
 			}
 			goto out_unknown;
 		case 16:
 			switch (desc->nr_channels) {
 			case 1:
-				result = V_0280A0_COLOR_16;
+				result = FMT_16;
 				goto out_word4;
 			case 2:
-				result = V_0280A0_COLOR_16_16;
+				result = FMT_16_16;
 				goto out_word4;
 			case 4:
-				result = V_0280A0_COLOR_16_16_16_16;
+				result = FMT_16_16_16_16;
 				goto out_word4;
 			}
 		}
 		goto out_unknown;
 
 	case UTIL_FORMAT_TYPE_FLOAT:
-		switch (desc->channel[0].size) {
+		switch (desc->channel[i].size) {
 		case 16:
 			switch (desc->nr_channels) {
 			case 1:
-				result = V_0280A0_COLOR_16_FLOAT;
+				result = FMT_16_FLOAT;
 				goto out_word4;
 			case 2:
-				result = V_0280A0_COLOR_16_16_FLOAT;
+				result = FMT_16_16_FLOAT;
 				goto out_word4;
 			case 4:
-				result = V_0280A0_COLOR_16_16_16_16_FLOAT;
+				result = FMT_16_16_16_16_FLOAT;
 				goto out_word4;
 			}
 			goto out_unknown;
 		case 32:
 			switch (desc->nr_channels) {
 			case 1:
-				result = V_0280A0_COLOR_32_FLOAT;
+				result = FMT_32_FLOAT;
 				goto out_word4;
 			case 2:
-				result = V_0280A0_COLOR_32_32_FLOAT;
+				result = FMT_32_32_FLOAT;
 				goto out_word4;
 			case 4:
-				result = V_0280A0_COLOR_32_32_32_32_FLOAT;
+				result = FMT_32_32_32_32_FLOAT;
 				goto out_word4;
 			}
 		}
-		
+
 	}
 out_word4:
 	if (word4_p)
@@ -636,81 +1032,3 @@ out_unknown:
 //	R600_ERR("Unable to handle texformat %d %s\n", format, util_format_name(format));
 	return ~0;
 }
-
-int r600_texture_from_depth(struct pipe_context *ctx, struct r600_resource_texture *rtexture, unsigned level)
-{
-	struct r600_screen *rscreen = r600_screen(ctx->screen);
-	int r;
-
-	if (!rtexture->depth) {
-		/* This shouldn't happen maybe print a warning */
-		return 0;
-	}
-	if (rtexture->uncompressed && !rtexture->dirty) {
-		/* Uncompressed bo already in good state */
-		return 0;
-	}
-
-	/* allocate uncompressed texture */
-	if (rtexture->uncompressed == NULL) {
-		rtexture->uncompressed = radeon_bo(rscreen->rw, 0, rtexture->size, 4096, NULL);
-		if (rtexture->uncompressed == NULL) {
-			return -ENOMEM;
-		}
-	}
-
-	/* render a rectangle covering whole buffer to uncompress depth */
-	r = r600_blit_uncompress_depth(ctx, rtexture, level);
-	if (r) {
-		return r;
-	}
-
-	rtexture->dirty = 0;
-	return 0;
-}
-
-
-
-int r600_texture_scissor(struct pipe_context *ctx, struct r600_resource_texture *rtexture, unsigned level)
-{
-	struct r600_screen *rscreen = r600_screen(ctx->screen);
-	struct r600_context *rctx = r600_context(ctx);
-
-	if (!rtexture->scissor[level].cpm4) {
-		rctx->vtbl->texture_state_scissor(rscreen, rtexture, level);
-	}
-	return 0;
-}
-
-int r600_texture_cb(struct pipe_context *ctx, struct r600_resource_texture *rtexture, unsigned cb, unsigned level)
-{
-	struct r600_screen *rscreen = r600_screen(ctx->screen);
-	struct r600_context *rctx = r600_context(ctx);
-
-	if (!rtexture->cb[cb][level].cpm4) {
-		rctx->vtbl->texture_state_cb(rscreen, rtexture, cb, level);
-	}
-	return 0;
-}
-
-int r600_texture_db(struct pipe_context *ctx, struct r600_resource_texture *rtexture, unsigned level)
-{
-	struct r600_screen *rscreen = r600_screen(ctx->screen);
-	struct r600_context *rctx = r600_context(ctx);
-
-	if (!rtexture->db[level].cpm4) {
-		rctx->vtbl->texture_state_db(rscreen, rtexture, level);
-	}
-	return 0;
-}
-
-int r600_texture_viewport(struct pipe_context *ctx, struct r600_resource_texture *rtexture, unsigned level)
-{
-	struct r600_screen *rscreen = r600_screen(ctx->screen);
-	struct r600_context *rctx = r600_context(ctx);
-
-	if (!rtexture->viewport[level].cpm4) {
-		rctx->vtbl->texture_state_viewport(rscreen, rtexture, level);
-	}
-	return 0;
-}