radeonsi: move r600_cs.h contents into si_pipe.h, si_build_pm4.h
[mesa.git] / src / gallium / drivers / radeon / r600_texture.c
index 1a0503bec6ee14a0fd166f505424b066be2dd229..cdca6c1f48da6a90963f3872cfec9d93631a0579 100644 (file)
@@ -1,5 +1,7 @@
 /*
  * Copyright 2010 Jerome Glisse <glisse@freedesktop.org>
+ * Copyright 2018 Advanced Micro Devices, Inc.
+ * All Rights Reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
  * copy of this software and associated documentation files (the "Software"),
@@ -22,7 +24,6 @@
  */
 
 #include "radeonsi/si_pipe.h"
-#include "r600_cs.h"
 #include "r600_query.h"
 #include "util/u_format.h"
 #include "util/u_log.h"
 #include "state_tracker/drm_driver.h"
 #include "amd/common/sid.h"
 
-static void r600_texture_discard_cmask(struct si_screen *sscreen,
-                                      struct r600_texture *rtex);
 static enum radeon_surf_mode
-r600_choose_tiling(struct si_screen *sscreen,
-                  const struct pipe_resource *templ);
+si_choose_tiling(struct si_screen *sscreen,
+                const struct pipe_resource *templ);
 
 
-bool si_prepare_for_dma_blit(struct r600_common_context *rctx,
+bool si_prepare_for_dma_blit(struct si_context *sctx,
                             struct r600_texture *rdst,
                             unsigned dst_level, unsigned dstx,
                             unsigned dsty, unsigned dstz,
@@ -51,7 +50,7 @@ bool si_prepare_for_dma_blit(struct r600_common_context *rctx,
                             unsigned src_level,
                             const struct pipe_box *src_box)
 {
-       if (!rctx->dma.cs)
+       if (!sctx->b.dma_cs)
                return false;
 
        if (rdst->surface.bpe != rsrc->surface.bpe)
@@ -90,12 +89,12 @@ bool si_prepare_for_dma_blit(struct r600_common_context *rctx,
                                                      src_box->height, src_box->depth))
                        return false;
 
-               r600_texture_discard_cmask(rctx->screen, rdst);
+               si_texture_discard_cmask(sctx->screen, rdst);
        }
 
        /* All requirements are met. Prepare textures for SDMA. */
        if (rsrc->cmask.size && rsrc->dirty_level_mask & (1 << src_level))
-               rctx->b.flush_resource(&rctx->b, &rsrc->resource.b.b);
+               sctx->b.b.flush_resource(&sctx->b.b, &rsrc->resource.b.b);
 
        assert(!(rsrc->dirty_level_mask & (1 << src_level)));
        assert(!(rdst->dirty_level_mask & (1 << dst_level)));
@@ -104,13 +103,13 @@ bool si_prepare_for_dma_blit(struct r600_common_context *rctx,
 }
 
 /* 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)
+static void si_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;
 
@@ -138,27 +137,27 @@ static void r600_copy_region_with_blit(struct pipe_context *pipe,
 }
 
 /* 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)
+static void si_copy_to_staging_texture(struct pipe_context *ctx, struct r600_transfer *rtransfer)
 {
-       struct r600_common_context *rctx = (struct r600_common_context*)ctx;
+       struct si_context *sctx = (struct si_context*)ctx;
        struct pipe_transfer *transfer = (struct pipe_transfer*)rtransfer;
        struct pipe_resource *dst = &rtransfer->staging->b.b;
        struct pipe_resource *src = transfer->resource;
 
        if (src->nr_samples > 1) {
-               r600_copy_region_with_blit(ctx, dst, 0, 0, 0, 0,
+               si_copy_region_with_blit(ctx, dst, 0, 0, 0, 0,
                                           src, transfer->level, &transfer->box);
                return;
        }
 
-       rctx->dma_copy(ctx, dst, 0, 0, 0, 0, src, transfer->level,
+       sctx->b.dma_copy(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)
+static void si_copy_from_staging_texture(struct pipe_context *ctx, struct r600_transfer *rtransfer)
 {
-       struct r600_common_context *rctx = (struct r600_common_context*)ctx;
+       struct si_context *sctx = (struct si_context*)ctx;
        struct pipe_transfer *transfer = (struct pipe_transfer*)rtransfer;
        struct pipe_resource *dst = transfer->resource;
        struct pipe_resource *src = &rtransfer->staging->b.b;
@@ -167,22 +166,22 @@ static void r600_copy_from_staging_texture(struct pipe_context *ctx, struct r600
        u_box_3d(0, 0, 0, transfer->box.width, transfer->box.height, transfer->box.depth, &sbox);
 
        if (dst->nr_samples > 1) {
-               r600_copy_region_with_blit(ctx, dst, transfer->level,
+               si_copy_region_with_blit(ctx, dst, transfer->level,
                                           transfer->box.x, transfer->box.y, transfer->box.z,
                                           src, 0, &sbox);
                return;
        }
 
-       rctx->dma_copy(ctx, dst, transfer->level,
+       sctx->b.dma_copy(ctx, dst, transfer->level,
                       transfer->box.x, transfer->box.y, transfer->box.z,
                       src, 0, &sbox);
 }
 
-static unsigned r600_texture_get_offset(struct si_screen *sscreen,
-                                       struct r600_texture *rtex, unsigned level,
-                                       const struct pipe_box *box,
-                                       unsigned *stride,
-                                       unsigned *layer_stride)
+static unsigned si_texture_get_offset(struct si_screen *sscreen,
+                                     struct r600_texture *rtex, unsigned level,
+                                     const struct pipe_box *box,
+                                     unsigned *stride,
+                                     unsigned *layer_stride)
 {
        if (sscreen->info.chip_class >= GFX9) {
                *stride = rtex->surface.u.gfx9.surf_pitch * rtex->surface.bpe;
@@ -217,16 +216,16 @@ static unsigned r600_texture_get_offset(struct si_screen *sscreen,
        }
 }
 
-static int r600_init_surface(struct si_screen *sscreen,
-                            struct radeon_surf *surface,
-                            const struct pipe_resource *ptex,
-                            enum radeon_surf_mode array_mode,
-                            unsigned pitch_in_bytes_override,
-                            unsigned offset,
-                            bool is_imported,
-                            bool is_scanout,
-                            bool is_flushed_depth,
-                            bool tc_compatible_htile)
+static int si_init_surface(struct si_screen *sscreen,
+                          struct radeon_surf *surface,
+                          const struct pipe_resource *ptex,
+                          enum radeon_surf_mode array_mode,
+                          unsigned pitch_in_bytes_override,
+                          unsigned offset,
+                          bool is_imported,
+                          bool is_scanout,
+                          bool is_flushed_depth,
+                          bool tc_compatible_htile)
 {
        const struct util_format_description *desc =
                util_format_description(ptex->format);
@@ -242,7 +241,7 @@ static int r600_init_surface(struct si_screen *sscreen,
                bpe = 4; /* stencil is allocated separately on evergreen */
        } else {
                bpe = util_format_get_blocksize(ptex->format);
-               assert(util_is_power_of_two(bpe));
+               assert(util_is_power_of_two_or_zero(bpe));
        }
 
        if (!is_flushed_depth && is_depth) {
@@ -267,7 +266,7 @@ static int r600_init_surface(struct si_screen *sscreen,
        }
 
        if (sscreen->info.chip_class >= VI &&
-           (ptex->flags & R600_RESOURCE_FLAG_DISABLE_DCC ||
+           (ptex->flags & SI_RESOURCE_FLAG_DISABLE_DCC ||
             ptex->format == PIPE_FORMAT_R9G9B9E5_FLOAT ||
             /* DCC MSAA array textures are disallowed due to incomplete clear impl. */
             (ptex->nr_samples >= 2 &&
@@ -289,7 +288,7 @@ static int r600_init_surface(struct si_screen *sscreen,
                flags |= RADEON_SURF_SHAREABLE;
        if (is_imported)
                flags |= RADEON_SURF_IMPORTED | RADEON_SURF_SHAREABLE;
-       if (!(ptex->flags & R600_RESOURCE_FLAG_FORCE_TILING))
+       if (!(ptex->flags & SI_RESOURCE_FLAG_FORCE_TILING))
                flags |= RADEON_SURF_OPTIMIZE_FOR_SPACE;
 
        r = sscreen->ws->surface_init(sscreen->ws, ptex, flags, bpe,
@@ -298,11 +297,21 @@ static int r600_init_surface(struct si_screen *sscreen,
                return r;
        }
 
+       unsigned pitch = pitch_in_bytes_override / bpe;
+
        if (sscreen->info.chip_class >= GFX9) {
-               assert(!pitch_in_bytes_override ||
-                      pitch_in_bytes_override == surface->u.gfx9.surf_pitch * bpe);
+               if (pitch) {
+                       surface->u.gfx9.surf_pitch = pitch;
+                       surface->u.gfx9.surf_slice_size =
+                               (uint64_t)pitch * surface->u.gfx9.surf_height * bpe;
+               }
                surface->u.gfx9.surf_offset = offset;
        } else {
+               if (pitch) {
+                       surface->u.legacy.level[0].nblk_x = pitch;
+                       surface->u.legacy.level[0].slice_size_dw =
+                               ((uint64_t)pitch * surface->u.legacy.level[0].nblk_y * bpe) / 4;
+               }
                if (offset) {
                        for (i = 0; i < ARRAY_SIZE(surface->u.legacy.level); ++i)
                                surface->u.legacy.level[i].offset += offset;
@@ -311,9 +320,9 @@ static int r600_init_surface(struct si_screen *sscreen,
        return 0;
 }
 
-static void r600_texture_init_metadata(struct si_screen *sscreen,
-                                      struct r600_texture *rtex,
-                                      struct radeon_bo_metadata *metadata)
+static void si_texture_init_metadata(struct si_screen *sscreen,
+                                    struct r600_texture *rtex,
+                                    struct radeon_bo_metadata *metadata)
 {
        struct radeon_surf *surface = &rtex->surface;
 
@@ -337,11 +346,11 @@ static void r600_texture_init_metadata(struct si_screen *sscreen,
        }
 }
 
-static void r600_surface_import_metadata(struct si_screen *sscreen,
-                                        struct radeon_surf *surf,
-                                        struct radeon_bo_metadata *metadata,
-                                        enum radeon_surf_mode *array_mode,
-                                        bool *is_scanout)
+static void si_surface_import_metadata(struct si_screen *sscreen,
+                                      struct radeon_surf *surf,
+                                      struct radeon_bo_metadata *metadata,
+                                      enum radeon_surf_mode *array_mode,
+                                      bool *is_scanout)
 {
        if (sscreen->info.chip_class >= GFX9) {
                if (metadata->u.gfx9.swizzle_mode > 0)
@@ -372,24 +381,28 @@ static void r600_surface_import_metadata(struct si_screen *sscreen,
        }
 }
 
-static void r600_eliminate_fast_color_clear(struct r600_common_context *rctx,
-                                           struct r600_texture *rtex)
+void si_eliminate_fast_color_clear(struct si_context *sctx,
+                                  struct r600_texture *rtex)
 {
-       struct si_screen *sscreen = rctx->screen;
-       struct pipe_context *ctx = &rctx->b;
+       struct si_screen *sscreen = sctx->screen;
+       struct pipe_context *ctx = &sctx->b.b;
 
        if (ctx == sscreen->aux_context)
                mtx_lock(&sscreen->aux_context_lock);
 
+       unsigned n = sctx->b.num_decompress_calls;
        ctx->flush_resource(ctx, &rtex->resource.b.b);
-       ctx->flush(ctx, NULL, 0);
+
+       /* Flush only if any fast clear elimination took place. */
+       if (n != sctx->b.num_decompress_calls)
+               ctx->flush(ctx, NULL, 0);
 
        if (ctx == sscreen->aux_context)
                mtx_unlock(&sscreen->aux_context_lock);
 }
 
-static void r600_texture_discard_cmask(struct si_screen *sscreen,
-                                      struct r600_texture *rtex)
+void si_texture_discard_cmask(struct si_screen *sscreen,
+                             struct r600_texture *rtex)
 {
        if (!rtex->cmask.size)
                return;
@@ -411,7 +424,7 @@ static void r600_texture_discard_cmask(struct si_screen *sscreen,
        p_atomic_inc(&sscreen->compressed_colortex_counter);
 }
 
-static bool r600_can_disable_dcc(struct r600_texture *rtex)
+static bool si_can_disable_dcc(struct r600_texture *rtex)
 {
        /* We can't disable DCC if it can be written by another process. */
        return rtex->dcc_offset &&
@@ -419,10 +432,10 @@ static bool r600_can_disable_dcc(struct r600_texture *rtex)
                !(rtex->resource.external_usage & PIPE_HANDLE_USAGE_WRITE));
 }
 
-static bool r600_texture_discard_dcc(struct si_screen *sscreen,
-                                    struct r600_texture *rtex)
+static bool si_texture_discard_dcc(struct si_screen *sscreen,
+                                  struct r600_texture *rtex)
 {
-       if (!r600_can_disable_dcc(rtex))
+       if (!si_can_disable_dcc(rtex))
                return false;
 
        assert(rtex->dcc_separate_buffer == NULL);
@@ -453,36 +466,36 @@ static bool r600_texture_discard_dcc(struct si_screen *sscreen,
  *   context 1 & 2 read garbage, because DCC is disabled, yet there are
  *   compressed tiled
  *
- * \param rctx  the current context if you have one, or rscreen->aux_context
+ * \param sctx  the current context if you have one, or rscreen->aux_context
  *              if you don't.
  */
-bool si_texture_disable_dcc(struct r600_common_context *rctx,
+bool si_texture_disable_dcc(struct si_context *sctx,
                            struct r600_texture *rtex)
 {
-       struct si_screen *sscreen = rctx->screen;
+       struct si_screen *sscreen = sctx->screen;
 
-       if (!r600_can_disable_dcc(rtex))
+       if (!si_can_disable_dcc(rtex))
                return false;
 
-       if (&rctx->b == sscreen->aux_context)
+       if (&sctx->b.b == sscreen->aux_context)
                mtx_lock(&sscreen->aux_context_lock);
 
        /* Decompress DCC. */
-       rctx->decompress_dcc(&rctx->b, rtex);
-       rctx->b.flush(&rctx->b, NULL, 0);
+       si_decompress_dcc(sctx, rtex);
+       sctx->b.b.flush(&sctx->b.b, NULL, 0);
 
-       if (&rctx->b == sscreen->aux_context)
+       if (&sctx->b.b == sscreen->aux_context)
                mtx_unlock(&sscreen->aux_context_lock);
 
-       return r600_texture_discard_dcc(sscreen, rtex);
+       return si_texture_discard_dcc(sscreen, rtex);
 }
 
-static void r600_reallocate_texture_inplace(struct r600_common_context *rctx,
-                                           struct r600_texture *rtex,
-                                           unsigned new_bind_flag,
-                                           bool invalidate_storage)
+static void si_reallocate_texture_inplace(struct si_context *sctx,
+                                         struct r600_texture *rtex,
+                                         unsigned new_bind_flag,
+                                         bool invalidate_storage)
 {
-       struct pipe_screen *screen = rctx->b.screen;
+       struct pipe_screen *screen = sctx->b.b.screen;
        struct r600_texture *new_tex;
        struct pipe_resource templ = rtex->resource.b.b;
        unsigned i;
@@ -497,7 +510,7 @@ static void r600_reallocate_texture_inplace(struct r600_common_context *rctx,
                        return;
 
                /* This fails with MSAA, depth, and compressed textures. */
-               if (r600_choose_tiling(rctx->screen, &templ) !=
+               if (si_choose_tiling(sctx->screen, &templ) !=
                    RADEON_SURF_MODE_LINEAR_ALIGNED)
                        return;
        }
@@ -513,16 +526,16 @@ static void r600_reallocate_texture_inplace(struct r600_common_context *rctx,
 
                        u_box_3d(0, 0, 0,
                                 u_minify(templ.width0, i), u_minify(templ.height0, i),
-                                util_max_layer(&templ, i) + 1, &box);
+                                util_num_layers(&templ, i), &box);
 
-                       rctx->dma_copy(&rctx->b, &new_tex->resource.b.b, i, 0, 0, 0,
+                       sctx->b.dma_copy(&sctx->b.b, &new_tex->resource.b.b, i, 0, 0, 0,
                                       &rtex->resource.b.b, i, &box);
                }
        }
 
        if (new_bind_flag == PIPE_BIND_LINEAR) {
-               r600_texture_discard_cmask(rctx->screen, rtex);
-               r600_texture_discard_dcc(rctx->screen, rtex);
+               si_texture_discard_cmask(sctx->screen, rtex);
+               si_texture_discard_dcc(sctx->screen, rtex);
        }
 
        /* Replace the structure fields of rtex. */
@@ -562,7 +575,7 @@ static void r600_reallocate_texture_inplace(struct r600_common_context *rctx,
 
        r600_texture_reference(&new_tex, NULL);
 
-       p_atomic_inc(&rctx->screen->dirty_tex_counter);
+       p_atomic_inc(&sctx->screen->dirty_tex_counter);
 }
 
 static uint32_t si_get_bo_metadata_word1(struct si_screen *sscreen)
@@ -582,7 +595,7 @@ static void si_query_opaque_metadata(struct si_screen *sscreen,
                PIPE_SWIZZLE_W
        };
        uint32_t desc[8], i;
-       bool is_array = util_resource_is_array_texture(res);
+       bool is_array = util_texture_is_array(res->target);
 
        /* DRM 2.x.x doesn't support this. */
        if (sscreen->info.drm_major != 3)
@@ -646,7 +659,7 @@ static void si_apply_opaque_metadata(struct si_screen *sscreen,
        /* Return if DCC is enabled. The texture should be set up with it
         * already.
         */
-       if (md->size_metadata >= 11 * 4 &&
+       if (md->size_metadata >= 10 * 4 && /* at least 2(header) + 8(desc) dwords */
            md->metadata[0] != 0 &&
            md->metadata[1] == si_get_bo_metadata_word1(sscreen) &&
            G_008F28_COMPRESSION_EN(desc[6])) {
@@ -660,22 +673,23 @@ static void si_apply_opaque_metadata(struct si_screen *sscreen,
        rtex->dcc_offset = 0;
 }
 
-static boolean r600_texture_get_handle(struct pipe_screen* screen,
-                                      struct pipe_context *ctx,
-                                      struct pipe_resource *resource,
-                                      struct winsys_handle *whandle,
-                                       unsigned usage)
+static boolean si_texture_get_handle(struct pipe_screen* screen,
+                                    struct pipe_context *ctx,
+                                    struct pipe_resource *resource,
+                                    struct winsys_handle *whandle,
+                                    unsigned usage)
 {
        struct si_screen *sscreen = (struct si_screen*)screen;
-       struct r600_common_context *rctx;
+       struct si_context *sctx;
        struct r600_resource *res = (struct r600_resource*)resource;
        struct r600_texture *rtex = (struct r600_texture*)resource;
        struct radeon_bo_metadata metadata;
        bool update_metadata = false;
        unsigned stride, offset, slice_size;
+       bool flush = false;
 
        ctx = threaded_context_unwrap_sync(ctx);
-       rctx = (struct r600_common_context*)(ctx ? ctx : sscreen->aux_context);
+       sctx = (struct si_context*)(ctx ? ctx : sscreen->aux_context);
 
        if (resource->target != PIPE_BUFFER) {
                /* This is not supported now, but it might be required for OpenCL
@@ -688,11 +702,12 @@ static boolean r600_texture_get_handle(struct pipe_screen* screen,
                if (sscreen->ws->buffer_is_suballocated(res->buf) ||
                    rtex->surface.tile_swizzle ||
                    (rtex->resource.flags & RADEON_FLAG_NO_INTERPROCESS_SHARING &&
+                    sscreen->info.has_local_buffers &&
                     whandle->type != DRM_API_HANDLE_TYPE_KMS)) {
                        assert(!res->b.is_shared);
-                       r600_reallocate_texture_inplace(rctx, rtex,
+                       si_reallocate_texture_inplace(sctx, rtex,
                                                        PIPE_BIND_SHARED, false);
-                       rctx->b.flush(&rctx->b, NULL, 0);
+                       flush = true;
                        assert(res->b.b.bind & PIPE_BIND_SHARED);
                        assert(res->flags & RADEON_FLAG_NO_SUBALLOC);
                        assert(!(res->flags & RADEON_FLAG_NO_INTERPROCESS_SHARING));
@@ -704,25 +719,30 @@ static boolean r600_texture_get_handle(struct pipe_screen* screen,
                 * access.
                 */
                if (usage & PIPE_HANDLE_USAGE_WRITE && rtex->dcc_offset) {
-                       if (si_texture_disable_dcc(rctx, rtex))
+                       if (si_texture_disable_dcc(sctx, rtex)) {
                                update_metadata = true;
+                               /* si_texture_disable_dcc flushes the context */
+                               flush = false;
+                       }
                }
 
                if (!(usage & PIPE_HANDLE_USAGE_EXPLICIT_FLUSH) &&
                    (rtex->cmask.size || rtex->dcc_offset)) {
                        /* Eliminate fast clear (both CMASK and DCC) */
-                       r600_eliminate_fast_color_clear(rctx, rtex);
+                       si_eliminate_fast_color_clear(sctx, rtex);
+                       /* eliminate_fast_color_clear flushes the context */
+                       flush = false;
 
                        /* Disable CMASK if flush_resource isn't going
                         * to be called.
                         */
                        if (rtex->cmask.size)
-                               r600_texture_discard_cmask(sscreen, rtex);
+                               si_texture_discard_cmask(sscreen, rtex);
                }
 
                /* Set metadata. */
                if (!res->b.is_shared || update_metadata) {
-                       r600_texture_init_metadata(sscreen, rtex, &metadata);
+                       si_texture_init_metadata(sscreen, rtex, &metadata);
                        si_query_opaque_metadata(sscreen, rtex, &metadata);
 
                        sscreen->ws->buffer_set_metadata(res->buf, &metadata);
@@ -740,8 +760,12 @@ static boolean r600_texture_get_handle(struct pipe_screen* screen,
                        slice_size = (uint64_t)rtex->surface.u.legacy.level[0].slice_size_dw * 4;
                }
        } else {
+               /* Buffer exports are for the OpenCL interop. */
                /* Move a suballocated buffer into a non-suballocated allocation. */
-               if (sscreen->ws->buffer_is_suballocated(res->buf)) {
+               if (sscreen->ws->buffer_is_suballocated(res->buf) ||
+                   /* A DMABUF export always fails if the BO is local. */
+                   (rtex->resource.flags & RADEON_FLAG_NO_INTERPROCESS_SHARING &&
+                    sscreen->info.has_local_buffers)) {
                        assert(!res->b.is_shared);
 
                        /* Allocate a new buffer with PIPE_BIND_SHARED. */
@@ -756,10 +780,11 @@ static boolean r600_texture_get_handle(struct pipe_screen* screen,
                        /* Copy the old buffer contents to the new one. */
                        struct pipe_box box;
                        u_box_1d(0, newb->width0, &box);
-                       rctx->b.resource_copy_region(&rctx->b, newb, 0, 0, 0, 0,
+                       sctx->b.b.resource_copy_region(&sctx->b.b, newb, 0, 0, 0, 0,
                                                     &res->b.b, 0, &box);
+                       flush = true;
                        /* Move the new buffer storage to the old pipe_resource. */
-                       si_replace_buffer_storage(&rctx->b, &res->b.b, newb);
+                       si_replace_buffer_storage(&sctx->b.b, &res->b.b, newb);
                        pipe_resource_reference(&newb, NULL);
 
                        assert(res->b.b.bind & PIPE_BIND_SHARED);
@@ -772,6 +797,9 @@ static boolean r600_texture_get_handle(struct pipe_screen* screen,
                slice_size = 0;
        }
 
+       if (flush)
+               sctx->b.b.flush(&sctx->b.b, NULL, 0);
+
        if (res->b.is_shared) {
                /* USAGE_EXPLICIT_FLUSH must be cleared if at least one user
                 * doesn't set it.
@@ -788,8 +816,8 @@ static boolean r600_texture_get_handle(struct pipe_screen* screen,
                                              slice_size, whandle);
 }
 
-static void r600_texture_destroy(struct pipe_screen *screen,
-                                struct pipe_resource *ptex)
+static void si_texture_destroy(struct pipe_screen *screen,
+                              struct pipe_resource *ptex)
 {
        struct r600_texture *rtex = (struct r600_texture*)ptex;
        struct r600_resource *resource = &rtex->resource;
@@ -805,7 +833,7 @@ static void r600_texture_destroy(struct pipe_screen *screen,
        FREE(rtex);
 }
 
-static const struct u_resource_vtbl r600_texture_vtbl;
+static const struct u_resource_vtbl si_texture_vtbl;
 
 /* The number of samples can be specified independently of the texture. */
 void si_texture_get_fmask_info(struct si_screen *sscreen,
@@ -823,6 +851,7 @@ void si_texture_get_fmask_info(struct si_screen *sscreen,
        if (sscreen->info.chip_class >= GFX9) {
                out->alignment = rtex->surface.u.gfx9.fmask_alignment;
                out->size = rtex->surface.u.gfx9.fmask_size;
+               out->tile_swizzle = rtex->surface.u.gfx9.fmask_tile_swizzle;
                return;
        }
 
@@ -838,13 +867,13 @@ void si_texture_get_fmask_info(struct si_screen *sscreen,
                bpe = 4;
                break;
        default:
-               R600_ERR("Invalid sample count for FMASK allocation.\n");
+               PRINT_ERR("Invalid sample count for FMASK allocation.\n");
                return;
        }
 
        if (sscreen->ws->surface_init(sscreen->ws, &templ, flags, bpe,
                                      RADEON_SURF_MODE_2D, &fmask)) {
-               R600_ERR("Got error in surface_init while allocating FMASK.\n");
+               PRINT_ERR("Got error in surface_init while allocating FMASK.\n");
                return;
        }
 
@@ -862,8 +891,8 @@ void si_texture_get_fmask_info(struct si_screen *sscreen,
        out->size = fmask.surf_size;
 }
 
-static void r600_texture_allocate_fmask(struct si_screen *sscreen,
-                                       struct r600_texture *rtex)
+static void si_texture_allocate_fmask(struct si_screen *sscreen,
+                                     struct r600_texture *rtex)
 {
        si_texture_get_fmask_info(sscreen, rtex,
                                    rtex->resource.b.b.nr_samples, &rtex->fmask);
@@ -922,12 +951,12 @@ void si_texture_get_cmask_info(struct si_screen *sscreen,
                out->slice_tile_max -= 1;
 
        out->alignment = MAX2(256, base_align);
-       out->size = (util_max_layer(&rtex->resource.b.b, 0) + 1) *
+       out->size = util_num_layers(&rtex->resource.b.b, 0) *
                    align(slice_bytes, base_align);
 }
 
-static void r600_texture_allocate_cmask(struct si_screen *sscreen,
-                                       struct r600_texture *rtex)
+static void si_texture_allocate_cmask(struct si_screen *sscreen,
+                                     struct r600_texture *rtex)
 {
        si_texture_get_cmask_info(sscreen, rtex, &rtex->cmask);
 
@@ -937,8 +966,8 @@ static void r600_texture_allocate_cmask(struct si_screen *sscreen,
        rtex->cb_color_info |= S_028C70_FAST_CLEAR(1);
 }
 
-static void r600_texture_get_htile_size(struct si_screen *sscreen,
-                                       struct r600_texture *rtex)
+static void si_texture_get_htile_size(struct si_screen *sscreen,
+                                     struct r600_texture *rtex)
 {
        unsigned cl_width, cl_height, width, height;
        unsigned slice_elements, slice_bytes, pipe_interleave_bytes, base_align;
@@ -1001,15 +1030,15 @@ static void r600_texture_get_htile_size(struct si_screen *sscreen,
 
        rtex->surface.htile_alignment = base_align;
        rtex->surface.htile_size =
-               (util_max_layer(&rtex->resource.b.b, 0) + 1) *
+               util_num_layers(&rtex->resource.b.b, 0) *
                align(slice_bytes, base_align);
 }
 
-static void r600_texture_allocate_htile(struct si_screen *sscreen,
-                                       struct r600_texture *rtex)
+static void si_texture_allocate_htile(struct si_screen *sscreen,
+                                     struct r600_texture *rtex)
 {
        if (sscreen->info.chip_class <= VI && !rtex->tc_compatible_htile)
-               r600_texture_get_htile_size(sscreen, rtex);
+               si_texture_get_htile_size(sscreen, rtex);
 
        if (!rtex->surface.htile_size)
                return;
@@ -1168,10 +1197,10 @@ void si_print_texture_info(struct si_screen *sscreen,
 
 /* Common processing for r600_texture_create and r600_texture_from_handle */
 static struct r600_texture *
-r600_texture_create_object(struct pipe_screen *screen,
-                          const struct pipe_resource *base,
-                          struct pb_buffer *buf,
-                          struct radeon_surf *surface)
+si_texture_create_object(struct pipe_screen *screen,
+                        const struct pipe_resource *base,
+                        struct pb_buffer *buf,
+                        struct radeon_surf *surface)
 {
        struct r600_texture *rtex;
        struct r600_resource *resource;
@@ -1184,7 +1213,7 @@ r600_texture_create_object(struct pipe_screen *screen,
        resource = &rtex->resource;
        resource->b.b = *base;
        resource->b.b.next = NULL;
-       resource->b.vtbl = &r600_texture_vtbl;
+       resource->b.vtbl = &si_texture_vtbl;
        pipe_reference_init(&resource->b.b.reference, 1);
        resource->b.b.screen = screen;
 
@@ -1232,20 +1261,21 @@ r600_texture_create_object(struct pipe_screen *screen,
                        rtex->can_sample_s = !rtex->surface.u.legacy.stencil_adjusted;
                }
 
-               if (!(base->flags & (R600_RESOURCE_FLAG_TRANSFER |
-                                    R600_RESOURCE_FLAG_FLUSHED_DEPTH))) {
+               if (!(base->flags & (SI_RESOURCE_FLAG_TRANSFER |
+                                    SI_RESOURCE_FLAG_FLUSHED_DEPTH))) {
                        rtex->db_compatible = true;
 
                        if (!(sscreen->debug_flags & DBG(NO_HYPERZ)))
-                               r600_texture_allocate_htile(sscreen, rtex);
+                               si_texture_allocate_htile(sscreen, rtex);
                }
        } else {
-               if (base->nr_samples > 1) {
-                       if (!buf) {
-                               r600_texture_allocate_fmask(sscreen, rtex);
-                               r600_texture_allocate_cmask(sscreen, rtex);
-                               rtex->cmask_buffer = &rtex->resource;
-                       }
+               if (base->nr_samples > 1 &&
+                   !buf &&
+                   !(sscreen->debug_flags & DBG(NO_FMASK))) {
+                       si_texture_allocate_fmask(sscreen, rtex);
+                       si_texture_allocate_cmask(sscreen, rtex);
+                       rtex->cmask_buffer = &rtex->resource;
+
                        if (!rtex->fmask.size || !rtex->cmask.size) {
                                FREE(rtex);
                                return NULL;
@@ -1320,7 +1350,7 @@ r600_texture_create_object(struct pipe_screen *screen,
                fprintf(stderr, "VM start=0x%"PRIX64"  end=0x%"PRIX64" | Texture %ix%ix%i, %i levels, %i samples, %s\n",
                        rtex->resource.gpu_address,
                        rtex->resource.gpu_address + rtex->resource.buf->size,
-                       base->width0, base->height0, util_max_layer(base, 0)+1, base->last_level+1,
+                       base->width0, base->height0, util_num_layers(base, 0), base->last_level+1,
                        base->nr_samples ? base->nr_samples : 1, util_format_short_name(base->format));
        }
 
@@ -1338,20 +1368,20 @@ r600_texture_create_object(struct pipe_screen *screen,
 }
 
 static enum radeon_surf_mode
-r600_choose_tiling(struct si_screen *sscreen,
+si_choose_tiling(struct si_screen *sscreen,
                   const struct pipe_resource *templ)
 {
        const struct util_format_description *desc = util_format_description(templ->format);
-       bool force_tiling = templ->flags & R600_RESOURCE_FLAG_FORCE_TILING;
+       bool force_tiling = templ->flags & SI_RESOURCE_FLAG_FORCE_TILING;
        bool is_depth_stencil = util_format_is_depth_or_stencil(templ->format) &&
-                               !(templ->flags & R600_RESOURCE_FLAG_FLUSHED_DEPTH);
+                               !(templ->flags & SI_RESOURCE_FLAG_FLUSHED_DEPTH);
 
        /* MSAA resources must be 2D tiled. */
        if (templ->nr_samples > 1)
                return RADEON_SURF_MODE_2D;
 
        /* Transfer resources should be linear. */
-       if (templ->flags & R600_RESOURCE_FLAG_TRANSFER)
+       if (templ->flags & SI_RESOURCE_FLAG_TRANSFER)
                return RADEON_SURF_MODE_LINEAR_ALIGNED;
 
        /* Avoid Z/S decompress blits by forcing TC-compatible HTILE on VI,
@@ -1411,7 +1441,7 @@ struct pipe_resource *si_texture_create(struct pipe_screen *screen,
 {
        struct si_screen *sscreen = (struct si_screen*)screen;
        struct radeon_surf surface = {0};
-       bool is_flushed_depth = templ->flags & R600_RESOURCE_FLAG_FLUSHED_DEPTH;
+       bool is_flushed_depth = templ->flags & SI_RESOURCE_FLAG_FLUSHED_DEPTH;
        bool tc_compatible_htile =
                sscreen->info.chip_class >= VI &&
                (templ->flags & PIPE_RESOURCE_FLAG_TEXTURING_MORE_LIKELY) &&
@@ -1422,8 +1452,8 @@ struct pipe_resource *si_texture_create(struct pipe_screen *screen,
 
        int r;
 
-       r = r600_init_surface(sscreen, &surface, templ,
-                             r600_choose_tiling(sscreen, templ), 0, 0,
+       r = si_init_surface(sscreen, &surface, templ,
+                             si_choose_tiling(sscreen, templ), 0, 0,
                              false, false, is_flushed_depth,
                              tc_compatible_htile);
        if (r) {
@@ -1431,13 +1461,13 @@ struct pipe_resource *si_texture_create(struct pipe_screen *screen,
        }
 
        return (struct pipe_resource *)
-              r600_texture_create_object(screen, templ, NULL, &surface);
+              si_texture_create_object(screen, templ, NULL, &surface);
 }
 
-static struct pipe_resource *r600_texture_from_handle(struct pipe_screen *screen,
-                                                     const struct pipe_resource *templ,
-                                                     struct winsys_handle *whandle,
-                                                      unsigned usage)
+static struct pipe_resource *si_texture_from_handle(struct pipe_screen *screen,
+                                                   const struct pipe_resource *templ,
+                                                   struct winsys_handle *whandle,
+                                                   unsigned usage)
 {
        struct si_screen *sscreen = (struct si_screen*)screen;
        struct pb_buffer *buf = NULL;
@@ -1459,16 +1489,16 @@ static struct pipe_resource *r600_texture_from_handle(struct pipe_screen *screen
                return NULL;
 
        sscreen->ws->buffer_get_metadata(buf, &metadata);
-       r600_surface_import_metadata(sscreen, &surface, &metadata,
+       si_surface_import_metadata(sscreen, &surface, &metadata,
                                     &array_mode, &is_scanout);
 
-       r = r600_init_surface(sscreen, &surface, templ, array_mode, stride,
+       r = si_init_surface(sscreen, &surface, templ, array_mode, stride,
                              offset, true, is_scanout, false, false);
        if (r) {
                return NULL;
        }
 
-       rtex = r600_texture_create_object(screen, templ, buf, &surface);
+       rtex = si_texture_create_object(screen, templ, buf, &surface);
        if (!rtex)
                return NULL;
 
@@ -1535,14 +1565,14 @@ bool si_init_flushed_depth_texture(struct pipe_context *ctx,
        resource.nr_samples = texture->nr_samples;
        resource.usage = staging ? PIPE_USAGE_STAGING : PIPE_USAGE_DEFAULT;
        resource.bind = texture->bind & ~PIPE_BIND_DEPTH_STENCIL;
-       resource.flags = texture->flags | R600_RESOURCE_FLAG_FLUSHED_DEPTH;
+       resource.flags = texture->flags | SI_RESOURCE_FLAG_FLUSHED_DEPTH;
 
        if (staging)
-               resource.flags |= R600_RESOURCE_FLAG_TRANSFER;
+               resource.flags |= SI_RESOURCE_FLAG_TRANSFER;
 
        *flushed_depth_texture = (struct r600_texture *)ctx->screen->resource_create(ctx->screen, &resource);
        if (*flushed_depth_texture == NULL) {
-               R600_ERR("failed to create temporary texture to hold flushed depth\n");
+               PRINT_ERR("failed to create temporary texture to hold flushed depth\n");
                return false;
        }
        return true;
@@ -1553,10 +1583,10 @@ bool si_init_flushed_depth_texture(struct pipe_context *ctx,
  * 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)
+static void si_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;
@@ -1564,7 +1594,7 @@ static void r600_init_temp_resource_from_box(struct pipe_resource *res,
        res->height0 = box->height;
        res->depth0 = 1;
        res->array_size = 1;
-       res->usage = flags & R600_RESOURCE_FLAG_TRANSFER ? PIPE_USAGE_STAGING : PIPE_USAGE_DEFAULT;
+       res->usage = flags & SI_RESOURCE_FLAG_TRANSFER ? PIPE_USAGE_STAGING : PIPE_USAGE_DEFAULT;
        res->flags = flags;
 
        /* We must set the correct texture target and dimensions for a 3D box. */
@@ -1576,10 +1606,10 @@ static void r600_init_temp_resource_from_box(struct pipe_resource *res,
        }
 }
 
-static bool r600_can_invalidate_texture(struct si_screen *sscreen,
-                                       struct r600_texture *rtex,
-                                       unsigned transfer_usage,
-                                       const struct pipe_box *box)
+static bool si_can_invalidate_texture(struct si_screen *sscreen,
+                                     struct r600_texture *rtex,
+                                     unsigned transfer_usage,
+                                     const struct pipe_box *box)
 {
        return !rtex->resource.b.is_shared &&
                !(transfer_usage & PIPE_TRANSFER_READ) &&
@@ -1590,10 +1620,10 @@ static bool r600_can_invalidate_texture(struct si_screen *sscreen,
                                                 box->depth);
 }
 
-static void r600_texture_invalidate_storage(struct r600_common_context *rctx,
-                                           struct r600_texture *rtex)
+static void si_texture_invalidate_storage(struct si_context *sctx,
+                                         struct r600_texture *rtex)
 {
-       struct si_screen *sscreen = rctx->screen;
+       struct si_screen *sscreen = sctx->screen;
 
        /* There is no point in discarding depth and tiled buffers. */
        assert(!rtex->is_depth);
@@ -1608,17 +1638,17 @@ static void r600_texture_invalidate_storage(struct r600_common_context *rctx,
 
        p_atomic_inc(&sscreen->dirty_tex_counter);
 
-       rctx->num_alloc_tex_transfer_bytes += rtex->size;
+       sctx->b.num_alloc_tex_transfer_bytes += rtex->size;
 }
 
-static void *r600_texture_transfer_map(struct pipe_context *ctx,
-                                      struct pipe_resource *texture,
-                                      unsigned level,
-                                      unsigned usage,
-                                      const struct pipe_box *box,
-                                      struct pipe_transfer **ptransfer)
+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_common_context *rctx = (struct r600_common_context*)ctx;
+       struct si_context *sctx = (struct si_context*)ctx;
        struct r600_texture *rtex = (struct r600_texture*)texture;
        struct r600_transfer *trans;
        struct r600_resource *buf;
@@ -1626,7 +1656,7 @@ static void *r600_texture_transfer_map(struct pipe_context *ctx,
        char *map;
        bool use_staging_texture = false;
 
-       assert(!(texture->flags & R600_RESOURCE_FLAG_TRANSFER));
+       assert(!(texture->flags & SI_RESOURCE_FLAG_TRANSFER));
        assert(box->width && box->height && box->depth);
 
        /* Depth textures use staging unconditionally. */
@@ -1635,15 +1665,15 @@ static void *r600_texture_transfer_map(struct pipe_context *ctx,
                 * On dGPUs, the staging texture is always faster.
                 * Only count uploads that are at least 4x4 pixels large.
                 */
-               if (!rctx->screen->info.has_dedicated_vram &&
+               if (!sctx->screen->info.has_dedicated_vram &&
                    level == 0 &&
                    box->width >= 4 && box->height >= 4 &&
                    p_atomic_inc_return(&rtex->num_level0_transfers) == 10) {
                        bool can_invalidate =
-                               r600_can_invalidate_texture(rctx->screen, rtex,
+                               si_can_invalidate_texture(sctx->screen, rtex,
                                                            usage, box);
 
-                       r600_reallocate_texture_inplace(rctx, rtex,
+                       si_reallocate_texture_inplace(sctx, rtex,
                                                        PIPE_BIND_LINEAR,
                                                        can_invalidate);
                }
@@ -1664,14 +1694,14 @@ static void *r600_texture_transfer_map(struct pipe_context *ctx,
                                rtex->resource.domains & RADEON_DOMAIN_VRAM ||
                                rtex->resource.flags & RADEON_FLAG_GTT_WC;
                /* Write & linear only: */
-               else if (si_rings_is_buffer_referenced(rctx, rtex->resource.buf,
-                                                        RADEON_USAGE_READWRITE) ||
-                        !rctx->ws->buffer_wait(rtex->resource.buf, 0,
+               else if (si_rings_is_buffer_referenced(sctx, rtex->resource.buf,
+                                                      RADEON_USAGE_READWRITE) ||
+                        !sctx->b.ws->buffer_wait(rtex->resource.buf, 0,
                                                RADEON_USAGE_READWRITE)) {
                        /* It's busy. */
-                       if (r600_can_invalidate_texture(rctx->screen, rtex,
+                       if (si_can_invalidate_texture(sctx->screen, rtex,
                                                        usage, box))
-                               r600_texture_invalidate_storage(rctx, rtex);
+                               si_texture_invalidate_storage(sctx, rtex);
                        else
                                use_staging_texture = true;
                }
@@ -1701,10 +1731,10 @@ static void *r600_texture_transfer_map(struct pipe_context *ctx,
                         */
                        struct pipe_resource resource;
 
-                       r600_init_temp_resource_from_box(&resource, texture, box, level, 0);
+                       si_init_temp_resource_from_box(&resource, texture, box, level, 0);
 
                        if (!si_init_flushed_depth_texture(ctx, &resource, &staging_depth)) {
-                               R600_ERR("failed to create temporary texture to hold untiled copy\n");
+                               PRINT_ERR("failed to create temporary texture to hold untiled copy\n");
                                FREE(trans);
                                return NULL;
                        }
@@ -1712,36 +1742,36 @@ static void *r600_texture_transfer_map(struct pipe_context *ctx,
                        if (usage & PIPE_TRANSFER_READ) {
                                struct pipe_resource *temp = ctx->screen->resource_create(ctx->screen, &resource);
                                if (!temp) {
-                                       R600_ERR("failed to create a temporary depth texture\n");
+                                       PRINT_ERR("failed to create a temporary depth texture\n");
                                        FREE(trans);
                                        return NULL;
                                }
 
-                               r600_copy_region_with_blit(ctx, temp, 0, 0, 0, 0, texture, level, box);
-                               rctx->blit_decompress_depth(ctx, (struct r600_texture*)temp, staging_depth,
-                                                           0, 0, 0, box->depth, 0, 0);
+                               si_copy_region_with_blit(ctx, temp, 0, 0, 0, 0, texture, level, box);
+                               si_blit_decompress_depth(ctx, (struct r600_texture*)temp, staging_depth,
+                                                        0, 0, 0, box->depth, 0, 0);
                                pipe_resource_reference(&temp, NULL);
                        }
 
                        /* Just get the strides. */
-                       r600_texture_get_offset(rctx->screen, staging_depth, level, NULL,
+                       si_texture_get_offset(sctx->screen, staging_depth, level, NULL,
                                                &trans->b.b.stride,
                                                &trans->b.b.layer_stride);
                } else {
                        /* XXX: only readback the rectangle which is being mapped? */
                        /* XXX: when discard is true, no need to read back from depth texture */
                        if (!si_init_flushed_depth_texture(ctx, texture, &staging_depth)) {
-                               R600_ERR("failed to create temporary texture to hold untiled copy\n");
+                               PRINT_ERR("failed to create temporary texture to hold untiled copy\n");
                                FREE(trans);
                                return NULL;
                        }
 
-                       rctx->blit_decompress_depth(ctx, rtex, staging_depth,
-                                                   level, level,
-                                                   box->z, box->z + box->depth - 1,
-                                                   0, 0);
+                       si_blit_decompress_depth(ctx, rtex, staging_depth,
+                                                level, level,
+                                                box->z, box->z + box->depth - 1,
+                                                0, 0);
 
-                       offset = r600_texture_get_offset(rctx->screen, staging_depth,
+                       offset = si_texture_get_offset(sctx->screen, staging_depth,
                                                         level, box,
                                                         &trans->b.b.stride,
                                                         &trans->b.b.layer_stride);
@@ -1753,40 +1783,40 @@ static void *r600_texture_transfer_map(struct pipe_context *ctx,
                struct pipe_resource resource;
                struct r600_texture *staging;
 
-               r600_init_temp_resource_from_box(&resource, texture, box, level,
-                                                R600_RESOURCE_FLAG_TRANSFER);
+               si_init_temp_resource_from_box(&resource, texture, box, level,
+                                                SI_RESOURCE_FLAG_TRANSFER);
                resource.usage = (usage & PIPE_TRANSFER_READ) ?
                        PIPE_USAGE_STAGING : PIPE_USAGE_STREAM;
 
                /* Create the temporary texture. */
                staging = (struct r600_texture*)ctx->screen->resource_create(ctx->screen, &resource);
                if (!staging) {
-                       R600_ERR("failed to create temporary texture to hold untiled copy\n");
+                       PRINT_ERR("failed to create temporary texture to hold untiled copy\n");
                        FREE(trans);
                        return NULL;
                }
                trans->staging = &staging->resource;
 
                /* Just get the strides. */
-               r600_texture_get_offset(rctx->screen, staging, 0, NULL,
+               si_texture_get_offset(sctx->screen, staging, 0, NULL,
                                        &trans->b.b.stride,
                                        &trans->b.b.layer_stride);
 
                if (usage & PIPE_TRANSFER_READ)
-                       r600_copy_to_staging_texture(ctx, trans);
+                       si_copy_to_staging_texture(ctx, trans);
                else
                        usage |= PIPE_TRANSFER_UNSYNCHRONIZED;
 
                buf = trans->staging;
        } else {
                /* the resource is mapped directly */
-               offset = r600_texture_get_offset(rctx->screen, rtex, level, box,
+               offset = si_texture_get_offset(sctx->screen, rtex, level, box,
                                                 &trans->b.b.stride,
                                                 &trans->b.b.layer_stride);
                buf = &rtex->resource;
        }
 
-       if (!(map = si_buffer_map_sync_with_rings(rctx, buf, usage))) {
+       if (!(map = si_buffer_map_sync_with_rings(sctx, buf, usage))) {
                r600_resource_reference(&trans->staging, NULL);
                FREE(trans);
                return NULL;
@@ -1796,10 +1826,10 @@ static void *r600_texture_transfer_map(struct pipe_context *ctx,
        return map + offset;
 }
 
-static void r600_texture_transfer_unmap(struct pipe_context *ctx,
-                                       struct pipe_transfer* transfer)
+static void si_texture_transfer_unmap(struct pipe_context *ctx,
+                                     struct pipe_transfer* transfer)
 {
-       struct r600_common_context *rctx = (struct r600_common_context*)ctx;
+       struct si_context *sctx = (struct si_context*)ctx;
        struct r600_transfer *rtransfer = (struct r600_transfer*)transfer;
        struct pipe_resource *texture = transfer->resource;
        struct r600_texture *rtex = (struct r600_texture*)texture;
@@ -1811,12 +1841,12 @@ static void r600_texture_transfer_unmap(struct pipe_context *ctx,
                                                  &rtransfer->staging->b.b, transfer->level,
                                                  &transfer->box);
                } else {
-                       r600_copy_from_staging_texture(ctx, rtransfer);
+                       si_copy_from_staging_texture(ctx, rtransfer);
                }
        }
 
        if (rtransfer->staging) {
-               rctx->num_alloc_tex_transfer_bytes += rtransfer->staging->buf->size;
+               sctx->b.num_alloc_tex_transfer_bytes += rtransfer->staging->buf->size;
                r600_resource_reference(&rtransfer->staging, NULL);
        }
 
@@ -1833,36 +1863,33 @@ static void r600_texture_transfer_unmap(struct pipe_context *ctx,
         *
         * The result is that the kernel memory manager is never a bottleneck.
         */
-       if (rctx->num_alloc_tex_transfer_bytes > rctx->screen->info.gart_size / 4) {
-               rctx->gfx.flush(rctx, PIPE_FLUSH_ASYNC, NULL);
-               rctx->num_alloc_tex_transfer_bytes = 0;
+       if (sctx->b.num_alloc_tex_transfer_bytes > sctx->screen->info.gart_size / 4) {
+               si_flush_gfx_cs(sctx, PIPE_FLUSH_ASYNC, NULL);
+               sctx->b.num_alloc_tex_transfer_bytes = 0;
        }
 
        pipe_resource_reference(&transfer->resource, NULL);
        FREE(transfer);
 }
 
-static const struct u_resource_vtbl r600_texture_vtbl =
+static const struct u_resource_vtbl si_texture_vtbl =
 {
        NULL,                           /* get_handle */
-       r600_texture_destroy,           /* resource_destroy */
-       r600_texture_transfer_map,      /* transfer_map */
+       si_texture_destroy,             /* resource_destroy */
+       si_texture_transfer_map,        /* transfer_map */
        u_default_transfer_flush_region, /* transfer_flush_region */
-       r600_texture_transfer_unmap,    /* transfer_unmap */
+       si_texture_transfer_unmap,      /* transfer_unmap */
 };
 
 /* DCC channel type categories within which formats can be reinterpreted
  * while keeping the same DCC encoding. The swizzle must also match. */
 enum dcc_channel_type {
-       dcc_channel_float32,
-       dcc_channel_uint32,
-       dcc_channel_sint32,
-       dcc_channel_float16,
-       dcc_channel_uint16,
-       dcc_channel_sint16,
+       dcc_channel_float,
+       /* uint and sint can be merged if we never use TC-compatible DCC clear
+        * encoding with the clear value of 1. */
+       dcc_channel_uint,
+       dcc_channel_sint,
        dcc_channel_uint_10_10_10_2,
-       dcc_channel_uint8,
-       dcc_channel_sint8,
        dcc_channel_incompatible,
 };
 
@@ -1881,23 +1908,15 @@ vi_get_dcc_channel_type(const struct util_format_description *desc)
 
        switch (desc->channel[i].size) {
        case 32:
-               if (desc->channel[i].type == UTIL_FORMAT_TYPE_FLOAT)
-                       return dcc_channel_float32;
-               if (desc->channel[i].type == UTIL_FORMAT_TYPE_UNSIGNED)
-                       return dcc_channel_uint32;
-               return dcc_channel_sint32;
        case 16:
+       case 8:
                if (desc->channel[i].type == UTIL_FORMAT_TYPE_FLOAT)
-                       return dcc_channel_float16;
+                       return dcc_channel_float;
                if (desc->channel[i].type == UTIL_FORMAT_TYPE_UNSIGNED)
-                       return dcc_channel_uint16;
-               return dcc_channel_sint16;
+                       return dcc_channel_uint;
+               return dcc_channel_sint;
        case 10:
                return dcc_channel_uint_10_10_10_2;
-       case 8:
-               if (desc->channel[i].type == UTIL_FORMAT_TYPE_UNSIGNED)
-                       return dcc_channel_uint8;
-               return dcc_channel_sint8;
        default:
                return dcc_channel_incompatible;
        }
@@ -1947,7 +1966,7 @@ bool vi_dcc_formats_are_incompatible(struct pipe_resource *tex,
 
 /* This can't be merged with the above function, because
  * vi_dcc_formats_compatible should be called only when DCC is enabled. */
-void vi_disable_dcc_if_incompatible_format(struct r600_common_context *rctx,
+void vi_disable_dcc_if_incompatible_format(struct si_context *sctx,
                                           struct pipe_resource *tex,
                                           unsigned level,
                                           enum pipe_format view_format)
@@ -1955,8 +1974,8 @@ void vi_disable_dcc_if_incompatible_format(struct r600_common_context *rctx,
        struct r600_texture *rtex = (struct r600_texture *)tex;
 
        if (vi_dcc_formats_are_incompatible(tex, level, view_format))
-               if (!si_texture_disable_dcc(rctx, (struct r600_texture*)tex))
-                       rctx->decompress_dcc(&rctx->b, rtex);
+               if (!si_texture_disable_dcc(sctx, (struct r600_texture*)tex))
+                       si_decompress_dcc(sctx, rtex);
 }
 
 struct pipe_surface *si_create_surface_custom(struct pipe_context *pipe,
@@ -1991,9 +2010,9 @@ struct pipe_surface *si_create_surface_custom(struct pipe_context *pipe,
        return &surface->base;
 }
 
-static struct pipe_surface *r600_create_surface(struct pipe_context *pipe,
-                                               struct pipe_resource *tex,
-                                               const struct pipe_surface *templ)
+static struct pipe_surface *si_create_surface(struct pipe_context *pipe,
+                                             struct pipe_resource *tex,
+                                             const struct pipe_surface *templ)
 {
        unsigned level = templ->u.tex.level;
        unsigned width = u_minify(tex->width0, level);
@@ -2029,8 +2048,8 @@ static struct pipe_surface *r600_create_surface(struct pipe_context *pipe,
                                          width, height);
 }
 
-static void r600_surface_destroy(struct pipe_context *pipe,
-                                struct pipe_surface *surface)
+static void si_surface_destroy(struct pipe_context *pipe,
+                              struct pipe_surface *surface)
 {
        pipe_resource_reference(&surface->texture, NULL);
        FREE(surface);
@@ -2098,49 +2117,49 @@ unsigned si_translate_colorswap(enum pipe_format format, bool do_endian_swap)
 
 /* PIPELINE_STAT-BASED DCC ENABLEMENT FOR DISPLAYABLE SURFACES */
 
-static void vi_dcc_clean_up_context_slot(struct r600_common_context *rctx,
+static void vi_dcc_clean_up_context_slot(struct si_context *sctx,
                                         int slot)
 {
        int i;
 
-       if (rctx->dcc_stats[slot].query_active)
-               vi_separate_dcc_stop_query(&rctx->b,
-                                          rctx->dcc_stats[slot].tex);
+       if (sctx->b.dcc_stats[slot].query_active)
+               vi_separate_dcc_stop_query(sctx,
+                                          sctx->b.dcc_stats[slot].tex);
 
-       for (i = 0; i < ARRAY_SIZE(rctx->dcc_stats[slot].ps_stats); i++)
-               if (rctx->dcc_stats[slot].ps_stats[i]) {
-                       rctx->b.destroy_query(&rctx->b,
-                                             rctx->dcc_stats[slot].ps_stats[i]);
-                       rctx->dcc_stats[slot].ps_stats[i] = NULL;
+       for (i = 0; i < ARRAY_SIZE(sctx->b.dcc_stats[slot].ps_stats); i++)
+               if (sctx->b.dcc_stats[slot].ps_stats[i]) {
+                       sctx->b.b.destroy_query(&sctx->b.b,
+                                             sctx->b.dcc_stats[slot].ps_stats[i]);
+                       sctx->b.dcc_stats[slot].ps_stats[i] = NULL;
                }
 
-       r600_texture_reference(&rctx->dcc_stats[slot].tex, NULL);
+       r600_texture_reference(&sctx->b.dcc_stats[slot].tex, NULL);
 }
 
 /**
  * Return the per-context slot where DCC statistics queries for the texture live.
  */
-static unsigned vi_get_context_dcc_stats_index(struct r600_common_context *rctx,
+static unsigned vi_get_context_dcc_stats_index(struct si_context *sctx,
                                               struct r600_texture *tex)
 {
        int i, empty_slot = -1;
 
        /* Remove zombie textures (textures kept alive by this array only). */
-       for (i = 0; i < ARRAY_SIZE(rctx->dcc_stats); i++)
-               if (rctx->dcc_stats[i].tex &&
-                   rctx->dcc_stats[i].tex->resource.b.b.reference.count == 1)
-                       vi_dcc_clean_up_context_slot(rctx, i);
+       for (i = 0; i < ARRAY_SIZE(sctx->b.dcc_stats); i++)
+               if (sctx->b.dcc_stats[i].tex &&
+                   sctx->b.dcc_stats[i].tex->resource.b.b.reference.count == 1)
+                       vi_dcc_clean_up_context_slot(sctx, i);
 
        /* Find the texture. */
-       for (i = 0; i < ARRAY_SIZE(rctx->dcc_stats); i++) {
+       for (i = 0; i < ARRAY_SIZE(sctx->b.dcc_stats); i++) {
                /* Return if found. */
-               if (rctx->dcc_stats[i].tex == tex) {
-                       rctx->dcc_stats[i].last_use_timestamp = os_time_get();
+               if (sctx->b.dcc_stats[i].tex == tex) {
+                       sctx->b.dcc_stats[i].last_use_timestamp = os_time_get();
                        return i;
                }
 
                /* Record the first seen empty slot. */
-               if (empty_slot == -1 && !rctx->dcc_stats[i].tex)
+               if (empty_slot == -1 && !sctx->b.dcc_stats[i].tex)
                        empty_slot = i;
        }
 
@@ -2149,66 +2168,64 @@ static unsigned vi_get_context_dcc_stats_index(struct r600_common_context *rctx,
                int oldest_slot = 0;
 
                /* Find the oldest slot. */
-               for (i = 1; i < ARRAY_SIZE(rctx->dcc_stats); i++)
-                       if (rctx->dcc_stats[oldest_slot].last_use_timestamp >
-                           rctx->dcc_stats[i].last_use_timestamp)
+               for (i = 1; i < ARRAY_SIZE(sctx->b.dcc_stats); i++)
+                       if (sctx->b.dcc_stats[oldest_slot].last_use_timestamp >
+                           sctx->b.dcc_stats[i].last_use_timestamp)
                                oldest_slot = i;
 
                /* Clean up the oldest slot. */
-               vi_dcc_clean_up_context_slot(rctx, oldest_slot);
+               vi_dcc_clean_up_context_slot(sctx, oldest_slot);
                empty_slot = oldest_slot;
        }
 
        /* Add the texture to the new slot. */
-       r600_texture_reference(&rctx->dcc_stats[empty_slot].tex, tex);
-       rctx->dcc_stats[empty_slot].last_use_timestamp = os_time_get();
+       r600_texture_reference(&sctx->b.dcc_stats[empty_slot].tex, tex);
+       sctx->b.dcc_stats[empty_slot].last_use_timestamp = os_time_get();
        return empty_slot;
 }
 
 static struct pipe_query *
-vi_create_resuming_pipestats_query(struct pipe_context *ctx)
+vi_create_resuming_pipestats_query(struct si_context *sctx)
 {
-       struct r600_query_hw *query = (struct r600_query_hw*)
-               ctx->create_query(ctx, PIPE_QUERY_PIPELINE_STATISTICS, 0);
+       struct si_query_hw *query = (struct si_query_hw*)
+               sctx->b.b.create_query(&sctx->b.b, PIPE_QUERY_PIPELINE_STATISTICS, 0);
 
-       query->flags |= R600_QUERY_HW_FLAG_BEGIN_RESUMES;
+       query->flags |= SI_QUERY_HW_FLAG_BEGIN_RESUMES;
        return (struct pipe_query*)query;
 }
 
 /**
  * Called when binding a color buffer.
  */
-void vi_separate_dcc_start_query(struct pipe_context *ctx,
+void vi_separate_dcc_start_query(struct si_context *sctx,
                                 struct r600_texture *tex)
 {
-       struct r600_common_context *rctx = (struct r600_common_context*)ctx;
-       unsigned i = vi_get_context_dcc_stats_index(rctx, tex);
+       unsigned i = vi_get_context_dcc_stats_index(sctx, tex);
 
-       assert(!rctx->dcc_stats[i].query_active);
+       assert(!sctx->b.dcc_stats[i].query_active);
 
-       if (!rctx->dcc_stats[i].ps_stats[0])
-               rctx->dcc_stats[i].ps_stats[0] = vi_create_resuming_pipestats_query(ctx);
+       if (!sctx->b.dcc_stats[i].ps_stats[0])
+               sctx->b.dcc_stats[i].ps_stats[0] = vi_create_resuming_pipestats_query(sctx);
 
        /* begin or resume the query */
-       ctx->begin_query(ctx, rctx->dcc_stats[i].ps_stats[0]);
-       rctx->dcc_stats[i].query_active = true;
+       sctx->b.b.begin_query(&sctx->b.b, sctx->b.dcc_stats[i].ps_stats[0]);
+       sctx->b.dcc_stats[i].query_active = true;
 }
 
 /**
  * Called when unbinding a color buffer.
  */
-void vi_separate_dcc_stop_query(struct pipe_context *ctx,
+void vi_separate_dcc_stop_query(struct si_context *sctx,
                                struct r600_texture *tex)
 {
-       struct r600_common_context *rctx = (struct r600_common_context*)ctx;
-       unsigned i = vi_get_context_dcc_stats_index(rctx, tex);
+       unsigned i = vi_get_context_dcc_stats_index(sctx, tex);
 
-       assert(rctx->dcc_stats[i].query_active);
-       assert(rctx->dcc_stats[i].ps_stats[0]);
+       assert(sctx->b.dcc_stats[i].query_active);
+       assert(sctx->b.dcc_stats[i].ps_stats[0]);
 
        /* pause or end the query */
-       ctx->end_query(ctx, rctx->dcc_stats[i].ps_stats[0]);
-       rctx->dcc_stats[i].query_active = false;
+       sctx->b.b.end_query(&sctx->b.b, sctx->b.dcc_stats[i].ps_stats[0]);
+       sctx->b.dcc_stats[i].query_active = false;
 }
 
 static bool vi_should_enable_separate_dcc(struct r600_texture *tex)
@@ -2219,7 +2236,7 @@ static bool vi_should_enable_separate_dcc(struct r600_texture *tex)
 }
 
 /* Called by fast clear. */
-void vi_separate_dcc_try_enable(struct r600_common_context *rctx,
+void vi_separate_dcc_try_enable(struct si_context *sctx,
                                struct r600_texture *tex)
 {
        /* The intent is to use this with shared displayable back buffers,
@@ -2238,7 +2255,7 @@ void vi_separate_dcc_try_enable(struct r600_common_context *rctx,
        /* Enable the DCC stat gathering. */
        if (!tex->dcc_gather_statistics) {
                tex->dcc_gather_statistics = true;
-               vi_separate_dcc_start_query(&rctx->b, tex);
+               vi_separate_dcc_start_query(sctx, tex);
        }
 
        if (!vi_should_enable_separate_dcc(tex))
@@ -2247,7 +2264,7 @@ void vi_separate_dcc_try_enable(struct r600_common_context *rctx,
        assert(tex->surface.num_dcc_levels);
        assert(!tex->dcc_separate_buffer);
 
-       r600_texture_discard_cmask(rctx->screen, tex);
+       si_texture_discard_cmask(sctx->screen, tex);
 
        /* Get a DCC buffer. */
        if (tex->last_dcc_separate_buffer) {
@@ -2257,8 +2274,8 @@ void vi_separate_dcc_try_enable(struct r600_common_context *rctx,
                tex->last_dcc_separate_buffer = NULL;
        } else {
                tex->dcc_separate_buffer = (struct r600_resource*)
-                       si_aligned_buffer_create(rctx->b.screen,
-                                                  R600_RESOURCE_FLAG_UNMAPPABLE,
+                       si_aligned_buffer_create(sctx->b.b.screen,
+                                                  SI_RESOURCE_FLAG_UNMAPPABLE,
                                                   PIPE_USAGE_DEFAULT,
                                                   tex->surface.dcc_size,
                                                   tex->surface.dcc_alignment);
@@ -2281,27 +2298,27 @@ void vi_separate_dcc_try_enable(struct r600_common_context *rctx,
 void vi_separate_dcc_process_and_reset_stats(struct pipe_context *ctx,
                                             struct r600_texture *tex)
 {
-       struct r600_common_context *rctx = (struct r600_common_context*)ctx;
+       struct si_context *sctx = (struct si_context*)ctx;
        struct pipe_query *tmp;
-       unsigned i = vi_get_context_dcc_stats_index(rctx, tex);
-       bool query_active = rctx->dcc_stats[i].query_active;
+       unsigned i = vi_get_context_dcc_stats_index(sctx, tex);
+       bool query_active = sctx->b.dcc_stats[i].query_active;
        bool disable = false;
 
-       if (rctx->dcc_stats[i].ps_stats[2]) {
+       if (sctx->b.dcc_stats[i].ps_stats[2]) {
                union pipe_query_result result;
 
                /* Read the results. */
-               ctx->get_query_result(ctx, rctx->dcc_stats[i].ps_stats[2],
+               ctx->get_query_result(ctx, sctx->b.dcc_stats[i].ps_stats[2],
                                      true, &result);
-               si_query_hw_reset_buffers(rctx,
-                                           (struct r600_query_hw*)
-                                           rctx->dcc_stats[i].ps_stats[2]);
+               si_query_hw_reset_buffers(sctx,
+                                         (struct si_query_hw*)
+                                         sctx->b.dcc_stats[i].ps_stats[2]);
 
                /* Compute the approximate number of fullscreen draws. */
                tex->ps_draw_ratio =
                        result.pipeline_statistics.ps_invocations /
                        (tex->resource.b.b.width0 * tex->resource.b.b.height0);
-               rctx->last_tex_ps_draw_ratio = tex->ps_draw_ratio;
+               sctx->b.last_tex_ps_draw_ratio = tex->ps_draw_ratio;
 
                disable = tex->dcc_separate_buffer &&
                          !vi_should_enable_separate_dcc(tex);
@@ -2311,17 +2328,17 @@ void vi_separate_dcc_process_and_reset_stats(struct pipe_context *ctx,
 
        /* stop the statistics query for ps_stats[0] */
        if (query_active)
-               vi_separate_dcc_stop_query(ctx, tex);
+               vi_separate_dcc_stop_query(sctx, tex);
 
        /* Move the queries in the queue by one. */
-       tmp = rctx->dcc_stats[i].ps_stats[2];
-       rctx->dcc_stats[i].ps_stats[2] = rctx->dcc_stats[i].ps_stats[1];
-       rctx->dcc_stats[i].ps_stats[1] = rctx->dcc_stats[i].ps_stats[0];
-       rctx->dcc_stats[i].ps_stats[0] = tmp;
+       tmp = sctx->b.dcc_stats[i].ps_stats[2];
+       sctx->b.dcc_stats[i].ps_stats[2] = sctx->b.dcc_stats[i].ps_stats[1];
+       sctx->b.dcc_stats[i].ps_stats[1] = sctx->b.dcc_stats[i].ps_stats[0];
+       sctx->b.dcc_stats[i].ps_stats[0] = tmp;
 
        /* create and start a new query as ps_stats[0] */
        if (query_active)
-               vi_separate_dcc_start_query(ctx, tex);
+               vi_separate_dcc_start_query(sctx, tex);
 
        if (disable) {
                assert(!tex->last_dcc_separate_buffer);
@@ -2335,9 +2352,9 @@ void vi_separate_dcc_process_and_reset_stats(struct pipe_context *ctx,
 }
 
 static struct pipe_memory_object *
-r600_memobj_from_handle(struct pipe_screen *screen,
-                       struct winsys_handle *whandle,
-                       bool dedicated)
+si_memobj_from_handle(struct pipe_screen *screen,
+                     struct winsys_handle *whandle,
+                     bool dedicated)
 {
        struct si_screen *sscreen = (struct si_screen*)screen;
        struct r600_memory_object *memobj = CALLOC_STRUCT(r600_memory_object);
@@ -2364,8 +2381,8 @@ r600_memobj_from_handle(struct pipe_screen *screen,
 }
 
 static void
-r600_memobj_destroy(struct pipe_screen *screen,
-                   struct pipe_memory_object *_memobj)
+si_memobj_destroy(struct pipe_screen *screen,
+                 struct pipe_memory_object *_memobj)
 {
        struct r600_memory_object *memobj = (struct r600_memory_object *)_memobj;
 
@@ -2374,10 +2391,10 @@ r600_memobj_destroy(struct pipe_screen *screen,
 }
 
 static struct pipe_resource *
-r600_texture_from_memobj(struct pipe_screen *screen,
-                        const struct pipe_resource *templ,
-                        struct pipe_memory_object *_memobj,
-                        uint64_t offset)
+si_texture_from_memobj(struct pipe_screen *screen,
+                      const struct pipe_resource *templ,
+                      struct pipe_memory_object *_memobj,
+                      uint64_t offset)
 {
        int r;
        struct si_screen *sscreen = (struct si_screen*)screen;
@@ -2391,7 +2408,7 @@ r600_texture_from_memobj(struct pipe_screen *screen,
 
        if (memobj->b.dedicated) {
                sscreen->ws->buffer_get_metadata(memobj->buf, &metadata);
-               r600_surface_import_metadata(sscreen, &surface, &metadata,
+               si_surface_import_metadata(sscreen, &surface, &metadata,
                                     &array_mode, &is_scanout);
        } else {
                /**
@@ -2421,14 +2438,14 @@ r600_texture_from_memobj(struct pipe_screen *screen,
 
        }
 
-       r = r600_init_surface(sscreen, &surface, templ,
+       r = si_init_surface(sscreen, &surface, templ,
                              array_mode, memobj->stride,
                              offset, true, is_scanout,
                              false, false);
        if (r)
                return NULL;
 
-       rtex = r600_texture_create_object(screen, templ, memobj->buf, &surface);
+       rtex = si_texture_create_object(screen, templ, memobj->buf, &surface);
        if (!rtex)
                return NULL;
 
@@ -2467,16 +2484,16 @@ static bool si_check_resource_capability(struct pipe_screen *screen,
 
 void si_init_screen_texture_functions(struct si_screen *sscreen)
 {
-       sscreen->b.resource_from_handle = r600_texture_from_handle;
-       sscreen->b.resource_get_handle = r600_texture_get_handle;
-       sscreen->b.resource_from_memobj = r600_texture_from_memobj;
-       sscreen->b.memobj_create_from_handle = r600_memobj_from_handle;
-       sscreen->b.memobj_destroy = r600_memobj_destroy;
+       sscreen->b.resource_from_handle = si_texture_from_handle;
+       sscreen->b.resource_get_handle = si_texture_get_handle;
+       sscreen->b.resource_from_memobj = si_texture_from_memobj;
+       sscreen->b.memobj_create_from_handle = si_memobj_from_handle;
+       sscreen->b.memobj_destroy = si_memobj_destroy;
        sscreen->b.check_resource_capability = si_check_resource_capability;
 }
 
-void si_init_context_texture_functions(struct r600_common_context *rctx)
+void si_init_context_texture_functions(struct si_context *sctx)
 {
-       rctx->b.create_surface = r600_create_surface;
-       rctx->b.surface_destroy = r600_surface_destroy;
+       sctx->b.b.create_surface = si_create_surface;
+       sctx->b.b.surface_destroy = si_surface_destroy;
 }