if (!rctx->dma.cs)
return false;
- if (util_format_get_blocksizebits(rdst->resource.b.b.format) !=
- util_format_get_blocksizebits(rsrc->resource.b.b.format))
+ if (rdst->surface.bpe != rsrc->surface.bpe)
return false;
/* MSAA: Blits don't exist in the real world. */
* src: Use the 3D path. DCC decompression is expensive.
* dst: Use the 3D path to compress the pixels with DCC.
*/
- if ((rsrc->dcc_offset && rsrc->surface.level[src_level].dcc_enabled) ||
- (rdst->dcc_offset && rdst->surface.level[dst_level].dcc_enabled))
+ if (vi_dcc_enabled(rsrc, src_level) ||
+ vi_dcc_enabled(rdst, dst_level))
return false;
/* CMASK as:
src, 0, &sbox);
}
-static unsigned r600_texture_get_offset(struct r600_texture *rtex, unsigned level,
- const struct pipe_box *box)
+static unsigned r600_texture_get_offset(struct r600_common_screen *rscreen,
+ struct r600_texture *rtex, unsigned level,
+ const struct pipe_box *box,
+ unsigned *stride,
+ unsigned *layer_stride)
{
- enum pipe_format format = rtex->resource.b.b.format;
-
- return rtex->surface.level[level].offset +
- box->z * rtex->surface.level[level].slice_size +
- box->y / util_format_get_blockheight(format) * rtex->surface.level[level].pitch_bytes +
- box->x / util_format_get_blockwidth(format) * util_format_get_blocksize(format);
+ if (rscreen->chip_class >= GFX9) {
+ *stride = rtex->surface.u.gfx9.surf_pitch * rtex->surface.bpe;
+ *layer_stride = rtex->surface.u.gfx9.surf_slice_size;
+
+ if (!box)
+ return 0;
+
+ /* Each texture is an array of slices. Each slice is an array
+ * of mipmap levels. */
+ return box->z * rtex->surface.u.gfx9.surf_slice_size +
+ rtex->surface.u.gfx9.offset[level] +
+ (box->y / rtex->surface.blk_h *
+ rtex->surface.u.gfx9.surf_pitch +
+ box->x / rtex->surface.blk_w) * rtex->surface.bpe;
+ } else {
+ *stride = rtex->surface.u.legacy.level[level].nblk_x *
+ rtex->surface.bpe;
+ *layer_stride = rtex->surface.u.legacy.level[level].slice_size;
+
+ if (!box)
+ return rtex->surface.u.legacy.level[level].offset;
+
+ /* Each texture is an array of mipmap levels. Each level is
+ * an array of slices. */
+ return rtex->surface.u.legacy.level[level].offset +
+ box->z * rtex->surface.u.legacy.level[level].slice_size +
+ (box->y / rtex->surface.blk_h *
+ rtex->surface.u.legacy.level[level].nblk_x +
+ box->x / rtex->surface.blk_w) * rtex->surface.bpe;
+ }
}
static int r600_init_surface(struct r600_common_screen *rscreen,
bpe = 4; /* stencil is allocated separately on evergreen */
} else {
bpe = util_format_get_blocksize(ptex->format);
- /* align byte per element on dword */
- if (bpe == 3) {
- bpe = 4;
- }
+ assert(util_is_power_of_two(bpe));
}
if (!is_flushed_depth && is_depth) {
flags |= RADEON_SURF_ZBUFFER;
if (tc_compatible_htile &&
- array_mode == RADEON_SURF_MODE_2D) {
+ (rscreen->chip_class >= GFX9 ||
+ array_mode == RADEON_SURF_MODE_2D)) {
/* TC-compatible HTILE only supports Z32_FLOAT.
- * Promote Z16 to Z32. DB->CB copies will convert
+ * GFX9 also supports Z16_UNORM.
+ * On VI, promote Z16 to Z32. DB->CB copies will convert
* the format for transfers.
*/
- bpe = 4;
+ if (rscreen->chip_class == VI)
+ bpe = 4;
+
flags |= RADEON_SURF_TC_COMPATIBLE_HTILE;
}
if (is_imported)
flags |= RADEON_SURF_IMPORTED;
+ if (!(ptex->flags & R600_RESOURCE_FLAG_FORCE_TILING))
+ flags |= RADEON_SURF_OPTIMIZE_FOR_SPACE;
r = rscreen->ws->surface_init(rscreen->ws, ptex, flags, bpe,
array_mode, surface);
return r;
}
- if (pitch_in_bytes_override && pitch_in_bytes_override != surface->level[0].pitch_bytes) {
- /* old ddx on evergreen over estimate alignment for 1d, only 1 level
- * for those
- */
- surface->level[0].nblk_x = pitch_in_bytes_override / bpe;
- surface->level[0].pitch_bytes = pitch_in_bytes_override;
- surface->level[0].slice_size = pitch_in_bytes_override * surface->level[0].nblk_y;
- }
+ if (rscreen->chip_class >= GFX9) {
+ assert(!pitch_in_bytes_override ||
+ pitch_in_bytes_override == surface->u.gfx9.surf_pitch * bpe);
+ surface->u.gfx9.surf_offset = offset;
+ } else {
+ if (pitch_in_bytes_override &&
+ pitch_in_bytes_override != surface->u.legacy.level[0].nblk_x * bpe) {
+ /* old ddx on evergreen over estimate alignment for 1d, only 1 level
+ * for those
+ */
+ surface->u.legacy.level[0].nblk_x = pitch_in_bytes_override / bpe;
+ surface->u.legacy.level[0].slice_size = pitch_in_bytes_override *
+ surface->u.legacy.level[0].nblk_y;
+ }
- if (offset) {
- for (i = 0; i < ARRAY_SIZE(surface->level); ++i)
- surface->level[i].offset += offset;
+ if (offset) {
+ for (i = 0; i < ARRAY_SIZE(surface->u.legacy.level); ++i)
+ surface->u.legacy.level[i].offset += offset;
+ }
}
return 0;
}
-static void r600_texture_init_metadata(struct r600_texture *rtex,
+static void r600_texture_init_metadata(struct r600_common_screen *rscreen,
+ struct r600_texture *rtex,
struct radeon_bo_metadata *metadata)
{
struct radeon_surf *surface = &rtex->surface;
memset(metadata, 0, sizeof(*metadata));
- metadata->microtile = surface->level[0].mode >= RADEON_SURF_MODE_1D ?
- RADEON_LAYOUT_TILED : RADEON_LAYOUT_LINEAR;
- metadata->macrotile = surface->level[0].mode >= RADEON_SURF_MODE_2D ?
- RADEON_LAYOUT_TILED : RADEON_LAYOUT_LINEAR;
- metadata->pipe_config = surface->pipe_config;
- metadata->bankw = surface->bankw;
- metadata->bankh = surface->bankh;
- metadata->tile_split = surface->tile_split;
- metadata->mtilea = surface->mtilea;
- metadata->num_banks = surface->num_banks;
- metadata->stride = surface->level[0].pitch_bytes;
- metadata->scanout = (surface->flags & RADEON_SURF_SCANOUT) != 0;
-}
-static void r600_dirty_all_framebuffer_states(struct r600_common_screen *rscreen)
-{
- p_atomic_inc(&rscreen->dirty_fb_counter);
+ if (rscreen->chip_class >= GFX9) {
+ metadata->u.gfx9.swizzle_mode = surface->u.gfx9.surf.swizzle_mode;
+ } else {
+ metadata->u.legacy.microtile = surface->u.legacy.level[0].mode >= RADEON_SURF_MODE_1D ?
+ RADEON_LAYOUT_TILED : RADEON_LAYOUT_LINEAR;
+ metadata->u.legacy.macrotile = surface->u.legacy.level[0].mode >= RADEON_SURF_MODE_2D ?
+ RADEON_LAYOUT_TILED : RADEON_LAYOUT_LINEAR;
+ metadata->u.legacy.pipe_config = surface->u.legacy.pipe_config;
+ metadata->u.legacy.bankw = surface->u.legacy.bankw;
+ metadata->u.legacy.bankh = surface->u.legacy.bankh;
+ metadata->u.legacy.tile_split = surface->u.legacy.tile_split;
+ metadata->u.legacy.mtilea = surface->u.legacy.mtilea;
+ metadata->u.legacy.num_banks = surface->u.legacy.num_banks;
+ metadata->u.legacy.stride = surface->u.legacy.level[0].nblk_x * surface->bpe;
+ metadata->u.legacy.scanout = (surface->flags & RADEON_SURF_SCANOUT) != 0;
+ }
}
static void r600_eliminate_fast_color_clear(struct r600_common_context *rctx,
struct pipe_context *ctx = &rctx->b;
if (ctx == rscreen->aux_context)
- pipe_mutex_lock(rscreen->aux_context_lock);
+ mtx_lock(&rscreen->aux_context_lock);
ctx->flush_resource(ctx, &rtex->resource.b.b);
ctx->flush(ctx, NULL, 0);
if (ctx == rscreen->aux_context)
- pipe_mutex_unlock(rscreen->aux_context_lock);
+ mtx_unlock(&rscreen->aux_context_lock);
}
static void r600_texture_discard_cmask(struct r600_common_screen *rscreen,
r600_resource_reference(&rtex->cmask_buffer, NULL);
/* Notify all contexts about the change. */
- r600_dirty_all_framebuffer_states(rscreen);
+ p_atomic_inc(&rscreen->dirty_tex_counter);
p_atomic_inc(&rscreen->compressed_colortex_counter);
}
{
/* We can't disable DCC if it can be written by another process. */
return rtex->dcc_offset &&
- (!rtex->resource.is_shared ||
+ (!rtex->resource.b.is_shared ||
!(rtex->resource.external_usage & PIPE_HANDLE_USAGE_WRITE));
}
rtex->dcc_offset = 0;
/* Notify all contexts about the change. */
- r600_dirty_all_framebuffer_states(rscreen);
+ p_atomic_inc(&rscreen->dirty_tex_counter);
return true;
}
return false;
if (&rctx->b == rscreen->aux_context)
- pipe_mutex_lock(rscreen->aux_context_lock);
+ mtx_lock(&rscreen->aux_context_lock);
/* Decompress DCC. */
rctx->decompress_dcc(&rctx->b, rtex);
rctx->b.flush(&rctx->b, NULL, 0);
if (&rctx->b == rscreen->aux_context)
- pipe_mutex_unlock(rscreen->aux_context_lock);
+ mtx_unlock(&rscreen->aux_context_lock);
return r600_texture_discard_dcc(rscreen, rtex);
}
if (rctx->chip_class < SI)
return;
- if (rtex->resource.is_shared ||
- rtex->surface.level[0].mode == RADEON_SURF_MODE_LINEAR_ALIGNED)
+ if (rtex->resource.b.is_shared ||
+ rtex->surface.is_linear)
return;
/* This fails with MSAA, depth, and compressed textures. */
rtex->cb_color_info = new_tex->cb_color_info;
rtex->cmask = new_tex->cmask; /* needed even without CMASK */
- assert(!rtex->htile_buffer);
+ assert(!rtex->htile_offset);
assert(!rtex->cmask.size);
assert(!rtex->fmask.size);
assert(!rtex->dcc_offset);
r600_texture_reference(&new_tex, NULL);
- r600_dirty_all_framebuffer_states(rctx->screen);
- p_atomic_inc(&rctx->screen->dirty_tex_descriptor_counter);
+ p_atomic_inc(&rctx->screen->dirty_tex_counter);
}
static boolean r600_texture_get_handle(struct pipe_screen* screen,
unsigned usage)
{
struct r600_common_screen *rscreen = (struct r600_common_screen*)screen;
- struct r600_common_context *rctx = (struct r600_common_context*)
- (ctx ? ctx : rscreen->aux_context);
+ struct r600_common_context *rctx;
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;
+
+ ctx = threaded_context_unwrap_sync(ctx);
+ rctx = (struct r600_common_context*)(ctx ? ctx : rscreen->aux_context);
/* This is not supported now, but it might be required for OpenCL
* interop in the future.
}
/* Set metadata. */
- if (!res->is_shared || update_metadata) {
- r600_texture_init_metadata(rtex, &metadata);
+ if (!res->b.is_shared || update_metadata) {
+ r600_texture_init_metadata(rscreen, rtex, &metadata);
if (rscreen->query_opaque_metadata)
rscreen->query_opaque_metadata(rscreen, rtex,
&metadata);
}
}
- if (res->is_shared) {
+ if (res->b.is_shared) {
/* USAGE_EXPLICIT_FLUSH must be cleared if at least one user
* doesn't set it.
*/
if (!(usage & PIPE_HANDLE_USAGE_EXPLICIT_FLUSH))
res->external_usage &= ~PIPE_HANDLE_USAGE_EXPLICIT_FLUSH;
} else {
- res->is_shared = true;
+ res->b.is_shared = true;
res->external_usage = usage;
}
- return rscreen->ws->buffer_get_handle(res->buf,
- rtex->surface.level[0].pitch_bytes,
- rtex->surface.level[0].offset,
- rtex->surface.level[0].slice_size,
- whandle);
+ if (rscreen->chip_class >= GFX9) {
+ offset = rtex->surface.u.gfx9.surf_offset;
+ stride = rtex->surface.u.gfx9.surf_pitch *
+ rtex->surface.bpe;
+ slice_size = rtex->surface.u.gfx9.surf_slice_size;
+ } else {
+ offset = rtex->surface.u.legacy.level[0].offset;
+ stride = rtex->surface.u.legacy.level[0].nblk_x *
+ rtex->surface.bpe;
+ slice_size = rtex->surface.u.legacy.level[0].slice_size;
+ }
+ return rscreen->ws->buffer_get_handle(res->buf, stride, offset,
+ slice_size, whandle);
}
static void r600_texture_destroy(struct pipe_screen *screen,
r600_texture_reference(&rtex->flushed_depth_texture, NULL);
- r600_resource_reference(&rtex->htile_buffer, NULL);
if (rtex->cmask_buffer != &rtex->resource) {
r600_resource_reference(&rtex->cmask_buffer, NULL);
}
memset(out, 0, sizeof(*out));
+ if (rscreen->chip_class >= GFX9) {
+ out->alignment = rtex->surface.u.gfx9.fmask_alignment;
+ out->size = rtex->surface.u.gfx9.fmask_size;
+ return;
+ }
+
templ.nr_samples = 1;
flags = rtex->surface.flags | RADEON_SURF_FMASK;
if (rscreen->chip_class <= CAYMAN) {
/* Use the same parameters and tile mode. */
- fmask.bankw = rtex->surface.bankw;
- fmask.bankh = rtex->surface.bankh;
- fmask.mtilea = rtex->surface.mtilea;
- fmask.tile_split = rtex->surface.tile_split;
+ fmask.u.legacy.bankw = rtex->surface.u.legacy.bankw;
+ fmask.u.legacy.bankh = rtex->surface.u.legacy.bankh;
+ fmask.u.legacy.mtilea = rtex->surface.u.legacy.mtilea;
+ fmask.u.legacy.tile_split = rtex->surface.u.legacy.tile_split;
if (nr_samples <= 4)
- fmask.bankh = 4;
+ fmask.u.legacy.bankh = 4;
}
switch (nr_samples) {
return;
}
- assert(fmask.level[0].mode == RADEON_SURF_MODE_2D);
+ assert(fmask.u.legacy.level[0].mode == RADEON_SURF_MODE_2D);
- out->slice_tile_max = (fmask.level[0].nblk_x * fmask.level[0].nblk_y) / 64;
+ out->slice_tile_max = (fmask.u.legacy.level[0].nblk_x * fmask.u.legacy.level[0].nblk_y) / 64;
if (out->slice_tile_max)
out->slice_tile_max -= 1;
- out->tile_mode_index = fmask.tiling_index[0];
- out->pitch_in_pixels = fmask.level[0].nblk_x;
- out->bank_height = fmask.bankh;
+ out->tile_mode_index = fmask.u.legacy.tiling_index[0];
+ out->pitch_in_pixels = fmask.u.legacy.level[0].nblk_x;
+ out->bank_height = fmask.u.legacy.bankh;
out->alignment = MAX2(256, fmask.surf_alignment);
out->size = fmask.surf_size;
}
unsigned num_pipes = rscreen->info.num_tile_pipes;
unsigned cl_width, cl_height;
+ if (rscreen->chip_class >= GFX9) {
+ out->alignment = rtex->surface.u.gfx9.cmask_alignment;
+ out->size = rtex->surface.u.gfx9.cmask_size;
+ return;
+ }
+
switch (num_pipes) {
case 2:
cl_width = 32;
}
rtex->cmask_buffer = (struct r600_resource *)
- r600_aligned_buffer_create(&rscreen->b, 0, PIPE_USAGE_DEFAULT,
+ r600_aligned_buffer_create(&rscreen->b,
+ R600_RESOURCE_FLAG_UNMAPPABLE,
+ PIPE_USAGE_DEFAULT,
rtex->cmask.size,
rtex->cmask.alignment);
if (rtex->cmask_buffer == NULL) {
unsigned slice_elements, slice_bytes, pipe_interleave_bytes, base_align;
unsigned num_pipes = rscreen->info.num_tile_pipes;
+ assert(rscreen->chip_class <= VI);
+
rtex->surface.htile_size = 0;
if (rscreen->chip_class <= EVERGREEN &&
/* HTILE is broken with 1D tiling on old kernels and CIK. */
if (rscreen->chip_class >= CIK &&
- rtex->surface.level[0].mode == RADEON_SURF_MODE_1D &&
+ rtex->surface.u.legacy.level[0].mode == RADEON_SURF_MODE_1D &&
rscreen->info.drm_major == 2 && rscreen->info.drm_minor < 38)
return;
static void r600_texture_allocate_htile(struct r600_common_screen *rscreen,
struct r600_texture *rtex)
{
- uint32_t clear_value;
-
- if (rtex->tc_compatible_htile) {
- clear_value = 0x0000030F;
- } else {
+ if (rscreen->chip_class <= VI && !rtex->tc_compatible_htile)
r600_texture_get_htile_size(rscreen, rtex);
- clear_value = 0;
- }
if (!rtex->surface.htile_size)
return;
- rtex->htile_buffer = (struct r600_resource*)
- r600_aligned_buffer_create(&rscreen->b, 0, PIPE_USAGE_DEFAULT,
- rtex->surface.htile_size,
- rtex->surface.htile_alignment);
- if (rtex->htile_buffer == NULL) {
- /* this is not a fatal error as we can still keep rendering
- * without htile buffer */
- R600_ERR("Failed to create buffer object for htile buffer.\n");
- } else {
- r600_screen_clear_buffer(rscreen, &rtex->htile_buffer->b.b,
- 0, rtex->surface.htile_size,
- clear_value, R600_COHERENCY_NONE);
- }
+ rtex->htile_offset = align(rtex->size, rtex->surface.htile_alignment);
+ rtex->size = rtex->htile_offset + rtex->surface.htile_size;
}
-void r600_print_texture_info(struct r600_texture *rtex, FILE *f)
+void r600_print_texture_info(struct r600_common_screen *rscreen,
+ struct r600_texture *rtex, FILE *f)
{
int i;
+ /* Common parameters. */
fprintf(f, " Info: npix_x=%u, npix_y=%u, npix_z=%u, blk_w=%u, "
"blk_h=%u, array_size=%u, last_level=%u, "
"bpe=%u, nsamples=%u, flags=0x%x, %s\n",
rtex->surface.bpe, rtex->resource.b.b.nr_samples,
rtex->surface.flags, util_format_short_name(rtex->resource.b.b.format));
+ if (rscreen->chip_class >= GFX9) {
+ fprintf(f, " Surf: size=%"PRIu64", slice_size=%"PRIu64", "
+ "alignment=%u, swmode=%u, epitch=%u, pitch=%u\n",
+ rtex->surface.surf_size,
+ rtex->surface.u.gfx9.surf_slice_size,
+ rtex->surface.surf_alignment,
+ rtex->surface.u.gfx9.surf.swizzle_mode,
+ rtex->surface.u.gfx9.surf.epitch,
+ rtex->surface.u.gfx9.surf_pitch);
+
+ if (rtex->fmask.size) {
+ fprintf(f, " FMASK: offset=%"PRIu64", size=%"PRIu64", "
+ "alignment=%u, swmode=%u, epitch=%u\n",
+ rtex->fmask.offset,
+ rtex->surface.u.gfx9.fmask_size,
+ rtex->surface.u.gfx9.fmask_alignment,
+ rtex->surface.u.gfx9.fmask.swizzle_mode,
+ rtex->surface.u.gfx9.fmask.epitch);
+ }
+
+ if (rtex->cmask.size) {
+ fprintf(f, " CMask: offset=%"PRIu64", size=%"PRIu64", "
+ "alignment=%u, rb_aligned=%u, pipe_aligned=%u\n",
+ rtex->cmask.offset,
+ rtex->surface.u.gfx9.cmask_size,
+ rtex->surface.u.gfx9.cmask_alignment,
+ rtex->surface.u.gfx9.cmask.rb_aligned,
+ rtex->surface.u.gfx9.cmask.pipe_aligned);
+ }
+
+ if (rtex->htile_offset) {
+ fprintf(f, " HTile: offset=%"PRIu64", size=%"PRIu64", alignment=%u, "
+ "rb_aligned=%u, pipe_aligned=%u\n",
+ rtex->htile_offset,
+ rtex->surface.htile_size,
+ rtex->surface.htile_alignment,
+ rtex->surface.u.gfx9.htile.rb_aligned,
+ rtex->surface.u.gfx9.htile.pipe_aligned);
+ }
+
+ if (rtex->dcc_offset) {
+ fprintf(f, " DCC: offset=%"PRIu64", size=%"PRIu64", "
+ "alignment=%u, pitch_max=%u, num_dcc_levels=%u\n",
+ rtex->dcc_offset, rtex->surface.dcc_size,
+ rtex->surface.dcc_alignment,
+ rtex->surface.u.gfx9.dcc_pitch_max,
+ rtex->surface.num_dcc_levels);
+ }
+
+ if (rtex->surface.u.gfx9.stencil_offset) {
+ fprintf(f, " Stencil: offset=%"PRIu64", swmode=%u, epitch=%u\n",
+ rtex->surface.u.gfx9.stencil_offset,
+ rtex->surface.u.gfx9.stencil.swizzle_mode,
+ rtex->surface.u.gfx9.stencil.epitch);
+ }
+ return;
+ }
+
fprintf(f, " Layout: size=%"PRIu64", alignment=%u, bankw=%u, "
"bankh=%u, nbanks=%u, mtilea=%u, tilesplit=%u, pipeconfig=%u, scanout=%u\n",
- rtex->surface.surf_size, rtex->surface.surf_alignment, rtex->surface.bankw,
- rtex->surface.bankh, rtex->surface.num_banks, rtex->surface.mtilea,
- rtex->surface.tile_split, rtex->surface.pipe_config,
+ rtex->surface.surf_size, rtex->surface.surf_alignment, rtex->surface.u.legacy.bankw,
+ rtex->surface.u.legacy.bankh, rtex->surface.u.legacy.num_banks, rtex->surface.u.legacy.mtilea,
+ rtex->surface.u.legacy.tile_split, rtex->surface.u.legacy.pipe_config,
(rtex->surface.flags & RADEON_SURF_SCANOUT) != 0);
if (rtex->fmask.size)
rtex->cmask.offset, rtex->cmask.size, rtex->cmask.alignment,
rtex->cmask.slice_tile_max);
- if (rtex->htile_buffer)
- fprintf(f, " HTile: size=%u, alignment=%u, TC_compatible = %u\n",
- rtex->htile_buffer->b.b.width0,
- rtex->htile_buffer->buf->alignment,
+ if (rtex->htile_offset)
+ fprintf(f, " HTile: offset=%"PRIu64", size=%"PRIu64", "
+ "alignment=%u, TC_compatible = %u\n",
+ rtex->htile_offset, rtex->surface.htile_size,
+ rtex->surface.htile_alignment,
rtex->tc_compatible_htile);
if (rtex->dcc_offset) {
for (i = 0; i <= rtex->resource.b.b.last_level; i++)
fprintf(f, " DCCLevel[%i]: enabled=%u, offset=%"PRIu64", "
"fast_clear_size=%"PRIu64"\n",
- i, rtex->surface.level[i].dcc_enabled,
- rtex->surface.level[i].dcc_offset,
- rtex->surface.level[i].dcc_fast_clear_size);
+ i, i < rtex->surface.num_dcc_levels,
+ rtex->surface.u.legacy.level[i].dcc_offset,
+ rtex->surface.u.legacy.level[i].dcc_fast_clear_size);
}
for (i = 0; i <= rtex->resource.b.b.last_level; i++)
fprintf(f, " Level[%i]: offset=%"PRIu64", slice_size=%"PRIu64", "
"npix_x=%u, npix_y=%u, npix_z=%u, nblk_x=%u, nblk_y=%u, "
- "pitch_bytes=%u, mode=%u, tiling_index = %u\n",
- i, rtex->surface.level[i].offset,
- rtex->surface.level[i].slice_size,
+ "mode=%u, tiling_index = %u\n",
+ i, rtex->surface.u.legacy.level[i].offset,
+ rtex->surface.u.legacy.level[i].slice_size,
u_minify(rtex->resource.b.b.width0, i),
u_minify(rtex->resource.b.b.height0, i),
u_minify(rtex->resource.b.b.depth0, i),
- rtex->surface.level[i].nblk_x,
- rtex->surface.level[i].nblk_y,
- rtex->surface.level[i].pitch_bytes,
- rtex->surface.level[i].mode,
- rtex->surface.tiling_index[i]);
+ rtex->surface.u.legacy.level[i].nblk_x,
+ rtex->surface.u.legacy.level[i].nblk_y,
+ rtex->surface.u.legacy.level[i].mode,
+ rtex->surface.u.legacy.tiling_index[i]);
if (rtex->surface.flags & RADEON_SURF_SBUFFER) {
fprintf(f, " StencilLayout: tilesplit=%u\n",
- rtex->surface.stencil_tile_split);
+ rtex->surface.u.legacy.stencil_tile_split);
for (i = 0; i <= rtex->resource.b.b.last_level; i++) {
fprintf(f, " StencilLevel[%i]: offset=%"PRIu64", "
"slice_size=%"PRIu64", npix_x=%u, "
"npix_y=%u, npix_z=%u, nblk_x=%u, nblk_y=%u, "
- "pitch_bytes=%u, mode=%u, tiling_index = %u\n",
- i, rtex->surface.stencil_level[i].offset,
- rtex->surface.stencil_level[i].slice_size,
+ "mode=%u, tiling_index = %u\n",
+ i, rtex->surface.u.legacy.stencil_level[i].offset,
+ rtex->surface.u.legacy.stencil_level[i].slice_size,
u_minify(rtex->resource.b.b.width0, i),
u_minify(rtex->resource.b.b.height0, i),
u_minify(rtex->resource.b.b.depth0, i),
- rtex->surface.stencil_level[i].nblk_x,
- rtex->surface.stencil_level[i].nblk_y,
- rtex->surface.stencil_level[i].pitch_bytes,
- rtex->surface.stencil_level[i].mode,
- rtex->surface.stencil_tiling_index[i]);
+ rtex->surface.u.legacy.stencil_level[i].nblk_x,
+ rtex->surface.u.legacy.stencil_level[i].nblk_y,
+ rtex->surface.u.legacy.stencil_level[i].mode,
+ rtex->surface.u.legacy.stencil_tiling_index[i]);
}
}
}
rtex->surface = *surface;
rtex->size = rtex->surface.surf_size;
- rtex->tc_compatible_htile = rtex->surface.htile_size != 0;
- assert(!!(rtex->surface.flags & RADEON_SURF_TC_COMPATIBLE_HTILE) ==
- rtex->tc_compatible_htile);
+ rtex->tc_compatible_htile = rtex->surface.htile_size != 0 &&
+ (rtex->surface.flags &
+ RADEON_SURF_TC_COMPATIBLE_HTILE);
- /* TC-compatible HTILE only supports Z32_FLOAT. */
- if (rtex->tc_compatible_htile)
- rtex->db_render_format = PIPE_FORMAT_Z32_FLOAT;
- else
+ /* TC-compatible HTILE:
+ * - VI only supports Z32_FLOAT.
+ * - GFX9 only supports Z32_FLOAT and Z16_UNORM. */
+ if (rtex->tc_compatible_htile) {
+ if (rscreen->chip_class >= GFX9 &&
+ base->format == PIPE_FORMAT_Z16_UNORM)
+ rtex->db_render_format = base->format;
+ else
+ rtex->db_render_format = PIPE_FORMAT_Z32_FLOAT;
+ } else {
rtex->db_render_format = base->format;
+ }
/* Tiled depth textures utilize the non-displayable tile order.
* This must be done after r600_setup_surface.
* Applies to R600-Cayman. */
- rtex->non_disp_tiling = rtex->is_depth && rtex->surface.level[0].mode >= RADEON_SURF_MODE_1D;
+ rtex->non_disp_tiling = rtex->is_depth && rtex->surface.u.legacy.level[0].mode >= RADEON_SURF_MODE_1D;
/* Applies to GCN. */
rtex->last_msaa_resolve_target_micro_mode = rtex->surface.micro_tile_mode;
if (base->flags & (R600_RESOURCE_FLAG_TRANSFER |
R600_RESOURCE_FLAG_FLUSHED_DEPTH) ||
rscreen->chip_class >= EVERGREEN) {
- rtex->can_sample_z = !rtex->surface.depth_adjusted;
- rtex->can_sample_s = !rtex->surface.stencil_adjusted;
+ if (rscreen->chip_class >= GFX9) {
+ rtex->can_sample_z = true;
+ rtex->can_sample_s = true;
+ } else {
+ rtex->can_sample_z = !rtex->surface.u.legacy.depth_adjusted;
+ rtex->can_sample_s = !rtex->surface.u.legacy.stencil_adjusted;
+ }
} else {
if (rtex->resource.b.b.nr_samples <= 1 &&
(rtex->resource.b.b.format == PIPE_FORMAT_Z16_UNORM ||
/* Initialize the cmask to 0xCC (= compressed state). */
r600_screen_clear_buffer(rscreen, &rtex->cmask_buffer->b.b,
rtex->cmask.offset, rtex->cmask.size,
- 0xCCCCCCCC, R600_COHERENCY_NONE);
+ 0xCCCCCCCC);
+ }
+ if (rtex->htile_offset) {
+ uint32_t clear_value = 0;
+
+ if (rscreen->chip_class >= GFX9 || rtex->tc_compatible_htile)
+ clear_value = 0x0000030F;
+
+ r600_screen_clear_buffer(rscreen, &rtex->resource.b.b,
+ rtex->htile_offset,
+ rtex->surface.htile_size,
+ clear_value);
}
/* Initialize DCC only if the texture is not being imported. */
r600_screen_clear_buffer(rscreen, &rtex->resource.b.b,
rtex->dcc_offset,
rtex->surface.dcc_size,
- 0xFFFFFFFF, R600_COHERENCY_NONE);
+ 0xFFFFFFFF);
}
/* Initialize the CMASK base register value. */
if (rscreen->debug_flags & DBG_TEX) {
puts("Texture:");
- r600_print_texture_info(rtex, stdout);
+ r600_print_texture_info(rscreen, rtex, stdout);
fflush(stdout);
}
{
const struct util_format_description *desc = util_format_description(templ->format);
bool force_tiling = templ->flags & R600_RESOURCE_FLAG_FORCE_TILING;
+ bool is_depth_stencil = util_format_is_depth_or_stencil(templ->format) &&
+ !(templ->flags & R600_RESOURCE_FLAG_FLUSHED_DEPTH);
/* MSAA resources must be 2D tiled. */
if (templ->nr_samples > 1)
if (templ->flags & R600_RESOURCE_FLAG_TRANSFER)
return RADEON_SURF_MODE_LINEAR_ALIGNED;
+ /* Avoid Z/S decompress blits by forcing TC-compatible HTILE on VI,
+ * which requires 2D tiling.
+ */
+ if (rscreen->chip_class == VI &&
+ is_depth_stencil &&
+ (templ->flags & PIPE_RESOURCE_FLAG_TEXTURING_MORE_LIKELY))
+ return RADEON_SURF_MODE_2D;
+
/* r600g: force tiling on TEXTURE_2D and TEXTURE_3D compute resources. */
if (rscreen->chip_class >= R600 && rscreen->chip_class <= CAYMAN &&
(templ->bind & PIPE_BIND_COMPUTE_RESOURCE) &&
/* Handle common candidates for the linear mode.
* Compressed textures and DB surfaces must always be tiled.
*/
- if (!force_tiling && !util_format_is_compressed(templ->format) &&
- (!util_format_is_depth_or_stencil(templ->format) ||
- templ->flags & R600_RESOURCE_FLAG_FLUSHED_DEPTH)) {
+ if (!force_tiling &&
+ !is_depth_stencil &&
+ !util_format_is_compressed(templ->format)) {
if (rscreen->debug_flags & DBG_NO_TILING)
return RADEON_SURF_MODE_LINEAR_ALIGNED;
/* Textures with a very small height are recommended to be linear. */
if (templ->target == PIPE_TEXTURE_1D ||
templ->target == PIPE_TEXTURE_1D_ARRAY ||
- templ->height0 <= 4)
+ /* Only very thin and long 2D textures should benefit from
+ * linear_aligned. */
+ (templ->width0 > 8 && templ->height0 <= 2))
return RADEON_SURF_MODE_LINEAR_ALIGNED;
/* Textures likely to be mapped often. */
int r;
struct radeon_bo_metadata metadata = {};
struct r600_texture *rtex;
+ bool is_scanout;
/* Support only 2D textures without mipmaps */
if ((templ->target != PIPE_TEXTURE_2D && templ->target != PIPE_TEXTURE_RECT) ||
rscreen->ws->buffer_get_metadata(buf, &metadata);
- surface.pipe_config = metadata.pipe_config;
- surface.bankw = metadata.bankw;
- surface.bankh = metadata.bankh;
- surface.tile_split = metadata.tile_split;
- surface.mtilea = metadata.mtilea;
- surface.num_banks = metadata.num_banks;
-
- if (metadata.macrotile == RADEON_LAYOUT_TILED)
- array_mode = RADEON_SURF_MODE_2D;
- else if (metadata.microtile == RADEON_LAYOUT_TILED)
- array_mode = RADEON_SURF_MODE_1D;
- else
- array_mode = RADEON_SURF_MODE_LINEAR_ALIGNED;
+ if (rscreen->chip_class >= GFX9) {
+ if (metadata.u.gfx9.swizzle_mode > 0)
+ array_mode = RADEON_SURF_MODE_2D;
+ else
+ array_mode = RADEON_SURF_MODE_LINEAR_ALIGNED;
+
+ is_scanout = metadata.u.gfx9.swizzle_mode == 0 ||
+ metadata.u.gfx9.swizzle_mode % 4 == 2;
+ } else {
+ surface.u.legacy.pipe_config = metadata.u.legacy.pipe_config;
+ surface.u.legacy.bankw = metadata.u.legacy.bankw;
+ surface.u.legacy.bankh = metadata.u.legacy.bankh;
+ surface.u.legacy.tile_split = metadata.u.legacy.tile_split;
+ surface.u.legacy.mtilea = metadata.u.legacy.mtilea;
+ surface.u.legacy.num_banks = metadata.u.legacy.num_banks;
+
+ if (metadata.u.legacy.macrotile == RADEON_LAYOUT_TILED)
+ array_mode = RADEON_SURF_MODE_2D;
+ else if (metadata.u.legacy.microtile == RADEON_LAYOUT_TILED)
+ array_mode = RADEON_SURF_MODE_1D;
+ else
+ array_mode = RADEON_SURF_MODE_LINEAR_ALIGNED;
+
+ is_scanout = metadata.u.legacy.scanout;
+ }
r = r600_init_surface(rscreen, &surface, templ, array_mode, stride,
- offset, true, metadata.scanout, false, false);
+ offset, true, is_scanout, false, false);
if (r) {
return NULL;
}
if (!rtex)
return NULL;
- rtex->resource.is_shared = true;
+ rtex->resource.b.is_shared = true;
rtex->resource.external_usage = usage;
if (rscreen->apply_opaque_metadata)
rscreen->apply_opaque_metadata(rscreen, rtex, &metadata);
+ /* Validate that addrlib arrived at the same surface parameters. */
+ if (rscreen->chip_class >= GFX9) {
+ assert(metadata.u.gfx9.swizzle_mode == surface.u.gfx9.surf.swizzle_mode);
+ }
+
return &rtex->resource.b.b;
}
{
/* r600g doesn't react to dirty_tex_descriptor_counter */
return rscreen->chip_class >= SI &&
- !rtex->resource.is_shared &&
+ !rtex->resource.b.is_shared &&
!(transfer_usage & PIPE_TRANSFER_READ) &&
rtex->resource.b.b.last_level == 0 &&
util_texrange_covers_whole_level(&rtex->resource.b.b, 0,
/* There is no point in discarding depth and tiled buffers. */
assert(!rtex->is_depth);
- assert(rtex->surface.level[0].mode == RADEON_SURF_MODE_LINEAR_ALIGNED);
+ assert(rtex->surface.is_linear);
/* Reallocate the buffer in the same pipe_resource. */
r600_alloc_resource(rscreen, &rtex->resource);
rtex->cmask.base_address_reg =
(rtex->resource.gpu_address + rtex->cmask.offset) >> 8;
- r600_dirty_all_framebuffer_states(rscreen);
- p_atomic_inc(&rscreen->dirty_tex_descriptor_counter);
+ p_atomic_inc(&rscreen->dirty_tex_counter);
rctx->num_alloc_tex_transfer_bytes += rtex->size;
}
bool use_staging_texture = false;
assert(!(texture->flags & R600_RESOURCE_FLAG_TRANSFER));
+ assert(box->width && box->height && box->depth);
/* Depth textures use staging unconditionally. */
if (!rtex->is_depth) {
/* Tiled textures need to be converted into a linear texture for CPU
* access. The staging texture is always linear and is placed in GART.
*
- * Reading from VRAM is slow, always use the staging texture in
- * this case.
+ * Reading from VRAM or GTT WC is slow, always use the staging
+ * texture in this case.
*
* Use the staging texture for uploads if the underlying BO
* is busy.
*/
- if (rtex->surface.level[0].mode >= RADEON_SURF_MODE_1D)
+ if (!rtex->surface.is_linear)
use_staging_texture = true;
else if (usage & PIPE_TRANSFER_READ)
- use_staging_texture = (rtex->resource.domains &
- RADEON_DOMAIN_VRAM) != 0;
+ use_staging_texture =
+ rtex->resource.domains & RADEON_DOMAIN_VRAM ||
+ rtex->resource.flags & RADEON_FLAG_GTT_WC;
/* Write & linear only: */
else if (r600_rings_is_buffer_referenced(rctx, rtex->resource.buf,
RADEON_USAGE_READWRITE) ||
trans = CALLOC_STRUCT(r600_transfer);
if (!trans)
return NULL;
- trans->transfer.resource = texture;
- trans->transfer.level = level;
- trans->transfer.usage = usage;
- trans->transfer.box = *box;
+ pipe_resource_reference(&trans->b.b.resource, texture);
+ trans->b.b.level = level;
+ trans->b.b.usage = usage;
+ trans->b.b.box = *box;
if (rtex->is_depth) {
struct r600_texture *staging_depth;
0, 0, 0, box->depth, 0, 0);
pipe_resource_reference(&temp, NULL);
}
- }
- else {
+
+ /* Just get the strides. */
+ r600_texture_get_offset(rctx->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 (!r600_init_flushed_depth_texture(ctx, texture, &staging_depth)) {
box->z, box->z + box->depth - 1,
0, 0);
- offset = r600_texture_get_offset(staging_depth, level, box);
+ offset = r600_texture_get_offset(rctx->screen, staging_depth,
+ level, box,
+ &trans->b.b.stride,
+ &trans->b.b.layer_stride);
}
- trans->transfer.stride = staging_depth->surface.level[level].pitch_bytes;
- trans->transfer.layer_stride = staging_depth->surface.level[level].slice_size;
trans->staging = (struct r600_resource*)staging_depth;
buf = trans->staging;
} else if (use_staging_texture) {
return NULL;
}
trans->staging = &staging->resource;
- trans->transfer.stride = staging->surface.level[0].pitch_bytes;
- trans->transfer.layer_stride = staging->surface.level[0].slice_size;
+
+ /* Just get the strides. */
+ r600_texture_get_offset(rctx->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);
buf = trans->staging;
} else {
/* the resource is mapped directly */
- trans->transfer.stride = rtex->surface.level[level].pitch_bytes;
- trans->transfer.layer_stride = rtex->surface.level[level].slice_size;
- offset = r600_texture_get_offset(rtex, level, box);
+ offset = r600_texture_get_offset(rctx->screen, rtex, level, box,
+ &trans->b.b.stride,
+ &trans->b.b.layer_stride);
buf = &rtex->resource;
}
return NULL;
}
- *ptransfer = &trans->transfer;
+ *ptransfer = &trans->b.b;
return map + offset;
}
rctx->num_alloc_tex_transfer_bytes = 0;
}
+ pipe_resource_reference(&transfer->resource, NULL);
FREE(transfer);
}
type1 == type2;
}
-void vi_dcc_disable_if_incompatible_format(struct r600_common_context *rctx,
+bool vi_dcc_formats_are_incompatible(struct pipe_resource *tex,
+ unsigned level,
+ enum pipe_format view_format)
+{
+ struct r600_texture *rtex = (struct r600_texture *)tex;
+
+ return vi_dcc_enabled(rtex, level) &&
+ !vi_dcc_formats_compatible(tex->format, view_format);
+}
+
+/* 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,
struct pipe_resource *tex,
unsigned level,
enum pipe_format view_format)
{
struct r600_texture *rtex = (struct r600_texture *)tex;
- if (rtex->dcc_offset &&
- rtex->surface.level[level].dcc_enabled &&
+ if (vi_dcc_enabled(rtex, level) &&
!vi_dcc_formats_compatible(tex->format, view_format))
if (!r600_texture_disable_dcc(rctx, (struct r600_texture*)tex))
rctx->decompress_dcc(&rctx->b, rtex);
struct pipe_surface *r600_create_surface_custom(struct pipe_context *pipe,
struct pipe_resource *texture,
const struct pipe_surface *templ,
+ unsigned width0, unsigned height0,
unsigned width, unsigned height)
{
- struct r600_common_context *rctx = (struct r600_common_context*)pipe;
- struct r600_texture *rtex = (struct r600_texture*)texture;
struct r600_surface *surface = CALLOC_STRUCT(r600_surface);
if (!surface)
surface->base.width = width;
surface->base.height = height;
surface->base.u = templ->u;
- surface->level_info = &rtex->surface.level[templ->u.tex.level];
- if (texture->target != PIPE_BUFFER)
- vi_dcc_disable_if_incompatible_format(rctx, texture,
- templ->u.tex.level,
- templ->format);
+ surface->width0 = width0;
+ surface->height0 = height0;
+ surface->dcc_incompatible =
+ texture->target != PIPE_BUFFER &&
+ vi_dcc_formats_are_incompatible(texture, templ->u.tex.level,
+ templ->format);
return &surface->base;
}
unsigned level = templ->u.tex.level;
unsigned width = u_minify(tex->width0, level);
unsigned height = u_minify(tex->height0, level);
+ unsigned width0 = tex->width0;
+ unsigned height0 = tex->height0;
if (tex->target != PIPE_BUFFER && templ->format != tex->format) {
const struct util_format_description *tex_desc
width = nblks_x * templ_desc->block.width;
height = nblks_y * templ_desc->block.height;
+
+ width0 = util_format_get_nblocksx(tex->format, width0);
+ height0 = util_format_get_nblocksy(tex->format, height0);
}
}
- return r600_create_surface_custom(pipe, tex, templ, width, height);
+ return r600_create_surface_custom(pipe, tex, templ,
+ width0, height0,
+ width, height);
}
static void r600_surface_destroy(struct pipe_context *pipe,
/* The intent is to use this with shared displayable back buffers,
* but it's not strictly limited only to them.
*/
- if (!tex->resource.is_shared ||
+ if (!tex->resource.b.is_shared ||
!(tex->resource.external_usage & PIPE_HANDLE_USAGE_EXPLICIT_FLUSH) ||
tex->resource.b.b.target != PIPE_TEXTURE_2D ||
tex->resource.b.b.last_level > 0 ||
if (!vi_should_enable_separate_dcc(tex))
return; /* stats show that DCC decompression is too expensive */
- assert(tex->surface.level[0].dcc_enabled);
+ assert(tex->surface.num_dcc_levels);
assert(!tex->dcc_separate_buffer);
r600_texture_discard_cmask(rctx->screen, tex);
tex->last_dcc_separate_buffer = NULL;
} else {
tex->dcc_separate_buffer = (struct r600_resource*)
- r600_aligned_buffer_create(rctx->b.screen, 0,
+ r600_aligned_buffer_create(rctx->b.screen,
+ R600_RESOURCE_FLAG_UNMAPPABLE,
PIPE_USAGE_DEFAULT,
tex->surface.dcc_size,
tex->surface.dcc_alignment);
memset(&uc, 0, sizeof(uc));
- if (util_format_get_blocksizebits(surface_format) == 128) {
+ if (rtex->surface.bpe == 16) {
/* DCC fast clear only:
* CLEAR_WORD0 = R = G = B
* CLEAR_WORD1 = A
bool main_value = false;
bool extra_value = false;
int extra_channel;
+
+ /* This is needed to get the correct DCC clear value for luminance formats.
+ * 1) Get the linear format (because the next step can't handle L8_SRGB).
+ * 2) Convert luminance to red. (the real hw format for luminance)
+ */
+ surface_format = util_format_linear(surface_format);
+ surface_format = util_format_luminance_to_red(surface_format);
+
const struct util_format_description *desc = util_format_description(surface_format);
if (desc->block.bits == 128 &&
if (surface_format == PIPE_FORMAT_R11G11B10_FLOAT ||
surface_format == PIPE_FORMAT_B5G6R5_UNORM ||
- surface_format == PIPE_FORMAT_B5G6R5_SRGB) {
+ surface_format == PIPE_FORMAT_B5G6R5_SRGB ||
+ util_format_is_alpha(surface_format)) {
extra_channel = -1;
} else if (desc->layout == UTIL_FORMAT_LAYOUT_PLAIN) {
if(r600_translate_colorswap(surface_format, false) <= 1)
unsigned level, unsigned clear_value)
{
struct pipe_resource *dcc_buffer;
- uint64_t dcc_offset;
+ uint64_t dcc_offset, clear_size;
- assert(rtex->dcc_offset && rtex->surface.level[level].dcc_enabled);
+ assert(vi_dcc_enabled(rtex, level));
if (rtex->dcc_separate_buffer) {
dcc_buffer = &rtex->dcc_separate_buffer->b.b;
dcc_offset = rtex->dcc_offset;
}
- dcc_offset += rtex->surface.level[level].dcc_offset;
+ if (rctx->chip_class >= GFX9) {
+ /* Mipmap level clears aren't implemented. */
+ assert(rtex->resource.b.b.last_level == 0);
+ /* MSAA needs a different clear size. */
+ assert(rtex->resource.b.b.nr_samples <= 1);
+ clear_size = rtex->surface.dcc_size;
+ } else {
+ dcc_offset += rtex->surface.u.legacy.level[level].dcc_offset;
+ clear_size = rtex->surface.u.legacy.level[level].dcc_fast_clear_size;
+ }
- rctx->clear_buffer(&rctx->b, dcc_buffer, dcc_offset,
- rtex->surface.level[level].dcc_fast_clear_size,
+ rctx->clear_buffer(&rctx->b, dcc_buffer, dcc_offset, clear_size,
clear_value, R600_COHERENCY_CB_META);
}
static void si_set_optimal_micro_tile_mode(struct r600_common_screen *rscreen,
struct r600_texture *rtex)
{
- if (rtex->resource.is_shared ||
+ if (rtex->resource.b.is_shared ||
rtex->resource.b.b.nr_samples <= 1 ||
rtex->surface.micro_tile_mode == rtex->last_msaa_resolve_target_micro_mode)
return;
- assert(rtex->surface.level[0].mode == RADEON_SURF_MODE_2D);
+ assert(rscreen->chip_class >= GFX9 ||
+ rtex->surface.u.legacy.level[0].mode == RADEON_SURF_MODE_2D);
assert(rtex->resource.b.b.last_level == 0);
- /* These magic numbers were copied from addrlib. It doesn't use any
- * definitions for them either. They are all 2D_TILED_THIN1 modes with
- * different bpp and micro tile mode.
- */
- if (rscreen->chip_class >= CIK) {
+ if (rscreen->chip_class >= GFX9) {
+ /* 4K or larger tiles only. 0 is linear. 1-3 are 256B tiles. */
+ assert(rtex->surface.u.gfx9.surf.swizzle_mode >= 4);
+
+ /* If you do swizzle_mode % 4, you'll get:
+ * 0 = Depth
+ * 1 = Standard,
+ * 2 = Displayable
+ * 3 = Rotated
+ *
+ * Depth-sample order isn't allowed:
+ */
+ assert(rtex->surface.u.gfx9.surf.swizzle_mode % 4 != 0);
+
switch (rtex->last_msaa_resolve_target_micro_mode) {
- case 0: /* displayable */
- rtex->surface.tiling_index[0] = 10;
+ case RADEON_MICRO_MODE_DISPLAY:
+ rtex->surface.u.gfx9.surf.swizzle_mode &= ~0x3;
+ rtex->surface.u.gfx9.surf.swizzle_mode += 2; /* D */
break;
- case 1: /* thin */
- rtex->surface.tiling_index[0] = 14;
+ case RADEON_MICRO_MODE_THIN:
+ rtex->surface.u.gfx9.surf.swizzle_mode &= ~0x3;
+ rtex->surface.u.gfx9.surf.swizzle_mode += 1; /* S */
break;
- case 3: /* rotated */
- rtex->surface.tiling_index[0] = 28;
+ case RADEON_MICRO_MODE_ROTATED:
+ rtex->surface.u.gfx9.surf.swizzle_mode &= ~0x3;
+ rtex->surface.u.gfx9.surf.swizzle_mode += 3; /* R */
+ break;
+ default: /* depth */
+ assert(!"unexpected micro mode");
+ return;
+ }
+ } else if (rscreen->chip_class >= CIK) {
+ /* These magic numbers were copied from addrlib. It doesn't use
+ * any definitions for them either. They are all 2D_TILED_THIN1
+ * modes with different bpp and micro tile mode.
+ */
+ switch (rtex->last_msaa_resolve_target_micro_mode) {
+ case RADEON_MICRO_MODE_DISPLAY:
+ rtex->surface.u.legacy.tiling_index[0] = 10;
+ break;
+ case RADEON_MICRO_MODE_THIN:
+ rtex->surface.u.legacy.tiling_index[0] = 14;
+ break;
+ case RADEON_MICRO_MODE_ROTATED:
+ rtex->surface.u.legacy.tiling_index[0] = 28;
break;
default: /* depth, thick */
assert(!"unexpected micro mode");
}
} else { /* SI */
switch (rtex->last_msaa_resolve_target_micro_mode) {
- case 0: /* displayable */
+ case RADEON_MICRO_MODE_DISPLAY:
switch (rtex->surface.bpe) {
case 1:
- rtex->surface.tiling_index[0] = 10;
+ rtex->surface.u.legacy.tiling_index[0] = 10;
break;
case 2:
- rtex->surface.tiling_index[0] = 11;
+ rtex->surface.u.legacy.tiling_index[0] = 11;
break;
default: /* 4, 8 */
- rtex->surface.tiling_index[0] = 12;
+ rtex->surface.u.legacy.tiling_index[0] = 12;
break;
}
break;
- case 1: /* thin */
+ case RADEON_MICRO_MODE_THIN:
switch (rtex->surface.bpe) {
case 1:
- rtex->surface.tiling_index[0] = 14;
+ rtex->surface.u.legacy.tiling_index[0] = 14;
break;
case 2:
- rtex->surface.tiling_index[0] = 15;
+ rtex->surface.u.legacy.tiling_index[0] = 15;
break;
case 4:
- rtex->surface.tiling_index[0] = 16;
+ rtex->surface.u.legacy.tiling_index[0] = 16;
break;
default: /* 8, 16 */
- rtex->surface.tiling_index[0] = 17;
+ rtex->surface.u.legacy.tiling_index[0] = 17;
break;
}
break;
rtex->surface.micro_tile_mode = rtex->last_msaa_resolve_target_micro_mode;
- p_atomic_inc(&rscreen->dirty_fb_counter);
- p_atomic_inc(&rscreen->dirty_tex_descriptor_counter);
+ p_atomic_inc(&rscreen->dirty_tex_counter);
}
void evergreen_do_fast_color_clear(struct r600_common_context *rctx,
struct pipe_framebuffer_state *fb,
struct r600_atom *fb_state,
- unsigned *buffers, unsigned *dirty_cbufs,
+ unsigned *buffers, ubyte *dirty_cbufs,
const union pipe_color_union *color)
{
int i;
}
/* only supported on tiled surfaces */
- if (tex->surface.level[0].mode < RADEON_SURF_MODE_1D) {
+ if (tex->surface.is_linear) {
continue;
}
* because there is no way to communicate the clear color among
* all clients
*/
- if (tex->resource.is_shared &&
+ if (tex->resource.b.is_shared &&
!(tex->resource.external_usage & PIPE_HANDLE_USAGE_EXPLICIT_FLUSH))
continue;
/* fast color clear with 1D tiling doesn't work on old kernels and CIK */
- if (tex->surface.level[0].mode == RADEON_SURF_MODE_1D &&
- rctx->chip_class >= CIK &&
+ if (rctx->chip_class == CIK &&
+ tex->surface.u.legacy.level[0].mode == RADEON_SURF_MODE_1D &&
rctx->screen->info.drm_major == 2 &&
rctx->screen->info.drm_minor < 38) {
continue;
!(rctx->screen->debug_flags & DBG_NO_DCC_FB)) {
vi_separate_dcc_try_enable(rctx, tex);
- /* Stoney can't do a CMASK-based clear, so all clears are
- * considered to be hypothetically slow clears, which
- * is weighed when determining to enable separate DCC.
+ /* RB+ isn't supported with a CMASK clear only on Stoney,
+ * so all clears are considered to be hypothetically slow
+ * clears, which is weighed when determining whether to
+ * enable separate DCC.
*/
if (tex->dcc_gather_statistics &&
rctx->family == CHIP_STONEY)
}
/* Try to clear DCC first, otherwise try CMASK. */
- if (tex->dcc_offset && tex->surface.level[0].dcc_enabled) {
+ if (vi_dcc_enabled(tex, 0)) {
uint32_t reset_value;
bool clear_words_needed;
+ /* TODO: fix DCC clear */
+ if (rctx->chip_class >= GFX9)
+ continue;
+
if (rctx->screen->debug_flags & DBG_NO_DCC_CLEAR)
continue;
vi_dcc_clear_level(rctx, tex, 0, reset_value);
- if (clear_words_needed)
- tex->dirty_level_mask |= 1 << fb->cbufs[i]->u.tex.level;
+ unsigned level_bit = 1 << fb->cbufs[i]->u.tex.level;
+ if (clear_words_needed) {
+ bool need_compressed_update = !tex->dirty_level_mask;
+
+ tex->dirty_level_mask |= level_bit;
+
+ if (need_compressed_update)
+ p_atomic_inc(&rctx->screen->compressed_colortex_counter);
+ }
tex->separate_dcc_dirty = true;
} else {
/* 128-bit formats are unusupported */
- if (util_format_get_blocksizebits(fb->cbufs[i]->format) > 64) {
+ if (tex->surface.bpe > 8) {
continue;
}
- /* Stoney/RB+ doesn't work with CMASK fast clear. */
+ /* RB+ doesn't work with CMASK fast clear on Stoney. */
if (rctx->family == CHIP_STONEY)
continue;
tex->cmask.offset, tex->cmask.size, 0,
R600_COHERENCY_CB_META);
+ bool need_compressed_update = !tex->dirty_level_mask;
+
tex->dirty_level_mask |= 1 << fb->cbufs[i]->u.tex.level;
+
+ if (need_compressed_update)
+ p_atomic_inc(&rctx->screen->compressed_colortex_counter);
}
/* We can change the micro tile mode before a full clear. */