{
*stride = rtex->surface.u.legacy.level[level].nblk_x *
rtex->surface.bpe;
- *layer_stride = rtex->surface.u.legacy.level[level].slice_size;
+ assert((uint64_t)rtex->surface.u.legacy.level[level].slice_size_dw * 4 <= UINT_MAX);
+ *layer_stride = (uint64_t)rtex->surface.u.legacy.level[level].slice_size_dw * 4;
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->z * (uint64_t)rtex->surface.u.legacy.level[level].slice_size_dw * 4 +
(box->y / rtex->surface.blk_h *
rtex->surface.u.legacy.level[level].nblk_x +
box->x / rtex->surface.blk_w) * rtex->surface.bpe;
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) {
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);
+ r = rscreen->ws->surface_init(rscreen->ws, ptex,
+ flags, bpe, array_mode, surface);
if (r) {
return r;
}
* 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;
+ surface->u.legacy.level[0].slice_size_dw =
+ ((uint64_t)pitch_in_bytes_override * surface->u.legacy.level[0].nblk_y) / 4;
}
if (offset) {
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,
&rtex->resource.b.b, i, &box);
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;
+ slice_size = (uint64_t)rtex->surface.u.legacy.level[0].slice_size_dw * 4;
} else {
/* Move a suballocated buffer into a non-suballocated allocation. */
if (rscreen->ws->buffer_is_suballocated(res->buf)) {
struct r600_resource *resource = &rtex->resource;
r600_texture_reference(&rtex->flushed_depth_texture, NULL);
+ pipe_resource_reference((struct pipe_resource**)&resource->immed_buffer, NULL);
if (rtex->cmask_buffer != &rtex->resource) {
r600_resource_reference(&rtex->cmask_buffer, NULL);
bpe *= 2;
}
- if (rscreen->ws->surface_init(rscreen->ws, &templ, flags, bpe,
- RADEON_SURF_MODE_2D, &fmask)) {
+ if (rscreen->ws->surface_init(rscreen->ws, &templ,
+ flags, bpe, RADEON_SURF_MODE_2D, &fmask)) {
R600_ERR("Got error in surface_init while allocating FMASK.\n");
return;
}
out->slice_tile_max = ((pitch_elements * height) / (128*128)) - 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);
}
p_atomic_inc(&rscreen->compressed_colortex_counter);
}
+void eg_resource_alloc_immed(struct r600_common_screen *rscreen,
+ struct r600_resource *res,
+ unsigned immed_size)
+{
+ res->immed_buffer = (struct r600_resource *)
+ pipe_buffer_create(&rscreen->b, PIPE_BIND_CUSTOM,
+ PIPE_USAGE_DEFAULT, immed_size);
+}
+
static void r600_texture_get_htile_size(struct r600_common_screen *rscreen,
struct r600_texture *rtex)
{
return;
}
- width = align(rtex->resource.b.b.width0, cl_width * 8);
- height = align(rtex->resource.b.b.height0, cl_height * 8);
+ width = align(rtex->surface.u.legacy.level[0].nblk_x, cl_width * 8);
+ height = align(rtex->surface.u.legacy.level[0].nblk_y, cl_height * 8);
slice_elements = (width * height) / (8 * 8);
slice_bytes = slice_elements * 4;
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);
}
rtex->cmask.slice_tile_max);
if (rtex->htile_offset)
- u_log_printf(log, " HTile: offset=%"PRIu64", size=%"PRIu64", "
+ u_log_printf(log, " HTile: offset=%"PRIu64", size=%u "
"alignment=%u\n",
rtex->htile_offset, rtex->surface.htile_size,
rtex->surface.htile_alignment);
"npix_x=%u, npix_y=%u, npix_z=%u, nblk_x=%u, nblk_y=%u, "
"mode=%u, tiling_index = %u\n",
i, rtex->surface.u.legacy.level[i].offset,
- rtex->surface.u.legacy.level[i].slice_size,
+ (uint64_t)rtex->surface.u.legacy.level[i].slice_size_dw * 4,
u_minify(rtex->resource.b.b.width0, i),
u_minify(rtex->resource.b.b.height0, i),
u_minify(rtex->resource.b.b.depth0, i),
"npix_y=%u, npix_z=%u, nblk_x=%u, nblk_y=%u, "
"mode=%u, tiling_index = %u\n",
i, rtex->surface.u.legacy.stencil_level[i].offset,
- rtex->surface.u.legacy.stencil_level[i].slice_size,
+ (uint64_t)rtex->surface.u.legacy.stencil_level[i].slice_size_dw * 4,
u_minify(rtex->resource.b.b.width0, i),
u_minify(rtex->resource.b.b.height0, i),
u_minify(rtex->resource.b.b.depth0, i),
r600_init_resource_fields(rscreen, resource, rtex->size,
rtex->surface.surf_alignment);
- /* Displayable surfaces are not suballocated. */
- if (resource->b.b.bind & PIPE_BIND_SCANOUT)
- resource->flags |= RADEON_FLAG_NO_SUBALLOC;
-
if (!r600_alloc_resource(rscreen, resource)) {
FREE(rtex);
return NULL;
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));
}
if (templ->bind & PIPE_BIND_LINEAR)
return RADEON_SURF_MODE_LINEAR_ALIGNED;
- /* Textures with a very small height are recommended to be linear. */
+ /* 1D textures should be linear - fixes image operations on 1d */
if (templ->target == PIPE_TEXTURE_1D ||
- templ->target == PIPE_TEXTURE_1D_ARRAY ||
- /* Only very thin and long 2D textures should benefit from
- * linear_aligned. */
- (templ->width0 > 8 && templ->height0 <= 2))
+ templ->target == PIPE_TEXTURE_1D_ARRAY)
return RADEON_SURF_MODE_LINEAR_ALIGNED;
/* Textures likely to be mapped often. */
templ->depth0 != 1 || templ->last_level != 0)
return NULL;
- buf = rscreen->ws->buffer_from_handle(rscreen->ws, whandle, &stride, &offset);
+ buf = rscreen->ws->buffer_from_handle(rscreen->ws, whandle,
+ rscreen->info.max_alignment,
+ &stride, &offset);
if (!buf)
return NULL;
* 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, RADEON_FLUSH_ASYNC, NULL);
+ rctx->gfx.flush(rctx, PIPE_FLUSH_ASYNC, NULL);
rctx->num_alloc_tex_transfer_bytes = 0;
}
desc->unpack_rgba_float(color.f, 0, data, 0, 1, 1);
if (screen->is_format_supported(screen, tex->format,
- tex->target, 0,
+ tex->target, 0, 0,
PIPE_BIND_RENDER_TARGET)) {
pipe->clear_render_target(pipe, sf, &color,
box->x, box->y,
!(tex->resource.external_usage & PIPE_HANDLE_USAGE_EXPLICIT_FLUSH))
continue;
+ /* Use a slow clear for small surfaces where the cost of
+ * the eliminate pass can be higher than the benefit of fast
+ * clear. AMDGPU-pro does this, but the numbers may differ.
+ *
+ * This helps on both dGPUs and APUs, even small ones.
+ */
+ if (tex->resource.b.b.nr_samples <= 1 &&
+ tex->resource.b.b.width0 * tex->resource.b.b.height0 <= 300 * 300)
+ continue;
+
{
/* 128-bit formats are unusupported */
if (tex->surface.bpe > 8) {
return NULL;
buf = rscreen->ws->buffer_from_handle(rscreen->ws, whandle,
+ rscreen->info.max_alignment,
&stride, &offset);
if (!buf) {
free(memobj);
pb_reference(&buf, memobj->buf);
rtex->resource.b.is_shared = true;
- rtex->resource.external_usage = PIPE_HANDLE_USAGE_READ_WRITE;
+ rtex->resource.external_usage = PIPE_HANDLE_USAGE_FRAMEBUFFER_WRITE;
if (rscreen->apply_opaque_metadata)
rscreen->apply_opaque_metadata(rscreen, rtex, &metadata);
projects / mesa.git / blobdiff