Repository navigation
#1795: window-space + macOS decorations blend in linear light, sampled as declared (ADR-044 §7) - #1799
Merged
Conversation
…d as declared (#1795) vk_hud_blend sampled every window-space source through a hard-coded R8G8B8A8_UNORM view and blended into the encoded atlas, so the layer's declared format was ignored (37 and 43 gave byte-identical atlases) and a B8G8R8A8 source would have been read with R and B swapped. - vk_hud_blend: views cached per (image, format); new vk_hud_blend_draw_no_layout_fmt() samples the source as a given format. Existing callers keep the legacy view. - vk_native: the window-space pass raw-copies the atlas into the private compose target (UNORM + MUTABLE, _SRGB view), blends each layer through that view with the source read as its declared format, and raw-copies back. Two atlas copies, only on frames with window-space layers. Falls back to the encoded path under the legacy hatch or if anything cannot be created. Measured on the win box (windowspace_handle_vk_win, opaque byte 38): _SRGB 38, UNORM 108 (one encode) - now identical to D3D11/D3D12/GL. The cube_handle_vk_win HUD's 50% backdrop lands at (7,7,45), byte-identical to D3D11. The comp_multi (macOS/Linux service) blends are the next step. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ampled as declared (#1795) The workspace chrome/cursor/overlay decorations on the macOS shared surface were sampled through vk_hud_blend's hard-coded R8G8B8A8_UNORM view (illegal on a non-mutable _SRGB/BGRA image) and blended into the encoded atlas, so a surface's declared format was ignored. The atlas is now created MUTABLE_FORMAT with its _SRGB sibling; decorations render through that view with each source read as its declared format. Legacy hatch / any creation failure keeps the old path (sticky, one WARN). Also adds the atlas's missing COLOR_ATTACHMENT usage. ORDERING: the macOS shell submits UNORM display-referred bytes today and must move to _SRGB (displayxr-shell-pvt#121) in the same release. Linux registers no workspace surfaces, so it has no such site. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…1795) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
dfattal
marked this pull request as ready for review
October 2, 2026 23:58
dfattal
added a commit
that referenced
this pull request
Oct 3, 2026
…clared (#1795) The over-flatten and the 2D-under backdrop flatten sampled every Local2D layer through the non-decoding view into a UNORM scratch, so the declared format was ignored (a true-linear UNORM layer rendered too dark). Both scratches are now B8G8R8A8_UNORM + MUTABLE_FORMAT with a {UNORM,_SRGB} format list (the #1799 compose-target shape): the UNORM view is what every reader samples, so they still see ENCODED bytes, and the flatten renders through an _SRGB view with a second flatten render pass + pipeline pair (vk_local2d_composite_init_srgb_flatten, opt-in, so the Linux/macOS/Android multi-compositor callers are unchanged). The source view follows the target: format-honest (true view) into an _SRGB attachment, non-decoding into a UNORM one. The legacy hatch never builds the _SRGB pipelines and keeps the old path exactly; so does the weave-on-scanout split, whose bridge planes are typed-UNORM D3D11 textures. Measured (hardware A/B pending): cube_handle_vk_win, DXR_LOCAL2D_PANEL=1, opaque checker authored 40/235 -- expect _SRGB panel 40/235; UNORM panel 110/246 (one encode); UNORM under DXR_COLOR_LEGACY_UNORM_ENCODED=1 40/235. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
dfattal
added a commit
that referenced
this pull request
Oct 3, 2026
…clared (#1795) The over-flatten and the 2D-under backdrop flatten sampled every Local2D layer through the non-decoding view into a UNORM scratch, so the declared format was ignored (a true-linear UNORM layer rendered too dark). Both scratches are now B8G8R8A8_UNORM + MUTABLE_FORMAT with a {UNORM,_SRGB} format list (the #1799 compose-target shape): the UNORM view is what every reader samples, so they still see ENCODED bytes, and the flatten renders through an _SRGB view with a second flatten render pass + pipeline pair (vk_local2d_composite_init_srgb_flatten, opt-in, so the Linux/macOS/Android multi-compositor callers are unchanged). The source view follows the target: format-honest (true view) into an _SRGB attachment, non-decoding into a UNORM one. The legacy hatch never builds the _SRGB pipelines and keeps the old path exactly; so does the weave-on-scanout split, whose bridge planes are typed-UNORM D3D11 textures. Measured (hardware A/B pending): cube_handle_vk_win, DXR_LOCAL2D_PANEL=1, opaque checker authored 40/235 -- expect _SRGB panel 40/235; UNORM panel 110/246 (one encode); UNORM under DXR_COLOR_LEGACY_UNORM_ENCODED=1 40/235. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fixes the window-space half of #1795 (ADR-044 §7). Draft until the macOS test in the
tmp-adr044-mac-colourchannel passes. The release containing this is v2.23.0 and must ship together with displayxr-shell-pvt#121.Changes
vk_hud_blend: source views are cached per (image, format), and the newvk_hud_blend_draw_no_layout_fmt()samples a source as a given format. Existing callers keep the legacyR8G8B8A8_UNORMview._SRGBview), blends each layer through that view with the source read as its declared format, then raw-copies back._SRGBview of the now-MUTABLE atlas, sampled as declared.COLOR_ATTACHMENTusage.Measured (win box, in-process vk_native, dev runtime)
_SRGBwindowspace_handle_vk_win(opaque byte 38)cube_handle_vk_winHUD 50% backdrop over (13,13,64)The hard-coded view's R/B swap for
B8G8R8A8sources goes away with this change.Ordering / compatibility
_SRGB) before this ships._SRGBonly on runtime ≥ v2.23.0. Shell Shell: intermittent black frame flash in multi-app mode #121 on an old runtime looks the same as today; the old shell on this runtime would be washed out, so the two ship together.Not yet verified
_SRGBclients may render too dark (decoded into the UNORM atlas, never re-encoded). Not changed here.🤖 Generated with Claude Code