Files
DroidDeck/app/src/main/cpp/waylandcomp/HDR_RECON.md
T

63 KiB
Raw Blame History

HDR on the Wayland backend - reconnaissance

2026-09-14: round 1 of Phase A is built (branch feat/wayland-hdr, opt-in, testing only) - see §10 at the end for what was built, how it is gated and how a tester's log proves it. The recon below is kept as written.

Written 2026-09-13. Recon only: no product code was changed, nothing was built, nothing was installed on the device. Branch surveyed: app feat/wayland-phase2, proton-wine feat/winewayland-desktop-11.0-2, Banners-Turnip wayland (Mesa 7cda7850, v26.3.0). Device reads are dumpsys / getprop / strings only.

Companion docs: ZERO_COPY_SPIKE.md (buffer path), WAYLAND_RUNTIME.md.


0. Answer in one paragraph

The game half of HDR is already shipping and nobody noticed. The Wayland Turnip inside wcp v6 (Proton-11.0-2.1-arm64ec-6/lib/libvulkan_freedreno_wayland.so, verified on the device) already contains Mesa's full wp_color_manager_v1 client implementation and advertises VK_EXT_swapchain_colorspace + VK_EXT_hdr_metadata; the shipping DXVK v3.1-gplasync already carries dxgi.enableHDR / DXVK_HDR and the VK_COLOR_SPACE_HDR10_ST2084_EXT path; winewayland.drv is a transparent 121-line shim that neither adds nor removes colour information. Everything that is missing is in our compositor, which advertises no colour protocol, offers only four 8-bit DRM formats over zwp_linux_dmabuf_v1, takes surfaceFormats[0] for its own swapchain, runs an 8-bit scene/effects chain, and never calls ASurfaceTransaction_setBufferDataSpace. The work is therefore app-side only - the memory note in project_bannerlator_wayland_phase4 that HDR "does need a new wcp" is wrong for the HDR10 (10-bit PQ) path; a new wcp is only needed for scRGB/FP16 zero-copy. Estimated 4–6 engineer-days to a plumbed, log-verifiable HDR10 path. But the Pocket FIT's panel reports supportedHdrTypes=[], supportedColorModes=[0] (NATIVE only), hdrSdrRatio not_available and displayBrightnessNits=-1, so nothing of it can be seen on this device, and turning it on there would actively cost us the zero-copy DEVICE-composition win. Verdict in §8: not now.


1. Game side - how far an HDR request gets today

1.1 What a Windows game does

Call Where it lands in DXVK Notes
IDXGIOutput6::GetDesc1 → DXGI_OUTPUT_DESC1::ColorSpace src/dxgi/dxgi_output.cpp:220-259 Never queries Vulkan. Reports RGB_FULL_G2084_NONE_P2020 purely because dxgi.enableHDR is on.
IDXGISwapChain3::CheckColorSpaceSupport src/dxgi/dxgi_swapchain.cpp:655-667 This is backed by Vulkan (below). Note it is a swapchain method, not a factory method.
IDXGISwapChain3::SetColorSpace1 dxgi_swapchain.cpp:672-688 → d3d11_swapchain.cpp:306-331 → Presenter::supportsColorSpace (dxvk_presenter.cpp:319-343) Returns E_INVALIDARG if the Vulkan surface does not list the colour space.
IDXGISwapChain4::SetHDRMetaData dxgi_swapchain.cpp:690-717 → vkSetHdrMetadataEXT Only DXGI_HDR_METADATA_TYPE_HDR10. Silently dropped if VK_EXT_hdr_metadata is absent (dxvk_presenter.cpp:143-149).

DXGI→Vulkan mapping (src/d3d11/d3d11_swapchain.cpp:21-28):

DXGI_COLOR_SPACE_RGB_FULL_G2084_NONE_P2020 -> VK_COLOR_SPACE_HDR10_ST2084_EXT
DXGI_COLOR_SPACE_RGB_FULL_G10_NONE_P709    -> VK_COLOR_SPACE_EXTENDED_SRGB_LINEAR_EXT  (scRGB)
DXGI_COLOR_SPACE_RGB_FULL_G22_NONE_P709    -> VK_COLOR_SPACE_SRGB_NONLINEAR_KHR

Swapchain formats DXVK will accept per colour space (Presenter::pickFormat, dxvk_presenter.cpp:1043-1081):

  • HDR10_ST2084 → A2R10G10B10_UNORM_PACK32, A2B10G10R10_UNORM_PACK32, E5B9G9R9_UFLOAT_PACK32, R16G16B16A16_UNORM, R16G16B16A16_SFLOAT
  • EXTENDED_SRGB_LINEAR (scRGB) → R16G16B16A16_SFLOAT only

VK_EXT_hdr_metadata is optional in both DXVK and vkd3d-proton. VK_EXT_swapchain_colorspace is not requested as an instance extension by DXVK's win32 WSI - it simply trusts the driver to report the extra VkColorSpaceKHR values from vkGetPhysicalDeviceSurfaceFormatsKHR. Mesa's Wayland WSI does exactly that (§2.3), so this is not a problem for us.

vkd3d-proton (D3D12) ships its own swapchain (libs/vkd3d/swapchain.c), same colour-space mapping (convert_color_space, :1717-1730), but no fallback table: if the requested HDR colour space is not in vkGetPhysicalDeviceSurfaceFormatsKHR, dxgi_vk_swap_chain_select_format (:1773-1791) returns false with the comment "Refuse to present unsupported HDR since it will look completely bogus", and the caller skips swapchain recreation for that cycle. A D3D12 game that force-sets HDR without checking can therefore stall presentation rather than degrade. DXGI enumeration for D3D12 titles is still DXVK's dxgi.dll, so DXVK_HDR still drives DXGI_OUTPUT_DESC1.

Verified in our shipped binaries (device, strings): contents/DXVK/v3.1-gplasync-0/system32/dxgi.dll contains dxgi.enableHDR, DXVK_HDR, VK_COLOR_SPACE_HDR10_ST2084_EXT; d3d11.dll/d3d9.dll/dxgi.dll contain VK_COLOR_SPACE_EXTENDED_SRGB_LINEAR_EXT; contents/VKD3D/3.1.0-wave64-relax-1/.../d3d12core.dll contains VK_EXT_hdr_metadata. Nothing needs rebuilding on the D3D side.

Proton-level gating: GE's proton script maps PROTON_ENABLE_HDR / PROTON_USE_HDR to DXVK_HDR=1 and nothing else. For us that is one container env var - no launcher work.

1.2 The failure mode today, exactly

DXVK's DXGI_OUTPUT_DESC1 will claim HDR10 as soon as DXVK_HDR=1, regardless of Vulkan. SetColorSpace1 then fails with E_INVALIDARG because our compositor advertises no colour management, so Mesa lists only SRGB_NONLINEAR + PASS_THROUGH (§2.3). Well-behaved games fall back to SDR. Games that trust GetDesc1 without checking CheckColorSpaceSupport can render an HDR-authored image into an SDR swapchain (washed out) - this is a pre-existing foot-gun, not something we introduce. Recommendation regardless of whether we do HDR: do not expose a DXVK_HDR container toggle until the compositor answers truthfully, or we invite exactly that bug class.

1.3 winewayland.drv - carries nothing for colour, and blocks nothing

/home/claude-user/proton-wine-wayland2-wt/dlls/winewayland.drv/vulkan.c is 121 lines: wayland_vulkan_surface_create wraps the wl_surface in a VkSurfaceKHR, wayland_map_instance_extensions aliases VK_KHR_win32_surface ⇄ VK_KHR_wayland_surface. There is no surface-format or colour-space filtering, so whatever Mesa advertises reaches DXVK unmodified. A grep for color|colour|hdr|st2084|bt2020|gamut|10.?bit across the whole driver returns only GDI colour-bitmap helpers - nothing colorimetric.

One real consequence: wayland_output.c publishes no EDID, and nothing in the tree writes the EDID registry value win32u/sysparams.c:815-818 reads. DXVK's src/wsi/win32/wsi_monitor_win32.cpp:288-361 therefore finds no EDID and falls back to its dummy HDR luminances (wsi_edid.h:38-56: min 0.01 / max 1499 / maxFullFrame 799 nits). A game will size its tonemapper for a 1499-nit display. Correctable later by synthesising an EDID in winewayland.drv (that would need a new wcp), but not a blocker.

1.4 Our bundled Turnip already speaks colour management - verified on the device

$ strings .../Proton/11.0-2.1-arm64ec-6/lib/libvulkan_freedreno_wayland.so
wp_color_manager_v1                        PRESENT
wp_color_management_surface_v1             PRESENT
wp_image_description_creator_params_v1     PRESENT
banner_ahb_v1                              PRESENT     (our zero-copy patch)
VK_EXT_swapchain_colorspace                PRESENT
VK_EXT_hdr_metadata                        PRESENT

The plain adrenotools driver the compositor itself loads (contents/adrenotools/Mesa Turnip v26.3.0-7cda785 (Android + Wayland)/libvulkan_freedreno.so) has the same colour-management symbols. Our own WSI patch is already aware of the protocol - patches/wayland/banner_ahb_wsi.py anchors one of its edits on Mesa's if (display->color_manager) wp_color_manager_v1_destroy(...) teardown line.

And the zero-copy patch already maps 10-bit to gralloc - banner_ahb_wsi.py:99-100, 225-237:

#define BANNER_AHB_FORMAT_R8G8B8A8_UNORM    1u
#define BANNER_AHB_FORMAT_R10G10B10A2_UNORM 0x2bu   /* AHARDWAREBUFFER_FORMAT_R10G10B10A2_UNORM */
...
case VK_FORMAT_A2B10G10R10_UNORM_PACK32:  return BANNER_AHB_FORMAT_R10G10B10A2_UNORM;

So a 10-bit HDR10 swapchain can already be gralloc-backed and go onto the display layer with the shipped wcp. VK_FORMAT_R16G16B16A16_SFLOAT is not mapped, so banner_ahb_skip_format (:239-242, applied at :606-617) hides FP16 surface formats whenever zero-copy is on - scRGB zero-copy is the one thing that needs a driver change (one enum: AHARDWAREBUFFER_FORMAT_R16G16B16A16_FLOAT = 0x16).

1.5 X11 has no path - confirmed

app/src/main/cpp/winlator/VulkanRendererContext.cpp:439 hardcodes swapchainFmt = VK_FORMAT_R8G8B8A8_UNORM and :492 hardcodes imageColorSpace = VK_COLOR_SPACE_SRGB_NONLINEAR_KHR. Upstream of that, the game presents through winex11.drv into our pure-Java X11 server, which has no colour extension of any kind. HDR on X11 would mean inventing a private X extension and teaching DXVK to use it. Correctly out of scope.


2. Wayland side

2.1 Protocol status

wp_color_management_v1 is merged and stable-ish: wayland-protocols MR !14, merged 2025-02-13, shipped in wayland-protocols 1.41, file staging/color-management/color-management-v1.xml. Interfaces are at version 3 upstream today (v2 added get_image_description/preferred_changed2; v3 added create_windows_bt2100 and the absolute_no_adaptation intent). Predecessors: zwp_linux_color_management_v1 → xx_color_management_v4 (the name most 2023–2025 shipping code targeted) → wp_* on landing. frog_color_management_v1 (Valve) is still vendored by gamescope alongside the real one and is still functionally load-bearing on the Deck.

Implementers: KWin (reference impl, most mature), Mutter (merged 2025-02-25 but still does not advertise enough features for automatic HDR), wlroots 0.20 / Sway 1.12 (HDR10 on the Vulkan renderer), gamescope (v1 semantics + frog), Weston (experimental, no tone mapping). Proton-CachyOS's changelog notes that its winewayland.drv windows_bt2100 path needs protocol v3 and that "no compositor supports it yet" - everybody is converged on v1.

We only need v1. That is what Mesa binds.

2.2 Minimum viable compositor subset

At bind, on the wp_color_manager_v1 global, send one event per supported item then done:

  • supported_intent: perceptual (mandatory anyway - it is all Mesa ever asks for)
  • supported_feature: parametric (gates create_parametric_creator); optionally set_mastering_display_primaries, and extended_target_volume only if we want scRGB
  • supported_primaries_named: bt2020 (HDR10), srgb (needed for scRGB and for Mesa's "use a colour surface for sRGB too" branch), display_p3 (wide gamut)
  • supported_tf_named: st2084_pq (HDR10), ext_linear (scRGB), srgb

Then implement: create_parametric_creator → wp_image_description_creator_params_v1 (set_primaries_named, set_tf_named, set_mastering_display_primaries, set_mastering_luminance, set_max_cll, set_max_fall) → create → emit wp_image_description_v1.ready (or failed) → wp_color_manager_v1.get_surface → wp_color_management_surface_v1.set_image_description(desc, intent), double-buffered on wl_surface.commit.

Not needed: ICC creator, wp_color_management_surface_feedback_v1, wp_image_description_info_v1, output image descriptions, v2/v3 requests. Mesa uses none of them.

2.3 What Mesa's WSI does - read from the exact commit we build

Read from src/vulkan/wsi/wsi_common_wayland.c at Mesa 7cda7850 (4121 lines - the commit recorded in banners-turnip-wayland/mesa_hash.txt). Landed upstream as MR !32038 (Xaver Hugl, merged 2025-02-25, Mesa 25.1).

  • Bind: registry_handle_global binds wp_color_manager_v1 at version 1 unconditionally when advertised (:1536-1538), and installs color_manager_listener (:1217-1223).

  • Colour-space list: wsi_wl_display_determine_colorspaces (:1122-1160) always adds VK_COLOR_SPACE_SRGB_NONLINEAR_KHR and VK_COLOR_SPACE_PASS_THROUGH_EXT; everything else comes from the static colorspace_mapping[] table (:1038-1118) and is added only if the compositor advertised that entry's primaries and transfer function (and, for entries flagged needs_extended_range, feature.extended_target_volume). Table entries include HDR10_ST2084 (bt2020 + st2084_pq), EXTENDED_SRGB_LINEAR (srgb + ext_linear + extended range), DISPLAY_P3_NONLINEAR, DISPLAY_P3_LINEAR, BT709_LINEAR, BT709_NONLINEAR, BT2020_LINEAR, HDR10_HLG, ADOBERGB_LINEAR.

  • Surface formats: wsi_wl_surface_get_formats2 (:2205-2246) emits the full cross product of display.colorspaces × display.formats with no bit-depth gating:

    u_vector_foreach(cs, &display.colorspaces) {
        u_vector_foreach(disp_fmt, &display.formats) {
            if (!(disp_fmt->flags & WSI_WL_FMT_ALPHA) || !(disp_fmt->flags & WSI_WL_FMT_OPAQUE))
                continue;
            out_fmt->surfaceFormat.format    = disp_fmt->vk_format;
            out_fmt->surfaceFormat.colorSpace = *cs;
        }
    }
    

    Consequence: if we advertise BT.2020 + PQ but keep only 8-bit DRM formats, Mesa will happily offer R8G8B8A8_UNORM + HDR10_ST2084 and DXVK will not take it (pickFormat's HDR10 list is 10-bit/16-bit only) - so there is no accidental 8-bit-PQ trap, but there is also no HDR unless we also advertise a 10-bit DRM format.

  • Apply: wsi_wl_swapchain_update_colorspace (:1323-1455) builds the parametric creator, blocks on ready/failed (it dispatches the queue in a loop), sets mastering primaries / luminance / max_cll / max_fall when should_use_hdr_metadata is set for that colour space, and calls set_image_description(..., RENDER_INTENT_PERCEPTUAL). is_hdr_metadata_legal (:1277-1299) drops illegal combinations rather than risking a protocol error. On failed without extended_target_volume it retries once without metadata.

  • VK_EXT_hdr_metadata: wsi_wl_swapchain_set_hdr_metadata (:1459-1463) just stashes VkHdrMetadataEXT; the real work is in update_colorspace above.

Nothing to build in the driver for HDR10. Mesa is done.

2.4 Our compositor advertises nothing colorimetric

Globals (compositor.c:2527-2538, plus ahb_swapchain.c:406, wl_clipboard.c:522-523, wl_text_input.c:275, wl_toplevel_icon.c:35): wl_compositor v6, wl_subcompositor, wp_viewporter, wl_output v2, xdg_wm_base, zwp_linux_dmabuf_v1 v3, wl_seat v5, banner_desktop_v1, wp_presentation v2, pointer constraints, relative pointer, banner_ahb_v1 v2, data device/control, text input, toplevel icon. No wp_color_manager_v1. protocols/ has no colour-management XML, and protocol C glue is pre-generated into generated/ (no wayland-scanner at app build time - CMakeLists.txt:1-4), so adding a protocol means generating and committing the files offline.

The dmabuf format table is 8-bit only (compositor.c:1323-1326):

#define DMABUF_NFMT 4
static struct { uint32_t fmt; uint64_t mods[DMABUF_NMOD]; int n; } g_dmabuf_fmts[DMABUF_NFMT] = {
    {DRM_ARGB8888}, {DRM_XRGB8888}, {DRM_ABGR8888}, {DRM_XBGR8888}};

Good news: the machinery around it is already format-agnostic. vkp_dmabuf_modifiers (vk_present.c) takes a DRM fourcc and asks the compositor's own Turnip via VkDrmFormatModifierPropertiesListEXT, and drm_to_vk (vk_present.c:105-123) already maps AB30/XB30 → A2B10G10R10_UNORM_PACK32, AR30/XR30 → A2R10G10B10_UNORM_PACK32, and AB4H/XB4H → R16G16B16A16_SFLOAT. Adding 10-bit is a table edit plus the matching FOURCC defines next to compositor.c:415-421.

The compositor's own output swapchain is oblivious (vk_present.c:301-311, 437-466): two instance extensions (VK_KHR_surface, VK_KHR_android_surface - no VK_EXT_swapchain_colorspace), and VkSurfaceFormatKHR chosen = fmts[0]; (:441) - the first format the Android WSI happens to return, with whatever colour space rides along. The scene image and the whole 13-pass effects chain are 8-bit (vk_present.c:54 SCENE_FMT = VK_FORMAT_R8G8B8A8_UNORM; effects_chain.c:29 FX_FORMAT = VK_FORMAT_R8G8B8A8_UNORM).

Naming collision, flag it in any UI work. The drawer already has an "HDR" effect. It is effects_chain.c:197 P_HDR / app/src/main/cpp/winlator/hdr.frag - a dual-radius bloom + contrast lift ported from HDREffect.java, HDRPower = 1.30, entirely SDR and entirely fake. Real HDR must be named something else ("HDR output", "HDR10") or users will conflate them.

2.5 Is there a protocol-free shortcut?

No. A DRM_FORMAT_ABGR2101010 buffer is equally valid as SDR BT.709 or as PQ BT.2020; the format and modifier describe byte layout, not colorimetry. That ambiguity is precisely why wp_color_management_v1 exists. We could add a set_dataspace request to our own banner_ahb_v1 and skip the standard protocol - but Mesa would never call it, so that would need a driver patch (new wcp) to buy something Mesa already gives us for free. Implement wp_color_manager_v1 v1 server-side; it is strictly cheaper.


3. Android side

3.1 The real symbols, their API levels, and what the device has

frameworks/native/include/android/surface_control.h:

Symbol API level Present in /system/lib64/libandroid.so on the Pocket FIT (API 34)?
ASurfaceTransaction_setBufferDataSpace(tx, sc, ADataSpace) 29 PRESENT
ASurfaceTransaction_setHdrMetadata_smpte2086(tx, sc, AHdrMetadata_smpte2086*) 29 PRESENT
ASurfaceTransaction_setHdrMetadata_cta861_3(tx, sc, AHdrMetadata_cta861_3*) 29 PRESENT
ASurfaceTransaction_setExtendedRangeBrightness(tx, sc, float cur, float desired) 34 PRESENT
ASurfaceTransaction_setDesiredHdrHeadroom(tx, sc, float) 35 absent (expected - device is 34)
ASurfaceTransaction_setLuts 36 n/a
ASurfaceTransaction_setColorSpace does not exist -
ASurfaceTransaction_setLuma does not exist -

(Checked with strings on the device's own libandroid.so; setEnableBackPressure and setFrameRateWithChangeStrategy confirmed present alongside, matching what sc_layer.c already dlsyms.)

Our dlsym table is sc_layer.c:82-97 - 15 entries, none of them colorimetric. The additions needed are setBufferDataSpace (the only mandatory one), plus optionally setHdrMetadata_smpte2086 / _cta861_3 and setExtendedRangeBrightness. All four are API ≤ 34, i.e. inside the range sc_layer.c already targets, so the existing "missing symbol → path unavailable" pattern covers older devices unchanged.

3.2 Dataspaces we would emit

ADataSpace = STANDARD (63<<16) | TRANSFER (31<<22) | RANGE (7<<27), all in the NDK since 29:

Constant Value Maps from
ADATASPACE_BT2020_PQ 163971072 VK_COLOR_SPACE_HDR10_ST2084_EXT
ADATASPACE_BT2020_HLG 168165376 VK_COLOR_SPACE_HDR10_HLG_EXT
ADATASPACE_SCRGB_LINEAR 406913024 VK_COLOR_SPACE_EXTENDED_SRGB_LINEAR_EXT
ADATASPACE_DISPLAY_P3 143261696 VK_COLOR_SPACE_DISPLAY_P3_NONLINEAR_EXT
ADATASPACE_SRGB 142671872 VK_COLOR_SPACE_SRGB_NONLINEAR_KHR (today's implicit default)

Every layer on the device right now reports dataspace=UNKNOWN (0), which SF treats as sRGB.

3.3 Buffer formats

AHARDWAREBUFFER_FORMAT_R10G10B10A2_UNORM = 0x2b (API 26) ↔ VK_FORMAT_A2B10G10R10_UNORM_PACK32

  • the normal 10-bit scanout format on Adreno DPUs. QTI gralloc on this device knows it: strings /vendor/lib64/libqdMetaData.so lists RGBA_1010102, RGBX_1010102, BGRA_1010102, BGRX_1010102. AHARDWAREBUFFER_FORMAT_R16G16B16A16_FLOAT = 0x16 (API 26) exists but is a compositing intermediate, not normally a scanout format - it was not found in the vendor gralloc strings, so treat FP16 gralloc allocation on this SoC as unproven. AHardwareBuffer_isSupported() (API 29) is the correct runtime probe, and we should use it rather than assume.

The compositor's own AHB pool is hardcoded 8-bit (sc_layer.c:302 .format = AHARDWAREBUFFER_FORMAT_R8G8B8A8_UNORM) and the imported view is built with DRM_ABGR8888 (sc_layer.c:326). That is the copy path; the zero-copy path (sc_layer_present_ahb, sc_layer.c:430) takes the game's own buffer and so already carries whatever format the game allocated - it just never labels it.

3.4 What SurfaceFlinger does with an HDR layer on an SDR panel

From services/surfaceflinger/CompositionEngine/src/Output.cpp, getBestDataspace() / pickColorProfile():

  • A BT2020_PQ layer sets outHdrDataSpace = BT2020_PQ but the output only adopts it when DisplayColorProfile::hasLegacyHdrSupport(BT2020_PQ) is true - i.e. when the Composer HAL reported HDR types for that display. Otherwise the output dataspace stays DISPLAY_P3 with COLORIMETRIC intent, and the layer is tone-mapped, not rejected.
  • Because the HWC for an SDR-only display cannot consume a PQ buffer, the standard validateDisplay() negotiation marks that layer CLIENT instead of DEVICE, and RenderEngine/Skia composites it with the libtonemap SkSL gain shader (libtonemap_LookupTonemapGain, uniforms in_libtonemap_displayMaxLuminance / in_libtonemap_inputMaxLuminance).
  • This is the trap for us. Our whole zero-copy win is that banner_wayland_game composites DEVICE (proven 2026-09-13: DEVICE with ROT_90 + 1.5× scale at 144 Hz, −4 pts GPU busy, −3 % power). Tagging that layer BT2020_PQ on a panel with no HDR types is very likely to drop it to CLIENT - i.e. HDR would buy a washed-out tone-map and pay for it with the zero-copy saving. Any implementation must gate the dataspace on the display's reported HDR types, not on the game's request.
  • setExtendedRangeBrightness is a different mechanism: it applies only to RANGE_EXTENDED (scRGB) layers and works by having SF dim the other layers by whitePointNits / maxWhitePoint (renderengine/LayerSettings.h) while leaving the extended layer undimmed, so the boosted layer reads brighter by simultaneous contrast without exceeding the panel ceiling. It needs the display to expose an HDR/SDR ratio. This panel does not (§4).
  • setDesiredHdrHeadroom (API 35) is the PQ/HLG equivalent and overrides/is overridden by setExtendedRangeBrightness - last call wins. Not available on the test device at all.

3.5 has_HDR_display vs getHdrCapabilities()

ro.surface_flinger.has_HDR_display is a build-time declaration (TARGET_HAS_HDR_DISPLAY) that the device has a wide-gamut, high-luminance panel; it gates SF-internal colour-management machinery. Display.getHdrCapabilities().getSupportedHdrTypes() is the runtime, per-display answer, built from the Composer HAL's getHdrCapabilities() for that connector, itself derived from the EDID CTA-861.3 Static Metadata Data Block. They can disagree, and on this device they do. An empty list does not mean rejection - it means "tone-map it."

3.6 External display

On hotplug, the framework queries getDisplayConfigs, getColorModes, getHdrCapabilities, getDisplayCapabilities for the new display handle exactly as for the internal panel, so Display.getHdrCapabilities() on an external Display is architecturally wired to reflect the monitor's EDID. Whether it actually does is a vendor-HWC question; empty lists for genuinely HDR10 USB-C monitors are a commonly reported Snapdragon gap. Unproven for the Pocket FIT - the only way to know is to plug one in.


4. What the test device can and cannot show

All read today from the Pocket FIT (Adreno 750, Android 14 / API 34), read-only:

DisplayDeviceInfo{"Built-in Screen": 1080 x 1920, ...
  supportedModes [... supportedHdrTypes=[] (all four modes) ],
  colorMode 0, supportedColorModes [0],
  hdrCapabilities HdrCapabilities{mSupportedHdrTypes=[], mMaxLuminance=500.0,
                                  mMaxAverageLuminance=500.0, mMinLuminance=0.0},
  hdrSdrRatio NaN, brightnessMaximum 1.0 ...}
mBaseDisplayInfo: ... hdrSdrRatio not_available, userDisabledHdrTypes []
dumpsys SurfaceFlinger:
  Wide-Color information:
    Device supports wide color: 1
    Device uses color management: 1
    DisplayColorSetting: Managed
    Display 4630946852602928515 color modes:
        ColorMode::NATIVE (0)
        Current color mode: ColorMode::NATIVE (0)
  colorMode=NATIVE (0) renderIntent=COLORIMETRIC (0) dataspace=UNKNOWN (0) targetDataspace=UNKNOWN (0)
  displayBrightnessNits=-1.000000 sdrWhitePointNits=-1.000000 displayBrightness=nullopt
  (every layer) dataspace=UNKNOWN (0) hdr metadata types=0
getprop:
  [ro.surface_flinger.has_HDR_display]:        [true]
  [ro.surface_flinger.has_wide_color_display]: [true]
  [ro.surface_flinger.wcg_composition_dataspace]: [143261696]   = ADATASPACE_DISPLAY_P3
  [persist.sys.sf.color_mode]:                 [9]              = ColorMode::DISPLAY_P3
  [vendor.display.enable_hdr10_gpu_target]:    [1]

Reading:

  • The build declares an HDR, wide-colour device and even asks for DISPLAY_P3 as the system colour mode (persist.sys.sf.color_mode=9), but the HWC reports supportedColorModes=[0] (NATIVE only) and no HDR types, so SF runs the display in NATIVE with COLORIMETRIC intent and the P3 request is inert. The props are aspirational; the HAL is the truth.
  • hdrSdrRatio not_available / NaN and displayBrightnessNits = -1 mean Display.isHdrSdrRatioAvailable() is false → setExtendedRangeBrightness has nothing to act on here. The "brighten highlights on an SDR panel" trick is unavailable on this device.
  • supportedColorModes=[0] also means Display-P3 wide gamut is not available either - not just HDR. A P3-tagged layer would be colour-converted into the panel's native gamut, i.e. into whatever the panel natively is, with no way for us to know what that is.
  • 500 nits max luminance is below the ~540-nit bar AOSP's own has_HDR_display doc cites, and there is no HDR type to signal anyway.

Conclusion: visual HDR cannot be demonstrated on the Pocket FIT's internal panel, and neither can wide gamut, and neither can extended-range brightness. The only things provable on this device are (a) 10-bit buffers end to end (banding reduction, measurable), and (b) plumbing correctness by logs and dumpsys.

Vulkan-side, for completeness (device strings):

Library VK_EXT_swapchain_colorspace VK_EXT_hdr_metadata
/system/lib64/libvulkan.so (platform loader - implements the Android swapchain) PRESENT PRESENT
/vendor/lib64/hw/vulkan.adreno.so (Qualcomm blob) absent absent
/vendor/lib64/hw/vulkan.turnip.so (vendor-shipped Turnip) PRESENT PRESENT
wcp v6 libvulkan_freedreno_wayland.so (game driver) PRESENT PRESENT

The platform loader contains the string native_window_set_buffers_data_space(%d) failed: %s (%d)

  • i.e. Android's own swapchain implementation is what turns a VkColorSpaceKHR into a surface dataspace. So for the copy path, the compositor's output HDR is genuinely just "enable VK_EXT_swapchain_colorspace at instance creation and pick a better surface format than fmts[0]". No driver work.

5. The gap list, ordered, with sizes

Sizes are engineer-days for someone who already knows this code. "Prove" = device-visible in logs or dumpsys, not by eye. App-only unless marked.

Phase A - make the truth reachable (no user-visible HDR yet)

# Repo Work Size
A1 app Advertise 10-bit DRM formats over zwp_linux_dmabuf_v1. Add AB30/XB30 (and optionally AR30/XR30, AB4H/XB4H) to compositor.c:1323-1326 + FOURCC defines near :415-421. vkp_dmabuf_modifiers and drm_to_vk already handle them, so this is a table edit + a log line. Risk: a client picking a 10-bit format while the scene/effects chain stays 8-bit - the import works (drm_to_vk), the blit downconverts. Gate behind a flag until A4. 0.5 d
A2 app Generate + commit color-management-v1 protocol glue. Fetch staging/color-management/color-management-v1.xml from wayland-protocols ≥ 1.41, run wayland-scanner offline, commit XML + generated/color-management-v1-{protocol.c,server-protocol.h}, add to CMakeLists.txt. Mechanical. 0.5 d
A3 app Implement wp_color_manager_v1 v1 server-side (new wl_color_mgmt.c, ~400–500 lines): the global + bind-time feature/intent/primaries/tf events; the parametric creator with set_primaries_named/set_tf_named/set_max_cll/set_max_fall/set_mastering_*; image-description objects with ready/failed; get_surface → wp_color_management_surface_v1 and set_image_description stored double-buffered on our struct surface. Watch the blocking loop: Mesa dispatches its queue waiting for ready, so we must send ready from the request handler's own flush, not defer it to the next frame, or a game stalls at swapchain creation. Advertise only what the current output can actually honour (see A5). 2 d
A4 app Carry the image description onto the Android layer. dlsym ASurfaceTransaction_setBufferDataSpace (+ setHdrMetadata_smpte2086 / _cta861_3) in sc_layer.c:82-97; map surface image description → ADataSpace + AHdrMetadata_*; apply in sc_layer_present_ahb (sc_layer.c:430-470, next to the existing setBuffer). Plumb the per-surface description from compositor.c through ahb_swapchain.c:219. Also set an explicit ADATASPACE_SRGB on the existing 8-bit path so we stop relying on UNKNOWN. 1 d
A5 app Capability gate. Read Display.getHdrCapabilities() / getSupportedHdrTypes() / isHdrSdrRatioAvailable() (and the external Display when ExternalDisplayController has one) in Java, pass down over JNI, and only then advertise bt2020/st2084_pq from A3. Without this we advertise HDR on a panel that will tone-map it and drop the layer to CLIENT (§3.4). Include a banner_log("color", ...) line naming exactly what was advertised and why. 0.5 d

Phase A total: ~4.5 days. App-only. No new wcp. End state: a game with DXVK_HDR=1 on a genuinely HDR display gets an HDR10 swapchain, its buffer reaches an ASurfaceControl layer tagged BT2020_PQ with real mastering metadata, and dumpsys SurfaceFlinger proves it.

Phase B - the copy path and the effects chain (only if A is kept)

# Repo Work Size
B1 app 10-bit output swapchain. Add VK_EXT_swapchain_colorspace to vk_present.c:301-311 (enabledExtensionCount 2 → 3), replace chosen = fmts[0] (:441) with a real preference walk (HDR10 + A2B10G10R10 when the display supports it, else R8G8B8A8_UNORM + sRGB). The Android platform loader does the rest. 0.5 d
B2 app Widen the scene + effects chain. SCENE_FMT (vk_present.c:54) and FX_FORMAT (effects_chain.c:29) → A2B10G10R10_UNORM_PACK32 (or FP16 for real HDR headroom). Every one of the 13 passes samples/writes it; the debanding pass in particular must be re-tuned because its strength is calibrated for 8-bit quantisation. Memory and bandwidth go up ~25 % (or 100 % for FP16). This is the expensive item and it regresses SDR perf if done unconditionally. 2 d
B3 app Layer pool at 10-bit. sc_layer.c:302 pool format → R10G10B10A2_UNORM when the chain is 10-bit, guarded by AHardwareBuffer_isSupported(), with the existing sniff_modifier/linear-fallback retry extended. 0.5 d

Phase B total: ~3 days. App-only. No new wcp.

Phase C - scRGB, and the one thing that needs a wcp

# Repo Work Size
C1 driver (new wcp) Map VK_FORMAT_R16G16B16A16_SFLOAT → AHARDWAREBUFFER_FORMAT_R16G16B16A16_FLOAT (0x16) in banner_ahb_format_for (patches/wayland/banner_ahb_wsi.py:225-237) so FP16 swapchains are not hidden by banner_ahb_skip_format when zero-copy is on. One enum + a rebuild of eight drivers. This is the only wcp-requiring item, and only for scRGB zero-copy. Blocked on proving FP16 gralloc allocation works on Adreno at all (§3.3). 0.5 d + a driver build cycle
C2 app Advertise extended_target_volume + ext_linear + AB4H/XB4H, emit ADATASPACE_SCRGB_LINEAR, and wire ASurfaceTransaction_setExtendedRangeBrightness from Display.getHdrSdrRatio(). 1 d
C3 proton-wine (new wcp) Optional, quality only. Synthesise an EDID with a CTA-861.3 HDR static metadata block in winewayland.drv so DXVK reports the real panel luminance in DXGI_OUTPUT_DESC1 instead of its 1499-nit dummy. 1–2 d

6. Cheap partial wins - and which are visible on a 500-nit panel

Idea Verdict on the Pocket FIT Notes
10-bit output (banding) Visible, and the best value here. This is the one real win. B1 alone (0.5 d) gets a 10-bit output swapchain; A1 (0.5 d) lets the game render 10-bit. Adreno 750 + the DPU handle RGBA_1010102 natively, and vendor.display.enable_hdr10_gpu_target=1 says the GPU target path expects it. Pairs with the existing debanding pass - arguably makes it unnecessary for 10-bit-capable games, which saves a full-screen pass. Visible on gradients (skyboxes, fog, dark interiors) where our deband pass currently earns its keep.
Display-P3 wide gamut on SDR Not visible. Do not build. supportedColorModes=[0] / ColorMode::NATIVE - the HWC exposes no P3 mode. persist.sys.sf.color_mode=9 asks for P3 and is ignored. Tagging a layer ADATASPACE_DISPLAY_P3 would make SF convert P3 → the panel's native gamut, which on a NATIVE-mode display is an unknown-to-us transform. Net effect on this device: at best nothing, at worst a colour shift.
setExtendedRangeBrightness to lift highlights Not available. hdrSdrRatio not_available / NaN, displayBrightnessNits=-1, sdrWhitePointNits=-1. Display.isHdrSdrRatioAvailable() is false, so there is no headroom to claim and SF has no nits data to dim the other layers against.
setDesiredHdrHeadroom Not available. API 35; device is API 34. Absent from libandroid.so.
Set ADATASPACE_SRGB explicitly on our layers Invisible but worth doing anyway (part of A4, ~0 extra cost). Every layer currently reports dataspace=UNKNOWN (0). It works today because SF treats UNKNOWN as sRGB, but it is an accident we are relying on, and it will become wrong the moment any colour management is enabled.

So: of the three "cheap wins" in the brief, exactly one - 10-bit - is real on this hardware, and it is worth about 1 engineer-day (A1 + B1) on its own, independent of HDR.


7. How we would ever prove it

7.1 Which games actually emit HDR under DXVK / VKD3D

Requirements: a flip-model swapchain at R10G10B10A2_UNORM or R16G16B16A16_FLOAT, an IDXGISwapChain3::SetColorSpace1 call, and DXVK_HDR=1 in the environment. Known-good candidates, easiest first:

  • D3D11 + scRGB: Shadow of the Tomb Raider, Resident Evil 2/3 Remake, Devil May Cry 5, Mass Effect Legendary Edition. RE Engine titles are the usual smoke test on Linux because they call SetColorSpace1 early and unconditionally when DXGI reports HDR.
  • D3D12 + HDR10: Cyberpunk 2077, Forza Horizon 5, Horizon Zero Dawn. These go through vkd3d-proton's stricter path (§1.1) - good for catching a half-done implementation, since a wrong answer stalls presentation instead of degrading.
  • Sanity harness, no game needed: anything that calls vkGetPhysicalDeviceSurfaceFormatsKHR under the layer. vulkaninfo is not available in the prefix, but a tiny Win32 Vulkan probe built against the layer's headers, or simply DXVK's own VK_INSTANCE_LAYERS-free log with DXVK_LOG_LEVEL=info and DXVK_HDR=1, will say whether the colour space was accepted. Start here - it is the cheapest signal and needs no HDR display at all.

7.2 Platform-side verification (not by eye)

Order of proof, each step gating the next:

  1. Compositor log. New banner_log("color", ...) lines: which primaries/tf we advertised and why; every image description created (primaries, tf, max_cll, max_fall); every dataspace set on the layer. This is the only step that works on any device.
  2. Guest log. DXVK_HDR=1 + DXVK_LOG_LEVEL=info → DXVK prints the chosen colour space and format at swapchain creation. If SetColorSpace1 was refused, DXVK says so. Mesa's WSI prints Not using HDR metadata to avoid protocol errors when is_hdr_metadata_legal rejects metadata
    • catching that string is a direct test of A3's metadata plumbing.
  3. dumpsys SurfaceFlinger | grep banner_wayland_game. Today it prints dataspace=UNKNOWN (0) hdr metadata types=0. Success = dataspace:0x9c60000 (BT2020_PQ, 163971072 = STANDARD_BT2020|TRANSFER_ST2084|RANGE_FULL) and hdr metadata types= non-zero. Also record composition: on the same line
    • if it flips DEVICE → CLIENT, HDR cost us the zero-copy win (§3.4) and that is a failure, not a success.
  4. dumpsys display. On whatever display is in use: supportedHdrTypes=[...] non-empty, colorMode ≠ 0, hdrSdrRatio ≠ not_available. This is the gate for steps 5–6; on the Pocket FIT's internal panel it will never pass.
  5. dumpsys SurfaceFlinger Wide-Color section. Current color mode should move off ColorMode::NATIVE (0), and the output line's dataspace=/targetDataspace= should stop being UNKNOWN (0).
  6. Only then, by eye.

7.3 Getting a display that can actually show it - three options, ranked

  1. External HDR display over USB-C (recommended, cheapest). The app already reparents the game into a Presentation on a secondary display - app/src/main/java/com/winlator/star/display/ExternalDisplayController.java, findPresentationDisplay() picks a DISPLAY_CATEGORY_PRESENTATION display that is not the activity's own. In the Wayland session the compositor's SurfaceView is the reparented view, so sc_layer.c's ASurfaceControl_createFromWindow(vkp_window(), ...) follows it automatically - no new plumbing to test on an external panel. Risk: Display.getHdrCapabilities() for a DP-alt-mode display depends on the vendor HWC reading the external EDID, and empty lists for genuinely HDR10 monitors are a common Snapdragon gap (§3.6). Step zero is a five-minute, zero-code test: plug an HDR monitor in and run dumpsys display | grep -A2 hdrCapabilities. If the external display reports HDR types, everything else follows. If it does not, options 2 and 3 are all that remain.
  2. The user's Galaxy Fold (Adreno 840). Samsung foldables/flagships do report HDR10+ on the internal panel, so this is the most likely device to show real HDR with no extra hardware. But the Fold is already the outstanding unknown for basic Wayland bring-up (checkpoint item 1: "Other GPUs… the Galaxy Fold (A840) logs are still outstanding"). Chasing HDR there before plain Wayland is proven there inverts the priority order.
  3. Plumbing-only verification on SDR. Everything in §7.2 steps 1–3 works on the Pocket FIT. This proves the code is correct but proves nothing about whether it looks right - and tone-mapping quality is the entire user-visible payload of HDR. Do not ship on this evidence alone.

8. Verdict

Not now. Do the 10-bit slice; park HDR behind the Fold.

The reasoning, plainly:

  • The effort is genuinely small and smaller than we thought. ~4.5 days to a plumbed HDR10 path, app-only, no new wcp - because Mesa's colour-management client is already inside the shipped wcp v6 driver and our zero-copy WSI patch already maps 10-bit to gralloc. The phase-4 memory note saying HDR "does need a new wcp" should be corrected: only scRGB/FP16 zero-copy (C1) does, and that is optional.
  • But the payload is zero on the only device we can test on. The Pocket FIT reports no HDR types, no colour modes but NATIVE, no HDR/SDR ratio and no brightness nits. We could not see the feature, could not tune the tone-mapping, and could not tell a correct implementation from a subtly wrong one. Shipping HDR proven only by log lines would be the exact opposite of this project's "device-proven, not CI-green" rule.
  • On this device it would be a net negative if enabled. §3.4: an HDR-dataspace layer on a display with no HDR types drops from DEVICE to CLIENT composition. That trades the measured zero-copy win (−4 pts GPU busy, −3 % power) for a GPU tone-map pass and a washed-out image. A capability gate (A5) prevents the harm, but a gate that is always closed on our test hardware is a feature nobody can exercise.
  • It is also out of order. The pre-4 checkpoint's ranked list puts "other GPUs" and "prove zero-copy where it can win, then make it the default" first, with multi-layer presentation in flight. HDR competes with those for the same sc_layer.c surface area - and multi-layer will rewrite exactly the code A4 touches. Landing HDR into sc_layer.c now means merging it twice.

What the user would actually see for the effort, honestly:

  • Today, Pocket FIT, full HDR (7.5 days): nothing. Identical picture, plus a dumpsys line that says BT2020_PQ, plus a risk of losing zero-copy.
  • Today, Pocket FIT, the 10-bit slice only (A1 + B1, ~1 day): slightly smoother gradients in dark scenes and skyboxes for games that offer a 10-bit swapchain, and a plausible route to retiring the debanding pass (one fewer fullscreen pass) for those games. Small, real, and it is the foundation Phase A sits on anyway - so it is not wasted work if HDR later goes ahead.
  • Later, on an HDR display (external monitor or the Fold), full Phase A: genuinely brighter highlights and a wider gamut in the ~15 PC titles that implement HDR properly, on the one Android emulator that can do it. That is a real headline feature and a real differentiator - but only once there is a panel to show it on.

Recommended sequence:

  1. Five minutes, zero code: plug an HDR monitor into the Pocket FIT's USB-C and run dumpsys display | grep -A2 hdrCapabilities. This single reading decides everything - it is the difference between "we can develop and prove HDR today on hardware we own" and "HDR is blocked on the Fold."
  2. ~1 day, independent of HDR: the 10-bit slice (A1 + B1), gated and A/B'd for banding and perf. Ship it as a banding/quality improvement on its own merits.
  3. Park Phase A (A2–A5) until either step 1 comes back positive or the Fold is in the loop and plain Wayland is proven there. Revisit after multi-layer presentation lands, so the sc_layer.c dataspace work is written once against the final layer API.

9. What I could not establish, and why

  • Whether FP16 (R16G16B16A16_FLOAT) AHardwareBuffers are allocatable on this SoC. Not in the vendor gralloc strings, and I did not install or run a probe binary (other agents are testing on the device). Decides whether scRGB zero-copy (C1) is possible at all. Cheap to settle later with a one-call AHardwareBuffer_isSupported() check inside the compositor's existing startup log.
  • Whether an external HDR display over USB-C actually reports HDR types on the Pocket FIT. No external display was connected and I did not connect one. This is the single highest-value unknown in the whole report (§8, step 1).
  • Whether an HDR-tagged layer really drops DEVICE → CLIENT on this HWC. The AOSP code path is clear (§3.4) but it is vendor-HWC behaviour and I did not test it - testing it would mean launching a game, which was out of scope.
  • What the Galaxy Fold (Adreno 840, Android version unknown to me) reports. Not connected. Its getHdrCapabilities() and isHdrSdrRatioAvailable() are the other decisive readings.
  • Exact Mesa line numbers. Cited against wsi_common_wayland.c fetched from gitlab.freedesktop.org/mesa/mesa at 7cda7850edd103ace21aac37d416d2fdf7a282e1, the commit in banners-turnip-wayland/mesa_hash.txt. They are accurate for that file; they will drift if the Turnip base moves.
  • DXVK/vkd3d-proton line numbers are cited against upstream doitsujin/dxvk@8759acd1 and HansKristian-Work/vkd3d-proton@2a38e799 (master as of today), not against the exact v3.1-gplasync / 3.1.0-wave64-relax builds we ship. The feature presence in our builds was verified independently by strings on the device (§1.1); the line numbers are for orientation.
  • Whether any of the games in §7.1 launch and run correctly under our Wayland backend at all. Untested. Picking the HDR test title is its own small exercise and should start from games already known to work in a Wayland container.

10. Round 1 - as built (2026-09-14, feat/wayland-hdr)

Phase A (A1–A5) for HDR10, opt-in, proven by the app's own logs because the first HDR panel in the loop is a tester's non-rooted Galaxy Fold (Adreno 840, Android API 37: HDR10/HLG/HDR10+, 1351 nits, HDR/SDR ratio available). App-only; no layer change.

10.1 The gate (wl_color_mgmt.c, banner_color_init)

Open only when ALL hold when the compositor starts:

Input Where it comes from
BANNER_WAYLAND_HDR=1 (or true/on) container or shortcut env (XServerDisplayActivity.resolvedWaylandHdrMode)
the game's display lists HDR10 DisplayHdrInfo.supportsHdr10 for hdrTargetDisplay() (the TV when the game is on it)
display layers with dataspace control sc_layer_can_tag_hdr() = SurfaceControl API + ASurfaceTransaction_setBufferDataSpace
the zero-copy global ahb_swapchain_advertised()

BANNER_WAYLAND_HDR=force skips the display row only (testing the negotiation on an SDR panel - SurfaceFlinger then tone-maps the layer). A closed gate advertises nothing (no wp_color_manager_v1 global, no 10-bit dma-buf rows), writes color … HDR gate CLOSED: <why> and HDR on screen: no, because <why>, and the session is otherwise byte-for-byte the old one. With the switch unset the module is silent unless the display lists HDR10 or DXVK_HDR=1 is set (one hint line). On an HDR10 display the app also turns zero-copy presentation on for the session (an HDR frame is only right on the game's own layer) and exports BANNER_WSI_AHB=1 to match.

10.2 What an open gate offers (§2.2, exactly what Mesa binds)

wp_color_manager_v1 version 1: intent perceptual; features parametric + set_mastering_display_primaries (Mesa only sends mastering luminance inside that feature's branch); primaries bt2020; transfer st2084_pq. Not advertised: srgb (so Mesa creates no colour surface for SDR swapchains at all), HLG (no DXGI swapchain uses it), ext_linear/extended_target_volume (scRGB = Phase C, needs FP16 gralloc + a layer change), ICC, windows_scrgb. zwp_linux_dmabuf_v1 gains AB30/XB30 (A2B10G10R10 - Mesa lists a VkFormat only when both the alpha and opaque fourcc are there), appended after the four 8-bit rows. Protocol glue generated from wayland-protocols 1.41 with wayland-scanner 1.24.0 (protocols/color-management-v1.xml).

Strictness: protocol errors only where a conforming client can never hit them (un-advertised feature/intent/primaries/tf, a property set twice, an incomplete set, a not-ready description). Everything a conforming client can produce on a bad day is logged and absorbed instead - out-of-range HDR metadata is dropped from what Android gets (and max_cll/max_fall of 0 = "unknown", which Mesa sends as-is), a second colour surface for the same wl_surface replaces the first, requests on inert objects are ignored. ready is sent from inside create (Mesa blocks its swapchain on it). Output and surface-feedback image descriptions answer failed (Mesa uses neither).

10.3 The layer path (sc_layer.c)

setBufferDataSpace (+ setHdrMetadata_smpte2086 / _cta861_3 when present) ride the same transaction as the buffer, per layer, only on change, re-sent on every new SurfaceControl (recovery swap, window change). A layer that was never tagged is never touched; once tagged, an untagged frame puts it back to UNKNOWN and clears the metadata. The zero-copy frame gets the surface's current description; the 8-bit pool copy of an HDR frame keeps the PQ tag (colours right, 8-bit precision, logged once); the effects pass and the overlay layer are always untagged.

10.4 When an HDR frame cannot be zero-copy - round 1's decision

No tone-mapping anywhere (the copy path is vkCmdBlitImage into an 8-bit sRGB swapchain; a PQ→SDR shader pass is its own project). Instead:

  • Effects on / frame generation on: an HDR fullscreen game keeps its display layer; the effects chain and frame generation are skipped for it (both are 8-bit SDR and would wash it out), and the base surface's black frame skips them too (vkp_render_plain) so generated presents cannot pace the loop. Logged when it starts and when it ends.
  • A window above the game on a display that cannot compose a second layer (phase 5's rotated+scaled rule is untouched), a windowed HDR game, zero-copy switched off, layer unavailable: the scene goes through the copy path and the HDR frames are shown untone-mapped (washed out); every such scene is counted and the reason logged.

10.5 Proof without root - what a tester's logs say

Download/Wayland-logs/wayland-*.log, tag color (plus the existing display capability line):

Line Means
HDR gate OPEN: … "Built-in Screen" … reports HDR types HDR10, … the offer is on, with its inputs and which Android colour calls exist
<program> bound wp_color_manager_v1 version 1 … Mesa saw it: VK_COLOR_SPACE_HDR10_ST2084_EXT is now listed
<window>: colour-management surface created a swapchain asked for a non-sRGB colour space
image description #N from <program>: BT.2020, ST 2084 (PQ); mastering …; max CLL …, max FALL … the game's colour space + vkSetHdrMetadataEXT values
<window> presents WxH buffers in XB30 (10-bit A2B10G10R10) …, gralloc RGBA1010102 (10-bit) the 10-bit swapchain really arrived
banner_wayland_game: dataspace BT2020_PQ (0x9c60000) set on the display layer … the tag + metadata SurfaceFlinger received
HDR last 10 s: N frames on the display layer tagged BT2020_PQ … steady state
display HDR/SDR ratio X (was Y; HDR frames on screen: yes …) Display.getHdrSdrRatio() - rises above 1.00 only when Android grants HDR headroom
HDR on screen: yes / tagged but NOT confirmed / no, because … the verdict (written on changes, at game exit and at session end; the last one counts)

DXVK's own <exe>_dxgi.log / <exe>_d3d11.log add Color space: VK_COLOR_SPACE_HDR10_ST2084_EXT and the 10-bit format at swapchain creation; Mesa's Not using HDR metadata to avoid protocol errors (stderr → wine_debug.log) means the game's metadata failed Mesa's own legality check.

11. Round 2 - as built (2026-09-14, feat/wayland-hdr; CI only, not device-proven yet)

Round 1 is proven on the tester's Fold (God of War, DXVK v3.1: 10-bit zero-copy, BT2020_PQ on the layer, HDR/SDR ratio 1.00 → 2.51, HDR on screen: yes). Round 2 makes it a setting and removes round 1's big limitation (§10.4: no tone-mapping, effects/frame generation skipped, washed-out copies). The gate itself is unchanged (§10.1) apart from its first input.

11.1 The setting, and what the launch does with it

  • Container.isWaylandHdr() (extraData waylandHdr) and the shortcut extra waylandHdr (1/0/unset = the container's) - one resolver, display/WaylandHdr.effective(), used by the three editors (container, game shortcut, XMB game settings) and the launch path. BANNER_WAYLAND_HDR in the environment still overrides it (1 / 0 / force). Editors grey the row with the reason when the built-in screen does not list HDR10 (WaylandHdr.unavailableReason), and say it applies from the next launch.
  • HDR on for the session (setting resolved on AND HDR10 display, or force): DXVK_HDR=1 is exported after the user's env merge unless the user set DXVK_HDR themselves.
  • Every Wayland session whose display lists HDR10 exports BANNER_WAYLAND_HDR_MAX_NITS, …_MAX_AVG_NITS, …_MIN_NITS (decimal nits from Display.getHdrCapabilities(); unknown values and a max of 0 are left out) - the contract the Wayland layer from versionCode 10 turns into EDID HDR metadata for DXGI (checked by the Wine side against DXVK's own EDID parser: the Fold's 1351/1351/0 → DXGI max 1345.43, max full-frame 1345.43, min 0.01). Never on an SDR display, whose EDID would then claim PQ support. One session environment: line logs them. Android has no live absolute nits; the live evidence stays the HDR/SDR ratio.
  • A "DXVK before 3.0 needs layer v10" warning was built and then removed: its premise was wrong (Wine already exposes VK_EXT_swapchain_colorspace at instance level; DXVK ≥ 2.1 looks for it in the device list, logs 0 on every Mesa system - the working v3.1 run logs 0 too - and never uses it for the HDR decision). Why God of War offered no HDR option on DXVK 2.4.1 is open (Fold A/B).

11.2 HDR-aware composition (hdr_compose.c, hdr_encode.frag, compositor.c hdr_plan)

Every scene with an HDR draw takes one of four routes (logged on each change, with the reason):

Route When What reaches the display
0 the HDR game alone fullscreen, zero-copy on, no effects, nothing above it (or one window the overlay layer can carry) the game's own 10-bit gralloc frames on its layer, BT2020_PQ (round 1)
1 effects on, a window above on a display that cannot take a second layer, a windowed game, zero-copy off the HDR picture: every draw blitted 1:1 into a 10-bit "mixed" image (each pixel still in its own encoding), ONE encode pass to PQ BT.2020 (SDR pixels sRGB → linear → BT.2020, placed at 203 nits, BANNER_WAYLAND_HDR_SDR_NITS overrides), the effects chain run on it in 10-bit (vkp_effects_set_formats), copied into a 10-bit (RGBA1010102) layer buffer, tagged with the game's description; 8-bit buffers if gralloc refuses 10-bit (logged)
2 frame generation on composed into PQ in the 8-bit scene image, through the frame-generation engine, presented through an HDR10 swapchain (VK_EXT_swapchain_colorspace enabled on the compositor instance in HDR sessions only; first A2B10G10R10 else FP16 pair the surface lists with HDR10_ST2084); where the surface lists none, tone-mapped into the ordinary swapchain. 8 bits through the engine: gradients may band
3 an HDR frame the compositor could not import layer only; effects/frame generation skipped for it (round 1's behaviour, now the rare case)

The tone map (mode 1 of hdr_encode.frag): PQ → nits → BT.2020 → BT.709, relative to SDR white, maxRGB curve linear to a 0.8 knee then an exponential roll-off that puts the content peak (the game's max CLL, else mastering max, else 1000 nits) at 98 %, sRGB-encoded. Which pixels are HDR comes from up to six top-first scene rects + a mask in the push constants (more windows over the game → the lowest are treated as SDR, logged once). A composed frame that cannot reach the layer falls back to route 2 for as long as that lasts (logged once each way).

11.3 The live switch (lead's addition): drawer → Graphics → "HDR output"

Shown only while the gate is open (the activity's 1 s HDR sampler flips waylandHdrAvailable). Per session, starts on, nothing saved. Off (banner_color_set_output(0), posted through the host queue to the compositor thread, logged, redraw): route 0 becomes route 1 with the tone map, the picture goes on the layer untagged on 8-bit buffers (the layer's dataspace returns to UNKNOWN); route 2 rebuilds the swapchain as SDR and tone-maps; route 3 cannot (nothing can read those frames) and stays HDR, said in the log. The game is told nothing (its DXVK_HDR and colour-manager offer were decided at launch). The HUD reads Wayland · HDR while HDR frames are on screen (tagged in the last 1.5 s and ratio > 1.01 where reported), Wayland · HDR off while the switch is off. Frames tone-mapped by the switch are counted separately in the verdict.

11.4 Live headroom (lead's addition after the second Fold run)

The Fold's second run (brightness slider at maximum) showed the ratio at 3.23 for a minute and then 1.00 for 3+ minutes with HDR frames on screen the whole time, while the verdict kept saying "rose to 3.23". Android simply gave the layer no headroom (SDR white already at the panel's limit). Now, from the ratio samples the app already sends (1 s + the display listener): the time with HDR frames on screen and the part of it with ratio > 1.01; a no-headroom streak (HDR frames on screen, ratio <= 1.01) that after 5 s writes no HDR headroom for 5 s … the screen brightness is probably at maximum … (and HDR headroom is back … when it ends), tags the 10 s line and the steady ratio line, puts the HUD on Wayland · HDR (no headroom) and the drawer row on the same hint (banner_color_hdr_state = 2); and the verdict reports headroom above 1.00 for P% of the HDR time (a of b s), now R plus the hint while the streak lasts. Cheap: no new sampling, a few integers under the existing lock.

Round-2 Fold run (20:23, run 34911468350): the first no-headroom streak was a screen recording, not the brightness: the user's recording (210.8 s, BT.709 SDR H.264, finalised 20:27:30, so started ~20:23:59) lines up with the ratio falling 2.98 → 1.00 at 20:24:00.9 and staying there to the end. So Android (at least Samsung's) turns HDR headroom off while the screen is recorded. Every hint now names both causes; whether maximum brightness does it too is still unmeasured (tester note step H).

11.5 The HUD's HDR line (round 2b, the user's layout request)

40.3ms · Wayland · HDR (no headroom) ran off the pill's right edge (the capsule's ends curve in, and the latency line is the bottom of the stack). The display-server label is plain Wayland again, and the HDR state is a line of its own directly under it, fed as a FusionHdr code (FusionHudView.setHdrState): HDR / HDR (no headroom) / HDR off / HDR tone-mapped (switch on, frame generation without an HDR10 swapchain) / HDR ready (gate open, no HDR frames right now). NONE - every session whose gate is closed - draws nothing, so those HUDs are unchanged to the pixel. Pill: the extra line makes the capsule taller and its ends rounder, so in HDR sessions fitCapsule() measures every glyph's ink against the two rounded ends (plus the outline and 2 sp) and shifts/widens the pill until nothing crosses. Full: an HDR row under DISP; Tiles: the Display tile's sub-line; Minimal: a line under the footer; Mega: a fragment after DISP in the bottom band. Only the Fusion HUD shows the display server at all; the classic, GameHub and GameNative HUDs never received it and are untouched.

11.6 Round 2d (Fold evidence from 2c)

  • The format scan was a false negative. FG with the AIO HDR card logged lists no HDR10 swapchain format (37/0 37/1000104001 … 37/1000104012): swap_init read the surface's list into VkSurfaceFormatKHR fmts[32], and with VK_EXT_swapchain_colorspace the Android WSI lists every format once per colour space (11 on the Fold) - the 10-bit/FP16 rows were past entry 32, and the log's 160-char buffer cut the list too. Now: all pairs read (heap), one line with the total, colour spaces per format and every HDR-capable pair (10-bit/FP16 first); HDR10 pick A2B10G10R10 → A2R10G10B10 → FP16 → 8-bit; a driver that refuses the HDR10 swapchain gets the SDR one at once. The only other fixed array near it (queue families, 16) does not hold surface formats or present modes; FIFO is the only present mode used, so no list is read.
  • VK_EXT_hdr_metadata enabled in HDR sessions where the device lists it; the HDR10 swapchain gets the game's SMPTE 2086 / CTA-861.3 once per swapchain and image description (none if the game sent none).
  • FP16 frame generation for HDR. The bridge and both engines already take a format (the chain, the ring); lsfg-vk itself runs HDR in R16G16B16A16_SFLOAT. With HDR frames through an HDR10 swapchain the scene (g_fgscene), the encode pass (new FP16 output), the effects and the engine run in FP16 where fge_format_ok (storage + linear sampling) passes; the PQ signal is fed as-is (the shaders' own HDR flags stay off - PQ is already perceptual). A chain that fails to build in FP16 is sticky inside both engines, so the bridge refuses the format for the session and restarts the engine in 8 bits instead of marking frame generation failed.
  • Evidence beside the headroom (app, non-root): PowerManager thermal status (+ listener) and getThermalHeadroom(10) every 10 s, Settings.System brightness + mode (+ observer), screenshots (registerScreenCaptureCallback, API 34) and screen recording (addScreenRecordingCallback, API 35, reflection) with the normal DETECT_SCREEN_CAPTURE / DETECT_SCREEN_RECORDING permissions. Changes are logged; the no-headroom lines, the 10 s line and the verdict carry the values and name the likely cause (recording / hot / brightness at maximum) or say none is visible. The verdict lists only the paths that carried frames, to leave room for it.

11.7 Round 2e (headroom request, display ceiling, coverage)

2d on the Fold: all pass (68 surface pairs, HDR10 swapchain A2B10G10R10, FP16 FG at 120 fps, 0 tone-mapped, VK_EXT_hdr_metadata, headroom 3.00 at brightness 255 auto). A tester's ROG Phone 9 Pro (Adreno 830, API 36): correct HDR frames, composed path, HDR layer ~80% of the 2400x1080 screen (windowed card), not hot, not recording, brightness 8-223 manual - and the HDR/SDR ratio at 1.00 the whole session.

  • ASurfaceTransaction_setDesiredHdrHeadroom (API 35) on the game layer, per HDR frame, only on change: content peak (max CLL, else mastering max, else the display's peak) / SDR white, capped at the display's highest ratio; 0 (no preference) when the layer stops carrying HDR; re-sent on a fresh SurfaceControl. The screen surface (HDR10 swapchain for frame generation) gets the same through SurfaceView.setDesiredHdrHeadroom (API 35, reflection; SurfaceControl.Transaction fallback), applied by the app from banner_color_screen_headroom().
  • Display.getHighestHdrSdrRatio() (Android 16+, reflection) in the display line and the verdict; 1.00 means no boost at all, and the request is then not made.
  • HDR layer coverage in the dataspace line, the composed-picture line and on geometry changes.
  • "brightness at maximum" only for MANUAL >= 250/255; no visible cause -> "this phone may not boost HDR from apps (the requested headroom was …, the display's highest ratio is …)".

11.8 What is still NOT done / not proven

  • Round 2's composition, switch and HUD line were device-tested in parts on the Fold (setting-driven gate, auto environment, composed picture, headroom lines, HUD); 2d is CI only.
  • The engines' own HDR flags (lsfg hdr_support / FSR3 HDR input) stay off: the PQ signal is treated as perceptual SDR-like data. If interpolation artefacts show only in HDR, that is the next thing to try.
  • Colour effects (brightness/contrast/saturation/gamma, the "HDR" bloom) operate on the PQ signal in route 1 and look stronger than in SDR; sharpening/AA/CRT/upscalers are perceptually fine.
  • scRGB (FP16) swapchains (Phase C) - unchanged: needs FP16 gralloc and a layer change.