ffmpeg-v4l2-request-fourier: route H.264 chroma DC Hadamard through daedalus-fourier (0011)

Substitutes H264DSPContext.chroma_dc_dequant_idct in the 4:2:0 / bit_depth=8 init path with a wrapper that composes the daedalus chroma DC Hadamard primitive (daedalus-fourier PR #25) with the qmul scaling FFmpeg's reference does in one fused function (h264idct_template.c::ff_h264_chroma_dc_dequant_idct). Algorithm per H.264 §8.5.11.1 / §8.5.11.2: 1. Extract 4 DCs from the scattered positions in the per-MB coefficient buffer (stride=32, xStride=16) 2. 2x2 Hadamard transform (daedalus primitive) 3. qmul scale + >> 7, write back to original positions Bit-exact against ff_h264_chroma_dc_dequant_idct_8_c. The Hadamard itself is gated by the fourier PR #23 7-case test suite (including the H·H = 4·I algebraic invariant), and the public-API parity test added in PR #25 confirms the src/ symbol matches the test ref. 4:2:2 chroma stays on the in-tree ff_h264_chroma422_dc_dequant_idct_c path — same chroma_format_idc<=1 gating shape as 0009 chroma deblock. Pin bump: _daedalus_fourier_commit / DAEDALUS_FOURIER_COMMIT bumped to b9f9ff2a (post-PR #25) so the build picks up the public daedalus_h264_chroma_dc_hadamard_2x2 symbol. Verified the patch applies cleanly on top of 0001-0010 against the pinned upstream commit b57fbbe5 on hertz.
Merge pull request 'ffmpeg-v4l2-request-fourier: route H.264 luma intra deblock through daedalus-fourier (0010)' (#99 ) from claude-noether/marfrit-packages:noether/h264-substitute-deblock-intra into main
2026-05-25 13:39:54 +02:00 · 2026-05-25 11:28:33 +00:00 · 2026-05-25 13:21:00 +02:00 · 2026-05-25 11:18:15 +00:00 · 2026-05-25 13:16:45 +02:00 · 2026-05-25 11:14:26 +00:00
25 changed files with 1734 additions and 15 deletions
@@ -1556,3 +1556,179 @@ jobs:
        if: always()
        run: rm -f /root/repo_pass /root/.ssh/id_ed25519

+  # -------------------------------------------------------------------------
+  # aish (arch=any) — pure LuaJIT, one .pkg.tar valid on every pacman target.
+  # Same dual-arch publish pattern as lmcp / claude-his.
+  # -------------------------------------------------------------------------
+  aish-any:
+    needs: lmcp-debian   # parallel with claude-his-any (pure-Lua sibling),
+                         # serialized via the shared arch-aarch64 runner.
+                         # Avoids needless wait through the fourier stack.
+    runs-on: arch-aarch64
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: skip if already published
+        id: skip-check
+        run: |
+          set -e
+          result=$(./.gitea/scripts/check-already-published.sh arch/aish)
+          echo "$result" >> "$GITHUB_OUTPUT"
+          echo "decision: $result"
+
+      - name: bootstrap runner (idempotent)
+        if: steps.skip-check.outputs.skip != '1'
+        run: |
+          set -e
+          retry() { for i in 1 2 3; do "$@" && return 0; rc=$?; echo "retry $i (exit=$rc)" >&2; sleep $((i*5)); done; return 1; }
+          retry pacman -Syu --noconfirm --needed base-devel git rsync gnupg openssh sudo luajit readline curl
+
+      - name: import signing key
+        if: steps.skip-check.outputs.skip != '1'
+        env:
+          PRIV: ${{ secrets.MARFRIT_REPO_PRIVATE_KEY }}
+          PASS: ${{ secrets.MARFRIT_REPO_PASSPHRASE }}
+        run: |
+          set -e
+          gpgconf --homedir /root/.gnupg --kill all 2>/dev/null || true
+          rm -rf /root/.gnupg /root/repo_pass
+          mkdir -m700 -p /root/.gnupg
+          printf '%s' "$PASS" > /root/repo_pass
+          chmod 600 /root/repo_pass
+          printf '%s\n' "$PRIV" | gpg --batch --import
+          echo "92D5E96D8F63C75E4116AA1FF5C8C4603D0D250C:6:" | gpg --import-ownertrust
+
+      - name: install deploy ssh key
+        if: steps.skip-check.outputs.skip != '1'
+        env:
+          KEY: ${{ secrets.MARFRIT_REPO_DEPLOY_KEY }}
+        run: |
+          mkdir -m700 -p /root/.ssh
+          printf '%s\n' "$KEY" > /root/.ssh/id_ed25519
+          chmod 600 /root/.ssh/id_ed25519
+          ssh-keyscan -t ed25519 nc.reauktion.de > /root/.ssh/known_hosts 2>/dev/null
+
+      - name: makepkg aish
+        if: steps.skip-check.outputs.skip != '1'
+        run: |
+          set -e
+          rm -rf /tmp/build-aish
+          cp -r arch/aish /tmp/build-aish
+          chown -R builder:builder /tmp/build-aish
+          cd /tmp/build-aish
+          sudo -u builder -H makepkg --nocheck --noconfirm --syncdeps --cleanbuild
+          ls -la *.pkg.tar.* | grep -v "\.sig$"
+
+      - name: sign aish
+        if: steps.skip-check.outputs.skip != '1'
+        run: |
+          set -e
+          cd /tmp/build-aish
+          for f in *.pkg.tar.xz *.pkg.tar.zst *.pkg.tar.gz; do
+            [ -f "$f" ] || continue
+            gpg --batch --pinentry-mode loopback --passphrase-file /root/repo_pass \
+                --detach-sign --yes -u 92D5E96D8F63C75E4116AA1FF5C8C4603D0D250C "$f"
+          done
+
+      - name: publish aish to both arches
+        if: steps.skip-check.outputs.skip != '1'
+        run: |
+          set -e
+          retry() { for i in 1 2 3; do "$@" && return 0; rc=$?; echo "retry $i (exit=$rc)" >&2; sleep $((i*5)); done; return 1; }
+          export GNUPGHOME=/root/.gnupg
+          printf 'pinentry-mode loopback\npassphrase-file /root/repo_pass\n' > /root/.gnupg/gpg.conf
+          printf 'allow-loopback-pinentry\n' > /root/.gnupg/gpg-agent.conf
+          gpg-connect-agent reloadagent /bye
+
+          for target in aarch64 x86_64; do
+              stage="/tmp/arch-stage-$target"
+              rm -rf "$stage"; mkdir -p "$stage"; cd "$stage"
+              for f in marfrit.db.tar.gz marfrit.db.tar.gz.sig marfrit.files.tar.gz marfrit.files.tar.gz.sig; do
+                curl -sSLf "https://packages.reauktion.de/arch/$target/$f" -o "$f" || rm -f "$f"
+              done
+              cp /tmp/build-aish/*.pkg.tar.* .
+              pkgs=()
+              for ext in xz zst gz; do
+                for f in *.pkg.tar.$ext; do [ -f "$f" ] && pkgs+=("$f"); done
+              done
+              if [ -f marfrit.db.tar.gz ]; then
+                for f in "${pkgs[@]}"; do
+                  name=$(echo "$f" | sed -E 's/-[0-9].*//')
+                  repo-remove --sign --key 92D5E96D8F63C75E4116AA1FF5C8C4603D0D250C \
+                    marfrit.db.tar.gz "$name" 2>/dev/null || true
+                done
+              fi
+              repo-add --new --sign --key 92D5E96D8F63C75E4116AA1FF5C8C4603D0D250C \
+                --verify marfrit.db.tar.gz "${pkgs[@]}"
+              ln -sf marfrit.db.tar.gz        marfrit.db
+              ln -sf marfrit.files.tar.gz     marfrit.files
+              ln -sf marfrit.db.tar.gz.sig    marfrit.db.sig
+              ln -sf marfrit.files.tar.gz.sig marfrit.files.sig
+              retry rsync -avL --copy-unsafe-links \
+                -e 'ssh -i /root/.ssh/id_ed25519' \
+                ./ "mfritsche@nc.reauktion.de:arch/$target/"
+          done
+
+      - name: wipe secrets
+        if: always()
+        run: rm -f /root/repo_pass /root/.ssh/id_ed25519
+
+  aish-debian:
+    needs: aish-any   # serialize after the Arch build to share the runner
+    runs-on: arch-aarch64
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: skip if already published
+        id: skip-check
+        run: |
+          set -e
+          result=$(./.gitea/scripts/check-already-published.sh debian/aish)
+          echo "$result" >> "$GITHUB_OUTPUT"
+          echo "decision: $result"
+
+      - name: install dpkg
+        if: steps.skip-check.outputs.skip != '1'
+        run: |
+          set -e
+          retry() { for i in 1 2 3; do "$@" && return 0; rc=$?; echo "retry $i (exit=$rc)" >&2; sleep $((i*5)); done; return 1; }
+          retry pacman -Syu --noconfirm --needed dpkg openssh rsync curl
+
+      - name: install hertz deploy ssh key
+        if: steps.skip-check.outputs.skip != '1'
+        env:
+          KEY: ${{ secrets.MARFRIT_REPO_HERTZ_KEY }}
+        run: |
+          mkdir -m700 -p /root/.ssh
+          printf '%s\n' "$KEY" > /root/.ssh/id_ed25519_hertz
+          chmod 600 /root/.ssh/id_ed25519_hertz
+          ssh-keyscan -t ed25519 hertz.fritz.box >> /root/.ssh/known_hosts 2>/dev/null
+
+      - name: build aish .deb
+        if: steps.skip-check.outputs.skip != '1'
+        run: |
+          set -e
+          cd debian/aish
+          ./build-deb.sh
+          ls -la *.deb
+
+      - name: upload + publish to suites
+        if: steps.skip-check.outputs.skip != '1'
+        run: |
+          set -e
+          retry() { for i in 1 2 3; do "$@" && return 0; rc=$?; echo "retry $i (exit=$rc)" >&2; sleep $((i*5)); done; return 1; }
+          cd debian/aish
+          DEB=$(ls aish_*.deb | head -1)
+          # Push the .deb into hertz's incoming dir via rrsync.
+          retry rsync -av -e 'ssh -i /root/.ssh/id_ed25519_hertz' "$DEB" \
+              marfritrepo@hertz.fritz.box:
+          # Trigger reprepro for each suite.
+          for suite in bookworm trixie; do
+              retry ssh -i /root/.ssh/id_ed25519_hertz marfritrepo@hertz.fritz.box \
+                  "publish-deb $suite $DEB"
+          done
+
+      - name: wipe secrets
+        if: always()
+        run: rm -f /root/.ssh/id_ed25519_hertz
+
@@ -0,0 +1,53 @@
+# Maintainer: Markus Fritsche <mfritsche@reauktion.de>
+# aish — AI-augmented conversational shell in LuaJIT.
+# Source of truth: git.reauktion.de/marfrit/aish
+
+pkgname=aish
+pkgver=0.1.0
+pkgrel=1
+pkgdesc="AI-augmented conversational shell (LuaJIT, FFI-only)"
+arch=('any')
+url="https://git.reauktion.de/marfrit/aish"
+license=('MIT')
+depends=('luajit' 'readline' 'curl')
+# The _tag back-translation handles both clean releases (no '_') and
+# pre-release pkgvers (e.g. 0.1.0_rc1 → v0.1.0-rc1).
+_tag="v${pkgver//_/-}"
+source=("${pkgname}-${pkgver}.tar.gz::https://git.reauktion.de/marfrit/aish/archive/${_tag}.tar.gz")
+sha256sums=('9ebc3939e028832e39391ae33efacb5ec9bcd99d123cbc8ca1cd6ca9a640b5b5')
+
+package() {
+    cd "${pkgname}"
+    local libdir="${pkgdir}/usr/share/lua/5.1/aish"
+
+    # Top-level modules
+    install -Dm644 main.lua     "${libdir}/main.lua"
+    install -Dm644 broker.lua   "${libdir}/broker.lua"
+    install -Dm644 context.lua  "${libdir}/context.lua"
+    install -Dm644 executor.lua "${libdir}/executor.lua"
+    install -Dm644 history.lua  "${libdir}/history.lua"
+    install -Dm644 mcp.lua      "${libdir}/mcp.lua"
+    install -Dm644 renderer.lua "${libdir}/renderer.lua"
+    install -Dm644 repl.lua     "${libdir}/repl.lua"
+    install -Dm644 router.lua   "${libdir}/router.lua"
+    install -Dm644 safety.lua   "${libdir}/safety.lua"
+    install -Dm644 secrets.lua  "${libdir}/secrets.lua"
+
+    # FFI bindings
+    install -Dm644 ffi/curl.lua     "${libdir}/ffi/curl.lua"
+    install -Dm644 ffi/libc.lua     "${libdir}/ffi/libc.lua"
+    install -Dm644 ffi/pty.lua      "${libdir}/ffi/pty.lua"
+    install -Dm644 ffi/readline.lua "${libdir}/ffi/readline.lua"
+
+    # Vendored dependencies
+    install -Dm644 vendor/dkjson.lua "${libdir}/vendor/dkjson.lua"
+
+    # Launch wrapper
+    install -Dm755 bin/aish "${pkgdir}/usr/bin/aish"
+
+    # Documentation + example config
+    install -Dm644 README.md  "${pkgdir}/usr/share/doc/${pkgname}/README.md"
+    install -Dm644 LICENSE    "${pkgdir}/usr/share/doc/${pkgname}/LICENSE"
+    install -Dm644 examples/config.lua \
+        "${pkgdir}/usr/share/doc/${pkgname}/examples/config.lua"
+}
@@ -8,13 +8,13 @@
 # NEXT.md alongside this PKGBUILD for the full rationale and the
 # validation log on PineTab2 (RK3566).
 #
-# Multi-arch: builds natively on x86_64 and aarch64. The x86_64 path
-# is primarily a development / CI host; the runtime target audience is
-# aarch64. The two patches are architecture-independent.
+# Cross-compiled from x86_64 using chromium's bundled clang (upstream
+# LLVM doesn't ship clang 23+ yet; chromium's internal fork is required).
+# Runtime target is aarch64. The three patches are architecture-independent.

 pkgname=chromium-fourier
-pkgver=147.0.7727.116
-pkgrel=2
+pkgver=148.0.7778.178
+pkgrel=1
 epoch=1
 pkgdesc='Chromium with V4L2VDA HW video decode unlocked for mainline Linux Wayland on Rockchip'
 arch=('aarch64' 'x86_64')
@@ -150,7 +150,6 @@ build() {
    'symbol_level=0'
    'is_cfi=false'
    'treat_warnings_as_errors=false'
-    'enable_nacl=false'
    'enable_widevine=false'

    # System toolchain (clang/lld from pacman)
@@ -73,7 +73,7 @@ diff --git a/ui/ozone/common/native_pixmap_egl_binding.cc b/ui/ozone/common/nati
 index 31877f4459..6855c1093e 100644
 --- a/ui/ozone/common/native_pixmap_egl_binding.cc
 +++ b/ui/ozone/common/native_pixmap_egl_binding.cc
-@@ -6,10 +6,13 @@
+@@ -6,9 +6,12 @@

 #include <array>

@@ -82,7 +82,6 @@ index 31877f4459..6855c1093e 100644
 #include "base/memory/scoped_refptr.h"
 +#include "base/no_destructor.h"
 #include "base/notreached.h"
- #include "base/numerics/safe_conversions.h"
 +#include "base/synchronization/lock.h"
 #include "ui/gfx/linux/drm_util_linux.h"
 #include "ui/gl/gl_bindings.h"
@@ -0,0 +1,92 @@
+From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
+From: claude-noether <claude-noether@noreply.localhost>
+Date: Sun, 25 May 2026 12:00:00 +0200
+Subject: [PATCH] avcodec/aarch64/h264dsp: route H.264 luma-h deblock through daedalus-fourier
+
+Sibling of 0005 (which substituted v_loop_filter_luma).  Same
+NEON-to-NEON substitution: H264DSPContext.h_loop_filter_luma →
+daedalus_recipe_dispatch_h264_deblock_luma_h.  The H kernel landed
+in daedalus-fourier PR #9 (CPU NEON only — no QPU shader yet).
+
+libavcodec.so ctx is no-QPU per the existing 0003-0005 / 0007
+pattern; we cannot assume Vulkan in arbitrary host processes
+(firefox-fourier RDD, mpv-fourier, etc.).
+
+Intra (bS=4) h_loop_filter_luma_intra stays on the in-tree NEON .S
+code; daedalus_h264_deblock_meta only covers the non-intra path.
+An intra-h substitution can land once daedalus-fourier exposes a
+dispatch helper (the kernel already exists internally per PR #11).
+
+Refs reauktion/daedalus-v4l2#11 — substitution arc step 2 cycle 8 H.
+---
+diff --git a/libavcodec/aarch64/h264_idct_daedalus.c b/libavcodec/aarch64/h264_idct_daedalus.c
+--- a/libavcodec/aarch64/h264_idct_daedalus.c	2026-05-25 13:09:33.694760715 +0200
+++ libavcodec/aarch64/h264_idct_daedalus.c	2026-05-25 13:09:33.715603719 +0200
+@@ -1,9 +1,10 @@
+ /*
+- * H.264 4x4 / 8x8 IDCT + luma-v deblock — daedalus-fourier substitution shims.
+ * H.264 4x4 / 8x8 IDCT + luma v/h deblock — daedalus-fourier substitution shims.
+  *
+  * Routes H264DSPContext.idct_add           → daedalus_recipe_dispatch_h264_idct4
+  *        H264DSPContext.idct8_add          → daedalus_recipe_dispatch_h264_idct8
+  *        H264DSPContext.v_loop_filter_luma → daedalus_recipe_dispatch_h264_deblock_luma_v
+ *        H264DSPContext.h_loop_filter_luma → daedalus_recipe_dispatch_h264_deblock_luma_h
+  * instead of the in-tree ff_h264_*_neon assembly.  The recipe layer
+  * picks the substrate (CPU NEON for cycles 6 + 7 by default; cycle 8
+  * is CPU primary with QPU opportunistic — the ctx below is no-QPU,
+@@ -45,6 +46,8 @@
+ void ff_h264_idct8_add_daedalus(uint8_t *dst, int16_t *block, int stride);
+ void ff_h264_v_loop_filter_luma_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                          int alpha, int beta, int8_t *tc0);
+void ff_h264_h_loop_filter_luma_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                         int alpha, int beta, int8_t *tc0);
+ 
+ void ff_h264_idct_add_daedalus(uint8_t *dst, int16_t *block, int stride)
+ {
+@@ -84,3 +87,22 @@
+     daedalus_recipe_dispatch_h264_deblock_luma_v(g_dctx, pix, (size_t)stride,
+                                                  1, &meta);
+ }
+
+void ff_h264_h_loop_filter_luma_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                         int alpha, int beta, int8_t *tc0)
+{
+    daedalus_h264_deblock_meta meta = {
+        .dst_off = 0,
+        .alpha   = alpha,
+        .beta    = beta,
+    };
+    meta.tc0[0] = tc0[0];
+    meta.tc0[1] = tc0[1];
+    meta.tc0[2] = tc0[2];
+    meta.tc0[3] = tc0[3];
+
+    pthread_once(&g_dctx_once, daedalus_ctx_init_once);
+
+    daedalus_recipe_dispatch_h264_deblock_luma_h(g_dctx, pix, (size_t)stride,
+                                                 1, &meta);
+}
+diff --git a/libavcodec/aarch64/h264dsp_init_aarch64.c b/libavcodec/aarch64/h264dsp_init_aarch64.c
+--- a/libavcodec/aarch64/h264dsp_init_aarch64.c	2026-05-25 13:09:33.695937103 +0200
+++ libavcodec/aarch64/h264dsp_init_aarch64.c	2026-05-25 13:09:33.715541700 +0200
+@@ -31,6 +31,8 @@
+                                          int alpha, int beta, int8_t *tc0);
+ void ff_h264_h_loop_filter_luma_neon(uint8_t *pix, ptrdiff_t stride, int alpha,
+                                      int beta, int8_t *tc0);
+void ff_h264_h_loop_filter_luma_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                         int alpha, int beta, int8_t *tc0);
+ void ff_h264_v_loop_filter_luma_intra_neon(uint8_t *pix, ptrdiff_t stride, int alpha,
+                                            int beta);
+ void ff_h264_h_loop_filter_luma_intra_neon(uint8_t *pix, ptrdiff_t stride, int alpha,
+@@ -117,7 +119,7 @@
+ 
+     if (have_neon(cpu_flags) && bit_depth == 8) {
+         c->v_loop_filter_luma   = ff_h264_v_loop_filter_luma_daedalus;
+-        c->h_loop_filter_luma   = ff_h264_h_loop_filter_luma_neon;
+        c->h_loop_filter_luma   = ff_h264_h_loop_filter_luma_daedalus;
+         c->v_loop_filter_luma_intra= ff_h264_v_loop_filter_luma_intra_neon;
+         c->h_loop_filter_luma_intra= ff_h264_h_loop_filter_luma_intra_neon;
+ 
+--
+2.47.3
+
@@ -0,0 +1,127 @@
+From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
+From: claude-noether <claude-noether@noreply.localhost>
+Date: Sun, 25 May 2026 12:00:00 +0200
+Subject: [PATCH] avcodec/aarch64/h264dsp: route H.264 chroma v/h deblock through daedalus-fourier
+
+Chroma siblings of 0005 (luma_v) and 0008 (luma_h).  Same
+NEON-to-NEON pattern via the daedalus recipe layer:
+
+  H264DSPContext.v_loop_filter_chroma →
+    daedalus_recipe_dispatch_h264_deblock_chroma_v
+  H264DSPContext.h_loop_filter_chroma →
+    daedalus_recipe_dispatch_h264_deblock_chroma_h
+
+Both kernels landed in daedalus-fourier PR #10.  Recipe table
+routes AUTO to CPU NEON (no chroma QPU shaders yet), so this
+is plumbing-only and stays bit-exact against the in-tree NEON.
+
+Intra chroma (bS=4) loop filters remain on in-tree NEON;
+daedalus_h264_deblock_meta covers the non-intra (bS<4) path.
+
+Refs reauktion/daedalus-v4l2#11 — substitution arc step 2 cycle 8 chroma.
+---
+diff --git a/libavcodec/aarch64/h264_idct_daedalus.c b/libavcodec/aarch64/h264_idct_daedalus.c
+--- a/libavcodec/aarch64/h264_idct_daedalus.c	2026-05-25 13:15:45.995368233 +0200
+++ libavcodec/aarch64/h264_idct_daedalus.c	2026-05-25 13:15:46.015839177 +0200
+@@ -1,10 +1,12 @@
+ /*
+- * H.264 4x4 / 8x8 IDCT + luma v/h deblock — daedalus-fourier substitution shims.
+ * H.264 4x4 / 8x8 IDCT + luma v/h + chroma v/h deblock — daedalus-fourier substitution shims.
+  *
+  * Routes H264DSPContext.idct_add           → daedalus_recipe_dispatch_h264_idct4
+  *        H264DSPContext.idct8_add          → daedalus_recipe_dispatch_h264_idct8
+- *        H264DSPContext.v_loop_filter_luma → daedalus_recipe_dispatch_h264_deblock_luma_v
+- *        H264DSPContext.h_loop_filter_luma → daedalus_recipe_dispatch_h264_deblock_luma_h
+ *        H264DSPContext.v_loop_filter_luma   → daedalus_recipe_dispatch_h264_deblock_luma_v
+ *        H264DSPContext.h_loop_filter_luma   → daedalus_recipe_dispatch_h264_deblock_luma_h
+ *        H264DSPContext.v_loop_filter_chroma → daedalus_recipe_dispatch_h264_deblock_chroma_v
+ *        H264DSPContext.h_loop_filter_chroma → daedalus_recipe_dispatch_h264_deblock_chroma_h
+  * instead of the in-tree ff_h264_*_neon assembly.  The recipe layer
+  * picks the substrate (CPU NEON for cycles 6 + 7 by default; cycle 8
+  * is CPU primary with QPU opportunistic — the ctx below is no-QPU,
+@@ -48,6 +50,10 @@
+                                          int alpha, int beta, int8_t *tc0);
+ void ff_h264_h_loop_filter_luma_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                          int alpha, int beta, int8_t *tc0);
+void ff_h264_v_loop_filter_chroma_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                           int alpha, int beta, int8_t *tc0);
+void ff_h264_h_loop_filter_chroma_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                           int alpha, int beta, int8_t *tc0);
+ 
+ void ff_h264_idct_add_daedalus(uint8_t *dst, int16_t *block, int stride)
+ {
+@@ -106,3 +112,41 @@
+     daedalus_recipe_dispatch_h264_deblock_luma_h(g_dctx, pix, (size_t)stride,
+                                                  1, &meta);
+ }
+
+void ff_h264_v_loop_filter_chroma_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                           int alpha, int beta, int8_t *tc0)
+{
+    daedalus_h264_deblock_meta meta = {
+        .dst_off = 0,
+        .alpha   = alpha,
+        .beta    = beta,
+    };
+    meta.tc0[0] = tc0[0];
+    meta.tc0[1] = tc0[1];
+    meta.tc0[2] = tc0[2];
+    meta.tc0[3] = tc0[3];
+
+    pthread_once(&g_dctx_once, daedalus_ctx_init_once);
+
+    daedalus_recipe_dispatch_h264_deblock_chroma_v(g_dctx, pix, (size_t)stride,
+                                                   1, &meta);
+}
+
+void ff_h264_h_loop_filter_chroma_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                           int alpha, int beta, int8_t *tc0)
+{
+    daedalus_h264_deblock_meta meta = {
+        .dst_off = 0,
+        .alpha   = alpha,
+        .beta    = beta,
+    };
+    meta.tc0[0] = tc0[0];
+    meta.tc0[1] = tc0[1];
+    meta.tc0[2] = tc0[2];
+    meta.tc0[3] = tc0[3];
+
+    pthread_once(&g_dctx_once, daedalus_ctx_init_once);
+
+    daedalus_recipe_dispatch_h264_deblock_chroma_h(g_dctx, pix, (size_t)stride,
+                                                   1, &meta);
+}
+diff --git a/libavcodec/aarch64/h264dsp_init_aarch64.c b/libavcodec/aarch64/h264dsp_init_aarch64.c
+--- a/libavcodec/aarch64/h264dsp_init_aarch64.c	2026-05-25 13:15:45.996482360 +0200
+++ libavcodec/aarch64/h264dsp_init_aarch64.c	2026-05-25 13:15:46.025604910 +0200
+@@ -39,8 +39,12 @@
+                                            int beta);
+ void ff_h264_v_loop_filter_chroma_neon(uint8_t *pix, ptrdiff_t stride, int alpha,
+                                        int beta, int8_t *tc0);
+void ff_h264_v_loop_filter_chroma_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                           int alpha, int beta, int8_t *tc0);
+ void ff_h264_h_loop_filter_chroma_neon(uint8_t *pix, ptrdiff_t stride, int alpha,
+                                        int beta, int8_t *tc0);
+void ff_h264_h_loop_filter_chroma_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                           int alpha, int beta, int8_t *tc0);
+ void ff_h264_h_loop_filter_chroma422_neon(uint8_t *pix, ptrdiff_t stride, int alpha,
+                                           int beta, int8_t *tc0);
+ void ff_h264_v_loop_filter_chroma_intra_neon(uint8_t *pix, ptrdiff_t stride,
+@@ -123,11 +127,11 @@
+         c->v_loop_filter_luma_intra= ff_h264_v_loop_filter_luma_intra_neon;
+         c->h_loop_filter_luma_intra= ff_h264_h_loop_filter_luma_intra_neon;
+ 
+-        c->v_loop_filter_chroma = ff_h264_v_loop_filter_chroma_neon;
+        c->v_loop_filter_chroma = ff_h264_v_loop_filter_chroma_daedalus;
+         c->v_loop_filter_chroma_intra = ff_h264_v_loop_filter_chroma_intra_neon;
+ 
+         if (chroma_format_idc <= 1) {
+-            c->h_loop_filter_chroma = ff_h264_h_loop_filter_chroma_neon;
+            c->h_loop_filter_chroma = ff_h264_h_loop_filter_chroma_daedalus;
+             c->h_loop_filter_chroma_intra = ff_h264_h_loop_filter_chroma_intra_neon;
+             c->h_loop_filter_chroma_mbaff_intra = ff_h264_h_loop_filter_chroma_mbaff_intra_neon;
+         } else {
+--
+2.47.3
+
@@ -0,0 +1,126 @@
+From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
+From: claude-noether <claude-noether@noreply.localhost>
+Date: Sun, 25 May 2026 12:30:00 +0200
+Subject: [PATCH] avcodec/aarch64/h264dsp: route H.264 luma intra deblock through daedalus-fourier
+
+Adds the bS=4 intra-strength variants of the already-substituted
+luma_v / luma_h deblock (0005, 0008).  Intra MBs and certain
+inter-MB edges (4x4 transform boundaries inside an Intra_NxN
+neighbour) force boundary strength to 4 per H.264 §8.7.2.1.
+
+  H264DSPContext.v_loop_filter_luma_intra →
+    daedalus_recipe_dispatch_h264_deblock_luma_v_intra
+  H264DSPContext.h_loop_filter_luma_intra →
+    daedalus_recipe_dispatch_h264_deblock_luma_h_intra
+
+Both kernels landed in daedalus-fourier PR #11.  Recipe table
+routes AUTO to CPU NEON (no intra QPU shaders yet) — plumbing-
+only NEON-to-NEON via daedalus, bit-exact against the in-tree
+FFmpeg NEON path.
+
+Signature differs from bS<4: no tc0 argument.  The wrapper
+passes daedalus_h264_deblock_meta with alpha/beta set; tc0[] is
+ignored by the intra dispatch (bS=4 hardcodes the strength).
+
+Chroma intra variants are deferred to a follow-up PR because the
+chroma path has a 4:2:0 / 4:2:2 split (chroma_format_idc gating)
+that needs explicit conditional substitution to avoid running
+the 4:2:0-only daedalus dispatch on 4:2:2 chroma.
+
+Refs reauktion/daedalus-v4l2#11 — substitution arc step 2 cycle 8 intra.
+---
+diff --git a/libavcodec/aarch64/h264_idct_daedalus.c b/libavcodec/aarch64/h264_idct_daedalus.c
+--- a/libavcodec/aarch64/h264_idct_daedalus.c	2026-05-25 13:18:54.992244965 +0200
+++ libavcodec/aarch64/h264_idct_daedalus.c	2026-05-25 13:20:12.338122217 +0200
+@@ -1,5 +1,5 @@
+ /*
+- * H.264 4x4 / 8x8 IDCT + luma v/h + chroma v/h deblock — daedalus-fourier substitution shims.
+ * H.264 4x4 / 8x8 IDCT + luma v/h (inter + intra) + chroma v/h deblock — daedalus-fourier substitution shims.
+  *
+  * Routes H264DSPContext.idct_add           → daedalus_recipe_dispatch_h264_idct4
+  *        H264DSPContext.idct8_add          → daedalus_recipe_dispatch_h264_idct8
+@@ -7,6 +7,8 @@
+  *        H264DSPContext.h_loop_filter_luma   → daedalus_recipe_dispatch_h264_deblock_luma_h
+  *        H264DSPContext.v_loop_filter_chroma → daedalus_recipe_dispatch_h264_deblock_chroma_v
+  *        H264DSPContext.h_loop_filter_chroma → daedalus_recipe_dispatch_h264_deblock_chroma_h
+ *        H264DSPContext.v_loop_filter_luma_intra → daedalus_recipe_dispatch_h264_deblock_luma_v_intra
+ *        H264DSPContext.h_loop_filter_luma_intra → daedalus_recipe_dispatch_h264_deblock_luma_h_intra
+  * instead of the in-tree ff_h264_*_neon assembly.  The recipe layer
+  * picks the substrate (CPU NEON for cycles 6 + 7 by default; cycle 8
+  * is CPU primary with QPU opportunistic — the ctx below is no-QPU,
+@@ -54,6 +56,10 @@
+                                            int alpha, int beta, int8_t *tc0);
+ void ff_h264_h_loop_filter_chroma_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                            int alpha, int beta, int8_t *tc0);
+void ff_h264_v_loop_filter_luma_intra_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                                int alpha, int beta);
+void ff_h264_h_loop_filter_luma_intra_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                                int alpha, int beta);
+ 
+ void ff_h264_idct_add_daedalus(uint8_t *dst, int16_t *block, int stride)
+ {
+@@ -150,3 +156,34 @@
+     daedalus_recipe_dispatch_h264_deblock_chroma_h(g_dctx, pix, (size_t)stride,
+                                                    1, &meta);
+ }
+
+void ff_h264_v_loop_filter_luma_intra_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                                int alpha, int beta)
+{
+    daedalus_h264_deblock_meta meta = {
+        .dst_off = 0,
+        .alpha   = alpha,
+        .beta    = beta,
+    };
+    /* tc0[] is ignored by the intra-strength dispatch (bS=4 hardcodes the strength). */
+
+    pthread_once(&g_dctx_once, daedalus_ctx_init_once);
+
+    daedalus_recipe_dispatch_h264_deblock_luma_v_intra(g_dctx, pix, (size_t)stride,
+                                                        1, &meta);
+}
+
+void ff_h264_h_loop_filter_luma_intra_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                                int alpha, int beta)
+{
+    daedalus_h264_deblock_meta meta = {
+        .dst_off = 0,
+        .alpha   = alpha,
+        .beta    = beta,
+    };
+
+    pthread_once(&g_dctx_once, daedalus_ctx_init_once);
+
+    daedalus_recipe_dispatch_h264_deblock_luma_h_intra(g_dctx, pix, (size_t)stride,
+                                                        1, &meta);
+}
+diff --git a/libavcodec/aarch64/h264dsp_init_aarch64.c b/libavcodec/aarch64/h264dsp_init_aarch64.c
+--- a/libavcodec/aarch64/h264dsp_init_aarch64.c	2026-05-25 13:18:54.993349573 +0200
+++ libavcodec/aarch64/h264dsp_init_aarch64.c	2026-05-25 13:20:12.338265830 +0200
+@@ -35,8 +35,12 @@
+                                          int alpha, int beta, int8_t *tc0);
+ void ff_h264_v_loop_filter_luma_intra_neon(uint8_t *pix, ptrdiff_t stride, int alpha,
+                                            int beta);
+void ff_h264_v_loop_filter_luma_intra_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                                int alpha, int beta);
+ void ff_h264_h_loop_filter_luma_intra_neon(uint8_t *pix, ptrdiff_t stride, int alpha,
+                                            int beta);
+void ff_h264_h_loop_filter_luma_intra_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                                int alpha, int beta);
+ void ff_h264_v_loop_filter_chroma_neon(uint8_t *pix, ptrdiff_t stride, int alpha,
+                                        int beta, int8_t *tc0);
+ void ff_h264_v_loop_filter_chroma_daedalus(uint8_t *pix, ptrdiff_t stride,
+@@ -124,8 +128,8 @@
+     if (have_neon(cpu_flags) && bit_depth == 8) {
+         c->v_loop_filter_luma   = ff_h264_v_loop_filter_luma_daedalus;
+         c->h_loop_filter_luma   = ff_h264_h_loop_filter_luma_daedalus;
+-        c->v_loop_filter_luma_intra= ff_h264_v_loop_filter_luma_intra_neon;
+-        c->h_loop_filter_luma_intra= ff_h264_h_loop_filter_luma_intra_neon;
+        c->v_loop_filter_luma_intra= ff_h264_v_loop_filter_luma_intra_daedalus;
+        c->h_loop_filter_luma_intra= ff_h264_h_loop_filter_luma_intra_daedalus;
+ 
+         c->v_loop_filter_chroma = ff_h264_v_loop_filter_chroma_daedalus;
+         c->v_loop_filter_chroma_intra = ff_h264_v_loop_filter_chroma_intra_neon;
+--
+2.47.3
+
@@ -0,0 +1,101 @@
+From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
+From: claude-noether <claude-noether@noreply.localhost>
+Date: Sun, 25 May 2026 13:00:00 +0200
+Subject: [PATCH] avcodec/aarch64/h264dsp: route H.264 chroma DC Hadamard through daedalus-fourier
+
+Substitutes H264DSPContext.chroma_dc_dequant_idct in the
+4:2:0 / bit_depth=8 init path with a wrapper that composes
+the daedalus chroma DC Hadamard primitive (fourier PR #25)
+with qmul scaling FFmpeg does in one fused function.
+
+Bit-exact against ff_h264_chroma_dc_dequant_idct_8_c.
+Hadamard correctness gated by fourier PR #23 test suite.
+
+4:2:2 chroma stays on the in-tree 422 variant (same
+gating shape as 0009 chroma deblock substitution).
+
+Requires daedalus-fourier commit b9f9ff2 or later (PR #25
+exposing the public Hadamard symbol).  Pin bumps in PKGBUILD
+and build-deb.sh come in the same commit.
+---
+diff --git a/libavcodec/aarch64/h264_idct_daedalus.c b/libavcodec/aarch64/h264_idct_daedalus.c
+--- a/libavcodec/aarch64/h264_idct_daedalus.c	2026-05-25 13:38:32.019491484 +0200
+++ libavcodec/aarch64/h264_idct_daedalus.c	2026-05-25 13:38:32.033821507 +0200
+@@ -1,5 +1,5 @@
+ /*
+- * H.264 4x4 / 8x8 IDCT + luma v/h (inter + intra) + chroma v/h deblock — daedalus-fourier substitution shims.
+ * H.264 4x4 / 8x8 IDCT + luma v/h (inter+intra) + chroma v/h deblock + chroma DC Hadamard — daedalus-fourier substitution shims.
+  *
+  * Routes H264DSPContext.idct_add           → daedalus_recipe_dispatch_h264_idct4
+  *        H264DSPContext.idct8_add          → daedalus_recipe_dispatch_h264_idct8
+@@ -9,6 +9,7 @@
+  *        H264DSPContext.h_loop_filter_chroma → daedalus_recipe_dispatch_h264_deblock_chroma_h
+  *        H264DSPContext.v_loop_filter_luma_intra → daedalus_recipe_dispatch_h264_deblock_luma_v_intra
+  *        H264DSPContext.h_loop_filter_luma_intra → daedalus_recipe_dispatch_h264_deblock_luma_h_intra
+ *        H264DSPContext.chroma_dc_dequant_idct   → daedalus_h264_chroma_dc_hadamard_2x2 + caller-side qmul
+  * instead of the in-tree ff_h264_*_neon assembly.  The recipe layer
+  * picks the substrate (CPU NEON for cycles 6 + 7 by default; cycle 8
+  * is CPU primary with QPU opportunistic — the ctx below is no-QPU,
+@@ -60,6 +61,7 @@
+                                                 int alpha, int beta);
+ void ff_h264_h_loop_filter_luma_intra_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                                 int alpha, int beta);
+void ff_h264_chroma_dc_dequant_idct_daedalus(int16_t *block, int qmul);
+ 
+ void ff_h264_idct_add_daedalus(uint8_t *dst, int16_t *block, int stride)
+ {
+@@ -187,3 +189,32 @@
+     daedalus_recipe_dispatch_h264_deblock_luma_h_intra(g_dctx, pix, (size_t)stride,
+                                                         1, &meta);
+ }
+
+/* Composes daedalus_h264_chroma_dc_hadamard_2x2 with the qmul scaling
+ * that FFmpeg's reference does in one fused function (h264idct_template.c
+ * ff_h264_chroma_dc_dequant_idct).
+ *
+ * The 4 DC coefficients are scattered across the per-MB coefficient
+ * buffer at offsets [r*stride + c*xStride] (stride=32, xStride=16).
+ * Extract into a contiguous int16[4], run the Hadamard, then apply
+ * the qmul scale and write back to the original positions.
+ *
+ * No daedalus ctx needed; the Hadamard is a pure stateless primitive.
+ */
+void ff_h264_chroma_dc_dequant_idct_daedalus(int16_t *block, int qmul)
+{
+    enum { stride = 32, xStride = 16 };
+    int16_t dc[4];
+
+    dc[0] = block[stride*0 + xStride*0];
+    dc[1] = block[stride*0 + xStride*1];
+    dc[2] = block[stride*1 + xStride*0];
+    dc[3] = block[stride*1 + xStride*1];
+
+    daedalus_h264_chroma_dc_hadamard_2x2(dc);
+
+    block[stride*0 + xStride*0] = (int16_t)((int)dc[0] * qmul >> 7);
+    block[stride*0 + xStride*1] = (int16_t)((int)dc[1] * qmul >> 7);
+    block[stride*1 + xStride*0] = (int16_t)((int)dc[2] * qmul >> 7);
+    block[stride*1 + xStride*1] = (int16_t)((int)dc[3] * qmul >> 7);
+}
+diff --git a/libavcodec/aarch64/h264dsp_init_aarch64.c b/libavcodec/aarch64/h264dsp_init_aarch64.c
+--- a/libavcodec/aarch64/h264dsp_init_aarch64.c	2026-05-25 13:38:32.020346459 +0200
+++ libavcodec/aarch64/h264dsp_init_aarch64.c	2026-05-25 13:38:32.033909804 +0200
+@@ -41,6 +41,7 @@
+                                            int beta);
+ void ff_h264_h_loop_filter_luma_intra_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                                 int alpha, int beta);
+void ff_h264_chroma_dc_dequant_idct_daedalus(int16_t *block, int qmul);
+ void ff_h264_v_loop_filter_chroma_neon(uint8_t *pix, ptrdiff_t stride, int alpha,
+                                        int beta, int8_t *tc0);
+ void ff_h264_v_loop_filter_chroma_daedalus(uint8_t *pix, ptrdiff_t stride,
+@@ -135,6 +136,7 @@
+         c->v_loop_filter_chroma_intra = ff_h264_v_loop_filter_chroma_intra_neon;
+ 
+         if (chroma_format_idc <= 1) {
+            c->chroma_dc_dequant_idct = ff_h264_chroma_dc_dequant_idct_daedalus;
+             c->h_loop_filter_chroma = ff_h264_h_loop_filter_chroma_daedalus;
+             c->h_loop_filter_chroma_intra = ff_h264_h_loop_filter_chroma_intra_neon;
+             c->h_loop_filter_chroma_mbaff_intra = ff_h264_h_loop_filter_chroma_mbaff_intra_neon;
+--
+2.47.3
+
@@ -30,7 +30,7 @@ epoch=2
 # daedalus-fourier pin.  209a421 = PR #2 merge (Phase 8c — public API
 # gains daedalus_recipe_dispatch_h264_qpel_mc20 + DAEDALUS_KERNEL_H264_QPEL_MC20).
 # Cycle 9 closes the libavcodec.so substitution arc started at cycle 6.
-_daedalus_fourier_commit='209a4218bcb98b91c04f07ad61513bb04adb13ad'
+_daedalus_fourier_commit='b9f9ff2a89c068aea54dcb52b543afddad28311e'  # PR #25 — public chroma DC Hadamard symbol
 pkgdesc='FFmpeg with V4L2 Request API hwaccel (Rockchip / Allwinner stateless decode)'
 arch=('aarch64')
 url='https://github.com/Kwiboo/FFmpeg'
@@ -94,8 +94,12 @@ source=("git+https://github.com/Kwiboo/FFmpeg.git#commit=${_commit}"
        '0004-h264-idct8-daedalus-fourier.patch'
        '0005-h264-deblock-luma-v-daedalus-fourier.patch'
        '0006-h264-restore-low-delay.patch'
-        '0007-h264-qpel-mc20-daedalus-fourier.patch')
-sha256sums=('SKIP' 'SKIP' 'SKIP' 'SKIP' 'SKIP' 'SKIP' 'SKIP' 'SKIP' 'SKIP')
+        '0007-h264-qpel-mc20-daedalus-fourier.patch'
+        '0008-h264-deblock-luma-h-daedalus-fourier.patch'
+        '0009-h264-deblock-chroma-daedalus-fourier.patch'
+        '0010-h264-deblock-luma-intra-daedalus-fourier.patch'
+        '0011-h264-chroma-dc-hadamard-daedalus-fourier.patch')
+sha256sums=('SKIP' 'SKIP' 'SKIP' 'SKIP' 'SKIP' 'SKIP' 'SKIP' 'SKIP' 'SKIP' 'SKIP' 'SKIP' 'SKIP' 'SKIP')

 pkgver() {
  cd "${_srcname}"
@@ -113,6 +117,10 @@ prepare() {
  patch -Np1 -i "${srcdir}/0005-h264-deblock-luma-v-daedalus-fourier.patch"
  patch -Np1 -i "${srcdir}/0006-h264-restore-low-delay.patch"
  patch -Np1 -i "${srcdir}/0007-h264-qpel-mc20-daedalus-fourier.patch"
+  patch -Np1 -i "${srcdir}/0008-h264-deblock-luma-h-daedalus-fourier.patch"
+  patch -Np1 -i "${srcdir}/0009-h264-deblock-chroma-daedalus-fourier.patch"
+  patch -Np1 -i "${srcdir}/0010-h264-deblock-luma-intra-daedalus-fourier.patch"
+  patch -Np1 -i "${srcdir}/0011-h264-chroma-dc-hadamard-daedalus-fourier.patch"
 }

 build() {
@@ -0,0 +1,50 @@
+From: marfrit-packages noether <claude-noether@reauktion.de>
+Subject: [PATCH] panvk: report fragmentStoresAndAtomics = true on Bifrost
+
+Backports Mesa main's unconditional advertisement of
+fragmentStoresAndAtomics for panvk (snapshot ref: src/panfrost/vulkan/
+panvk_vX_physical_device.c at commit-time 2026-05-06; the line reads
+`.fragmentStoresAndAtomics = true,` on main with no PAN_ARCH gate).
+
+Motivation: Chromium Dawn's WebGPU initializer in
+third_party/dawn/src/dawn/native/vulkan/PhysicalDeviceVk.cpp:250
+unconditionally rejects any Vulkan adapter that doesn't advertise this
+feature, causing Dawn to fall back to the SwiftShader CPU adapter
+on PineTab2 / RK3566 / Mali-G52 r1 MC1 (PAN_ARCH 7). With this patch the
+device advertises true, satisfying Dawn's gate. Tracked at
+https://git.reauktion.de/marfrit/panvk-bifrost/issues/2.
+
+The disjunction with `instance->force_enable_shader_atomics` is
+preserved as a kill-switch: in compiler terms it's dead code
+(`true || X == true`), but it leaves the DRI option
+`pan_force_enable_shader_atomics` semantically wired so future
+rebases or downstream debugging can see the link to the runtime knob.
+
+Caveat: the existing DRI option's description in src/util/driconf.h
+still labels this as "may not work reliably and is for debug purposes
+only". Mesa main's choice to ship it as default-on for all panvk
+architectures (including Bifrost, which is non-conformant per the
+PAN_I_WANT_A_BROKEN_VULKAN_DRIVER gate) reflects an upstream judgment
+that the practical risk is acceptable. Verify-before-ship for this
+package: dEQP-VK.glsl.atomic_operations.* + dEQP-VK.image.store.*
+deltas vs the r4 baseline must show no new fails. Pass counts may rise
+(tests that previously NotSupported now run); the load-bearing line is
+the Failed column staying at zero.
+
+---
+ src/panfrost/vulkan/panvk_vX_physical_device.c | 3 +--
+ 1 file changed, 1 insertion(+), 2 deletions(-)
+
+diff --git a/src/panfrost/vulkan/panvk_vX_physical_device.c b/src/panfrost/vulkan/panvk_vX_physical_device.c
+--- a/src/panfrost/vulkan/panvk_vX_physical_device.c
+++ b/src/panfrost/vulkan/panvk_vX_physical_device.c
+@@ -280,8 +280,7 @@
+       .vertexPipelineStoresAndAtomics =
+          (PAN_ARCH >= 13 && instance->enable_vertex_pipeline_stores_atomics) ||
+          instance->force_enable_shader_atomics,
+-      .fragmentStoresAndAtomics =
+-         (PAN_ARCH >= 10) || instance->force_enable_shader_atomics,
+      .fragmentStoresAndAtomics = true || instance->force_enable_shader_atomics,
+       .shaderTessellationAndGeometryPointSize = false,
+       .shaderImageGatherExtended = true,
+       .shaderStorageImageExtendedFormats = true,
@@ -0,0 +1,51 @@
+From: marfrit-packages noether <claude-noether@reauktion.de>
+Subject: [PATCH] panvk: advertise VK_EXT_legacy_dithering on Bifrost
+
+Backports Mesa main's flip — vanilla 26.0.6 doesn't have the extension
+in the panvk advertisement list; main does (line 172 / 647 on snapshot
+617da94, 2026-05-06).
+
+VK_EXT_legacy_dithering exposes the classic OpenGL-style dithering
+behavior to Vulkan apps. Pure-software composition; no new HW path.
+ARM's own libmali driver release r51p0 (BXODROIDN2PL, Aug 2024) lists
+this extension in its Vulkan implementation for ODROID-N2 boards
+using the same Mali-G52 architecture family — confirms ARM ships it
+for Mali-G52-class hardware.
+
+Consumer benefit: dithering matters for low-bit-depth framebuffers
+(RGB565 / RGB5A1 — common on portable / battery-saving renders)
+where banding is visible. DXVK / vkd3d-proton both opt in when
+available.
+
+Verify-before-ship: vulkaninfo lists the extension and
+VkPhysicalDeviceLegacyDitheringFeaturesEXT.legacyDithering == true.
+
+Cross-refs:
+  - marfrit/panvk-bifrost research/r6_r7_mali_g52_feature_audit_2026-05-24.md
+  - ARM blob r51p0 strings dump (in-blob extension confirmed)
+
+---
+ src/panfrost/vulkan/panvk_vX_physical_device.c | 5 +++++
+ 1 file changed, 5 insertions(+)
+
+diff --git a/src/panfrost/vulkan/panvk_vX_physical_device.c b/src/panfrost/vulkan/panvk_vX_physical_device.c
+--- a/src/panfrost/vulkan/panvk_vX_physical_device.c
+++ b/src/panfrost/vulkan/panvk_vX_physical_device.c
+@@ -156,6 +156,7 @@
+       .EXT_image_drm_format_modifier = true,
+       .EXT_image_robustness = true,
+       .EXT_index_type_uint8 = true,
+      .EXT_legacy_dithering = true,
+       .EXT_line_rasterization = true,
+       .EXT_load_store_op_none = true,
+       .EXT_non_seamless_cube_map = true,
+@@ -552,6 +553,9 @@
+
+       /* VK_EXT_multisampled_render_to_single_sampled */
+       .multisampledRenderToSingleSampled = true,
+
+      /* VK_EXT_legacy_dithering */
+      .legacyDithering = true,
+    };
+ }
+
@@ -0,0 +1,103 @@
+From: marfrit-packages noether <claude-noether@reauktion.de>
+Subject: [PATCH] panvk-bifrost: fix XFB store channel-extract for packed varyings
+
+iter19 — fixes a reliable SIGSEGV during vkCreateGraphicsPipeline on any
+shader that uses XFB-bound varyings declared with non-zero `layout
+(component=N)` qualifiers. Surfaced by
+dEQP-VK.transform_feedback.simple.holes_vert; backtrace lands 11 frames
+into libvulkan_panfrost.so called from `vkt::TransformFeedback::
+TransformFeedbackHolesInstance::iterate`.
+
+Root cause: `lower_xfb_output_iter17` (and upstream `lower_xfb_output`,
+which carries a `// TODO` on the same assertion) computes the source-
+channel mask as `mask << channel_idx`, where `channel_idx` is the
+varying-location component (0..3) but `src` only contains channels for
+the source-side range starting at `nir_intrinsic_component(intr)`. For
+`flat out float vegeta` declared with `component=2`, NIR emits
+`store_output src=<vec1>, component=2`, and the lowering computes
+`mask << 2` against a single-component src — out-of-range; the
+resulting malformed nir_def then segfaults inside downstream NIR
+constant-folding (`nir_constant_expressions.c::evaluate_*`).
+
+The assertion `assert(nir_intrinsic_component(intr) == 0)` was inherited
+from upstream `pan_nir_lower_xfb.c` as a documented `// TODO`; release
+builds (-DNDEBUG) elide it. The fix translates `channel_idx` to the
+source-channel space by subtracting `nir_intrinsic_component(intr)`
+before shifting the mask, and replaces the elided asserts with explicit
+release-mode guards (the patch closes the same release-mode-elision
+class as the original bug).
+
+Verified on PineTab2 (Mali-G52 r1 MC1, PAN_ARCH 7) against vulkan-cts
+1.3.10.0:
+  - holes_vert / holes_extra_draw_vert no longer SIGSEGV (now Fail on
+    color-check; that is a separate iter20 finding — the rasterized
+    varying gets removed alongside the XFB-bound one).
+  - basic_*: 36/36 Pass. depth_clip_*: 1 Pass + 4 NotSupported.
+    lines_or_triangles*: 16 NotSupported. 0 Fail across the full set.
+  - holes_geom / holes_extra_draw_geom remain NotSupported
+    (geometryShader not on G52) — unchanged.
+
+Caveat: max_output_components_64/_128/_256 were never reached on the
+r5 sweep (watchdog killed transform_feedback after the holes_vert
+crash). With this fix in place, those tests now run and surface
+*their own pre-existing* coredumps — confirmed on shipped r6 baseline
+too. They are NOT regressions from this patch; they are latent crashes
+unmasked by it. iter20+ territory.
+
+Phase 5 (2nd-model) review: APPROVE WITH CHANGES (non-blocking).
+Changes applied: release-mode defensive guards on both preconditions
+plus a dispatcher-side comment clarifying the i*2+j semantics.
+
+Cross-refs:
+  - iter19/phase{0,1,2,3}_holes_vert*.md in panvk-bifrost repo
+
+---
+ src/panfrost/vulkan/panvk_vX_xfb_lower.c | 24 +++++++++++++++++++++---
+ 1 file changed, 21 insertions(+), 3 deletions(-)
+
+diff --git a/src/panfrost/vulkan/panvk_vX_xfb_lower.c b/src/panfrost/vulkan/panvk_vX_xfb_lower.c
+@@ -339,7 +339,20 @@
+                         unsigned buffer, unsigned offset_words)
+ {
+    assert(buffer < MAX_XFB_BUFFERS);
+-   assert(nir_intrinsic_component(intr) == 0);
+
+   /* iter19: nir_intrinsic_component(intr) is the source-channel base —
+    * for a packed varying like `layout (location=0, component=2) flat out
+    * float vegeta`, NIR emits store_output with component=2 and a single-
+    * component src. The XFB iteration index `channel_idx` (0..3) is the
+    * varying-location component, not the source channel. Translate by
+    * subtracting the base before shifting the mask. Fixes the long-
+    * standing `assert(nir_intrinsic_component(intr) == 0) // TODO` in
+    * upstream pan_nir_lower_xfb that surfaces on holes_vert. */
+   const unsigned base_comp = nir_intrinsic_component(intr);
+   /* Defensive against release-build elision: this is precisely the
+    * bug class the patch is fixing, so don't re-introduce it. */
+   if (channel_idx < base_comp)
+      return;
+ 
+    uint16_t stride = b->shader->info.xfb_stride[buffer] * 4;
+    assert(stride != 0);
+@@ -357,7 +370,11 @@
+ 
+    nir_def *src = intr->src[0].ssa;
+    nir_component_mask_t mask = nir_component_mask(num_components);
+-   nir_def *value = nir_channels(b, src, mask << channel_idx);
+   const unsigned src_channel = channel_idx - base_comp;
+   /* Same defensive class as the channel_idx >= base_comp guard above. */
+   if (src_channel + num_components > src->num_components)
+      return;
+   nir_def *value = nir_channels(b, src, mask << src_channel);
+ 
+    /* Topology dispatch ladder. LIST first (fast path). */
+    nir_push_if(b, nir_ieq_imm(b, topology, PANVK_XFB_TOPO_LIST));
+@@ -465,6 +482,9 @@
+       for (unsigned j = 0; j < 2; ++j) {
+          if (!xfb.out[j].num_components)
+             continue;
+         /* `i*2+j` is the varying-location component (0..3) — io_xfb covers
+          * slots 0..1, io_xfb2 covers 2..3. The leaf translates this into
+          * a source-channel index by subtracting nir_intrinsic_component(intr). */
+          lower_xfb_output_iter17(b, intr, i * 2 + j, xfb.out[j].num_components,
+                                  xfb.out[j].buffer, xfb.out[j].offset);
+          progress = true;
@@ -30,7 +30,7 @@

 pkgname=mesa-panvk-bifrost
 _mesaver=26.0.6
-pkgver=26.0.6.r4
+pkgver=26.0.6.r7
 pkgrel=1
 pkgdesc="Patched Mesa libvulkan_panfrost.so exposing Bifrost-gen Mali to Vulkan apps (panvk-bifrost campaign)"
 arch=('aarch64')
@@ -81,6 +81,9 @@ source=(
    "0002-panvk-expose-vulkan-1.1-1.2-on-bifrost.patch"
    "0003-panvk-bifrost-vk-ext-transform-feedback.patch"
    "0004-panvk-bifrost-xfb-primitive-decomposition.patch"
+    "0005-panvk-bifrost-fragment-stores-atomics.patch"
+    "0006-panvk-bifrost-legacy-dithering.patch"
+    "0007-panvk-bifrost-xfb-component-base-fix.patch"
    "brave-vulkan"
    "icd.json"
 )
@@ -92,6 +95,9 @@ sha256sums=(
    'SKIP'
    'SKIP'
    'SKIP'
+    'SKIP'
+    'SKIP'
+    'SKIP'
 )

 prepare() {
@@ -127,6 +133,37 @@ prepare() {
    # Phase-doc context: ~/src/panvk-bifrost/iter17/phase{0,1,2,4,5,6,8}_*.md.
    patch -p1 < "${srcdir}/0004-panvk-bifrost-xfb-primitive-decomposition.patch"

+    # r5 (2026-05-23): advertise .fragmentStoresAndAtomics = true on Bifrost
+    # to satisfy Chromium Dawn's WebGPU init gate
+    # (third_party/dawn/src/dawn/native/vulkan/PhysicalDeviceVk.cpp:250).
+    # Backports Mesa main's unconditional flip (same line as on main as of
+    # 2026-05-06). Disjunction with instance->force_enable_shader_atomics
+    # is preserved as a documented kill-switch even though the compiler
+    # folds it away. Closes marfrit/panvk-bifrost#2.
+    # Verify-before-ship: dEQP-VK.glsl.atomic_operations.* and
+    # dEQP-VK.image.store.* show no new Failed vs r4 baseline.
+    patch -p1 < "${srcdir}/0005-panvk-bifrost-fragment-stores-atomics.patch"
+
+    # r6 (2026-05-25): advertise VK_EXT_legacy_dithering. Backports Mesa
+    # main's unconditional flip. Pure-software composition; vk_render_pass
+    # already gates on enabled_features.legacyDithering and panvk_vX_blend
+    # + pan_format already plumb the dithered BLEND descriptor (BFMT2 table
+    # has MALI_BLEND_AU encodings for RGB565/RGB5A1/RGBA4/RGB10A2 on
+    # PAN_ARCH 7). Closes the EXT_legacy_dithering gap surfaced by
+    # marfrit/panvk-bifrost research/r6_r7_*. ARM blob r51p0 confirms the
+    # extension as Mali-G52-architecture supported.
+    patch -p1 < "${srcdir}/0006-panvk-bifrost-legacy-dithering.patch"
+
+    # r7 (2026-05-25): XFB store channel-extract fix for packed varyings.
+    # Eliminates a reliable SIGSEGV in vkCreateGraphicsPipeline whenever
+    # an XFB-bound vertex output is declared with non-zero
+    # `layout (component=N)`. Surfaced by dEQP-VK.transform_feedback.
+    # simple.holes_vert (now Fails on color-check rather than crashing;
+    # the color-check residual is a separate iter20 finding).
+    # Phase-doc context: ~/src/panvk-bifrost/iter19/phase{0,1,2,3}_*.md.
+    # Phase 5 reviewed; release-mode-elision defensive guards applied.
+    patch -p1 < "${srcdir}/0007-panvk-bifrost-xfb-component-base-fix.patch"
+
    # Sanity-check the patches landed.
    grep -q "KHR_robustness2 = true," src/panfrost/vulkan/panvk_vX_physical_device.c
    grep -q "EXT_robustness2 = true," src/panfrost/vulkan/panvk_vX_physical_device.c
@@ -138,9 +175,17 @@ prepare() {
    test -f src/panfrost/vulkan/jm/panvk_vX_cmd_xfb.c
    # iter17 sanity: pan_nir_lower_xfb call site has been replaced; new file present.
    grep -q "panvk_per_arch(nir_lower_xfb)" src/panfrost/vulkan/panvk_vX_shader.c
+    # r5 sanity: fragmentStoresAndAtomics = true patch landed
+    grep -q "fragmentStoresAndAtomics = true ||" src/panfrost/vulkan/panvk_vX_physical_device.c
+    # r6 sanity: VK_EXT_legacy_dithering advertised
+    grep -q '\.EXT_legacy_dithering = true,' src/panfrost/vulkan/panvk_vX_physical_device.c
+    grep -q '\.legacyDithering = true,' src/panfrost/vulkan/panvk_vX_physical_device.c
    grep -q "xfb_topology" src/panfrost/vulkan/panvk_shader.h
    grep -q "panvk_xfb_topology" src/panfrost/vulkan/panvk_shader.h
    test -f src/panfrost/vulkan/panvk_vX_xfb_lower.c
+    # r7 sanity: XFB channel-base correction landed
+    grep -q "iter19: nir_intrinsic_component(intr) is the source-channel base" src/panfrost/vulkan/panvk_vX_xfb_lower.c
+    grep -q "mask << src_channel" src/panfrost/vulkan/panvk_vX_xfb_lower.c
 }

 build() {
@@ -0,0 +1,85 @@
+#!/bin/bash
+# Build aish_<ver>_all.deb from this directory using dpkg-deb directly.
+# Run from inside the runner container, which has dpkg installed.
+#
+# Matches the lmcp build-deb.sh pattern: no dh/debhelper, no Build-Depends
+# beyond `dpkg`, structurally a normal apt package (Architecture: all).
+set -euo pipefail
+
+PKGVER=0.1.0
+UPSTREAM_TAG=v0.1.0
+PKGREL=1
+AISH_TARBALL_SHA256=9ebc3939e028832e39391ae33efacb5ec9bcd99d123cbc8ca1cd6ca9a640b5b5
+HERE=$(dirname "$(readlink -f "$0")")
+
+# Reproducible build: pin all file mtimes + ar member timestamps to a fixed
+# epoch tied to this packaging release (aish v0.1.0 — 2026-05-25 00:00 UTC).
+# Without this, repeat builds produce different byte streams and reprepro
+# refuses re-includes with "size expected: X, got: Y".
+export SOURCE_DATE_EPOCH=1779667200
+
+work=$(mktemp -d)
+trap "rm -rf $work" EXIT
+
+cd "$work"
+curl --connect-timeout 10 --max-time 600 --retry 3 --retry-delay 5 -sSLfo aish.tar.gz \
+    "https://git.reauktion.de/marfrit/aish/archive/${UPSTREAM_TAG}.tar.gz"
+echo "$AISH_TARBALL_SHA256  aish.tar.gz" | sha256sum -c
+tar xzf aish.tar.gz
+
+ROOT="$work/pkgroot"
+LIBDIR="$ROOT/usr/share/lua/5.1/aish"
+mkdir -p "$ROOT/DEBIAN" \
+         "$LIBDIR/ffi" \
+         "$LIBDIR/vendor" \
+         "$ROOT/usr/bin" \
+         "$ROOT/usr/share/doc/aish/examples"
+
+# Top-level modules
+for m in main broker context executor history mcp renderer repl router safety secrets; do
+    cp "aish/${m}.lua" "$LIBDIR/${m}.lua"
+done
+
+# FFI bindings
+for m in curl libc pty readline; do
+    cp "aish/ffi/${m}.lua" "$LIBDIR/ffi/${m}.lua"
+done
+
+# Vendored dependencies
+cp aish/vendor/dkjson.lua "$LIBDIR/vendor/dkjson.lua"
+
+# Launch wrapper
+install -m 755 aish/bin/aish "$ROOT/usr/bin/aish"
+
+# Documentation + example config
+cp aish/README.md          "$ROOT/usr/share/doc/aish/"
+cp aish/LICENSE            "$ROOT/usr/share/doc/aish/"
+cp aish/examples/config.lua "$ROOT/usr/share/doc/aish/examples/"
+cp "$HERE/debian/copyright" "$ROOT/usr/share/doc/aish/copyright"
+cp "$HERE/debian/changelog" "$ROOT/usr/share/doc/aish/changelog.Debian"
+gzip -9 -n "$ROOT/usr/share/doc/aish/changelog.Debian"
+
+cat > "$ROOT/DEBIAN/control" <<EOF
+Package: aish
+Version: ${PKGVER}-${PKGREL}
+Section: shells
+Priority: optional
+Architecture: all
+Depends: luajit, libreadline8t64 | libreadline8, libcurl4t64 | libcurl4
+Maintainer: Markus Fritsche <mfritsche@reauktion.de>
+Homepage: https://git.reauktion.de/marfrit/aish
+Description: AI-augmented conversational shell (LuaJIT, FFI-only)
+ aish is an interactive REPL that interleaves shell execution and
+ language-model conversation against llama.cpp HTTP brokers. Pure
+ LuaJIT 2.x with FFI bindings to libcurl, GNU readline, and libc.
+ .
+ Modules install under /usr/share/lua/5.1/aish/. The launcher is
+ /usr/bin/aish. Example configuration is at
+ /usr/share/doc/aish/examples/config.lua (copy to
+ ~/.config/aish/config.lua and adapt).
+EOF
+
+# Build the .deb. Output to current dir of the caller.
+DEB_OUT=aish_${PKGVER}-${PKGREL}_all.deb
+dpkg-deb --root-owner-group --build "$ROOT" "$HERE/$DEB_OUT"
+echo "built: $HERE/$DEB_OUT"
@@ -0,0 +1,14 @@
+aish (0.1.0-1) bookworm trixie; urgency=medium
+
+  * Initial release packaged for marfrit overlay repo. Phases 0-10
+    complete (102 closed issues): local llama.cpp + cloud broker
+    routing via hossenfelder, MCP tool calls with confirm-gate and
+    per-tool auto_approve, Chuck Norris autonomous mode with
+    destructive-op heuristic, cross-session memory.jsonl, multi-model
+    routing + GBNF grammar passthrough, project file-tree context,
+    cost/usage observability, /tokenize endpoint integration, project
+    overlay (.aish.lua + sha256-pinned trust ledger), cloud preplanner
+    → local executor split.
+  * Source-of-truth: git.reauktion.de/marfrit/aish, tagged v0.1.0.
+
+ -- Markus Fritsche <mfritsche@reauktion.de>  Mon, 25 May 2026 00:00:00 +0000
@@ -0,0 +1,20 @@
+Source: aish
+Section: shells
+Priority: optional
+Maintainer: Markus Fritsche <mfritsche@reauktion.de>
+Standards-Version: 4.6.2
+Homepage: https://git.reauktion.de/marfrit/aish
+
+Package: aish
+Architecture: all
+Depends: ${misc:Depends}, luajit, libreadline8t64 | libreadline8, libcurl4t64 | libcurl4
+Description: AI-augmented conversational shell (LuaJIT, FFI-only)
+ aish is an interactive REPL that interleaves shell execution and language-
+ model conversation against llama.cpp HTTP brokers. Implementation is pure
+ LuaJIT 2.x with FFI bindings to libcurl, GNU readline, and libc — no C
+ extensions, no build step.
+ .
+ Modules install under /usr/share/lua/5.1/aish/. The launcher is
+ /usr/bin/aish. Example configuration is at
+ /usr/share/doc/aish/examples/config.lua (copy to ~/.config/aish/config.lua
+ and adapt).
@@ -0,0 +1,30 @@
+Format: https://www.debian.org/doc/packaging-manuals/copyright-format/1.0/
+Upstream-Name: aish
+Source: https://git.reauktion.de/marfrit/aish
+
+Files: *
+Copyright: 2026 Markus Fritsche <mfritsche@reauktion.de>
+License: MIT
+ Permission is hereby granted, free of charge, to any person obtaining a copy
+ of this software and associated documentation files (the "Software"), to deal
+ in the Software without restriction, including without limitation the rights
+ to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ copies of the Software, and to permit persons to whom the Software is
+ furnished to do so, subject to the following conditions:
+ .
+ The above copyright notice and this permission notice shall be included in
+ all copies or substantial portions of the Software.
+ .
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND.
+
+Files: vendor/dkjson.lua
+Copyright: 2010-2014 David Heiko Kolf
+License: MIT
+ Permission is hereby granted, free of charge, to any person obtaining a copy
+ of this software and associated documentation files (the "Software"), to deal
+ in the Software without restriction, including the rights to use, copy,
+ modify, merge, publish, distribute, sublicense, and/or sell copies of the
+ Software, and to permit persons to whom the Software is furnished to do so,
+ subject to the following conditions: the above copyright notice and this
+ permission notice shall be included in all copies or substantial portions of
+ the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND.
@@ -0,0 +1,150 @@
+#!/bin/bash
+# Package pre-built chromium-fourier artifacts into a .deb.
+#
+# Chromium can't be compiled natively on any available aarch64 runner
+# (clang version wall — chromium requires its internal clang fork).
+# The build is cross-compiled on CT 220 (data, x86_64 Ryzen 7).
+# This script expects the build artifacts to exist at BUILD_DIR
+# (default: fetched from CT 220 via SSH).
+#
+# Sibling Arch package: ../../arch/chromium-fourier/PKGBUILD
+set -euo pipefail
+
+PKGVER=148.0.7778.178
+EPOCH=1
+PKGREL=1
+ARCH=arm64
+
+HERE=$(dirname "$(readlink -f "$0")")
+export SOURCE_DATE_EPOCH=1779854400  # 2026-05-24 09:00 UTC
+
+BUILD_DIR="${BUILD_DIR:-}"
+
+work=$(mktemp -d)
+trap "rm -rf $work" EXIT
+
+if [ -z "$BUILD_DIR" ]; then
+    echo "BUILD_DIR not set — fetching artifacts from CT 220 on data..."
+    BUILD_DIR="$work/artifacts"
+    mkdir -p "$BUILD_DIR"
+    ssh root@data "pct exec 220 -- tar -cf - -C /build/chromium/src/out/Default \
+        chrome chrome_crashpad_handler \
+        libEGL.so libGLESv2.so libvk_swiftshader.so libvulkan.so.1 \
+        vk_swiftshader_icd.json \
+        chrome_100_percent.pak chrome_200_percent.pak resources.pak \
+        v8_context_snapshot.bin snapshot_blob.bin icudtl.dat \
+        locales/" | tar -xf - -C "$BUILD_DIR"
+fi
+
+ROOT="$work/pkgroot"
+
+install -Dm755 "$BUILD_DIR/chrome"                   "$ROOT/usr/lib/chromium/chromium"
+install -Dm755 "$BUILD_DIR/chrome_crashpad_handler"  "$ROOT/usr/lib/chromium/chrome_crashpad_handler"
+
+for so in libEGL.so libGLESv2.so libvk_swiftshader.so libvulkan.so.1; do
+    [ -f "$BUILD_DIR/$so" ] && install -Dm755 "$BUILD_DIR/$so" "$ROOT/usr/lib/chromium/$so"
+done
+
+for icd in "$BUILD_DIR"/*_icd.json; do
+    [ -f "$icd" ] && install -Dm644 "$icd" "$ROOT/usr/lib/chromium/$(basename "$icd")"
+done
+
+for f in chrome_100_percent.pak chrome_200_percent.pak resources.pak \
+         v8_context_snapshot.bin snapshot_blob.bin icudtl.dat; do
+    [ -f "$BUILD_DIR/$f" ] && install -Dm644 "$BUILD_DIR/$f" "$ROOT/usr/lib/chromium/$f"
+done
+
+if [ -d "$BUILD_DIR/locales" ]; then
+    install -dm755 "$ROOT/usr/lib/chromium/locales"
+    cp -r "$BUILD_DIR/locales/"* "$ROOT/usr/lib/chromium/locales/"
+fi
+
+install -dm755 "$ROOT/usr/bin"
+cat > "$ROOT/usr/bin/chromium-fourier" <<'LAUNCHER'
+#!/bin/bash
+USER_HANDLES_VULKAN=0
+for arg in "$@"; do
+  case "$arg" in
+    --use-vulkan*|--enable-features=*Vulkan*|--disable-features=*Vulkan*|--use-angle=vulkan*)
+      USER_HANDLES_VULKAN=1
+      break
+      ;;
+  esac
+done
+
+vulkan_default=()
+if [ "$USER_HANDLES_VULKAN" = 0 ]; then
+  vulkan_default=(--disable-features=Vulkan)
+fi
+
+exec /usr/lib/chromium/chromium \
+  --ozone-platform=wayland \
+  --use-gl=angle --use-angle=gles \
+  --enable-features=AcceleratedVideoDecoder \
+  "${vulkan_default[@]}" \
+  "$@"
+LAUNCHER
+chmod 0755 "$ROOT/usr/bin/chromium-fourier"
+
+mkdir -p "$ROOT/usr/share/doc/chromium-fourier" "$ROOT/DEBIAN"
+install -Dm644 "$HERE/debian/copyright" \
+    "$ROOT/usr/share/doc/chromium-fourier/copyright"
+install -Dm644 "$HERE/debian/changelog" \
+    "$ROOT/usr/share/doc/chromium-fourier/changelog.Debian"
+gzip -9 -n "$ROOT/usr/share/doc/chromium-fourier/changelog.Debian"
+
+ISIZE=$(du -sk "$ROOT" | awk '{print $1}')
+cat > "$ROOT/DEBIAN/control" <<EOF
+Package: chromium-fourier
+Version: ${EPOCH}:${PKGVER}-${PKGREL}
+Section: web
+Priority: optional
+Architecture: ${ARCH}
+Installed-Size: ${ISIZE}
+Depends: libasound2,
+         libatk-bridge2.0-0,
+         libatk1.0-0,
+         libcairo2,
+         libcups2,
+         libdbus-1-3,
+         libdrm2,
+         libexpat1,
+         libfontconfig1,
+         libfreetype6,
+         libgbm1,
+         libglib2.0-0,
+         libgtk-3-0,
+         libnspr4,
+         libnss3,
+         libpango-1.0-0,
+         libpulse0,
+         libva2,
+         libwayland-client0,
+         libx11-6,
+         libxcb1,
+         libxkbcommon0,
+         libpipewire-0.3-0,
+         fonts-liberation,
+         v4l-utils
+Provides: www-browser
+Conflicts: chromium
+Maintainer: Markus Fritsche <mfritsche@reauktion.de>
+Homepage: https://www.chromium.org/
+Description: Chromium with V4L2 HW video decode for Rockchip (Wayland + mainline)
+ Chromium ${PKGVER} with three patches enabling V4L2 hardware video
+ decoding on mainline Linux / Wayland for Rockchip SoCs (RK3566 hantro,
+ RK3588 VDPU381).
+ .
+ Cross-compiled from x86_64 using chromium's bundled clang (upstream
+ LLVM cannot compile chromium). Runtime target is aarch64.
+ .
+ Patches: enable-v4l2-decoder-default, wayland-allow-direct-egl-gles2,
+ nv12-external-oes-on-modifier-external-only.
+ .
+ Launcher at /usr/bin/chromium-fourier defaults to Wayland + ANGLE/GLES
+ with Vulkan disabled (panvk on RK3566 breaks V4L2 dispatch).
+EOF
+
+DEB_OUT="chromium-fourier_${EPOCH}%3a${PKGVER}-${PKGREL}_${ARCH}.deb"
+dpkg-deb --root-owner-group --build "$ROOT" "$HERE/$DEB_OUT"
+echo "built: $HERE/$DEB_OUT"
@@ -0,0 +1,8 @@
+chromium-fourier (1:148.0.7778.178-1) trixie; urgency=medium
+
+  * Chromium 148.0.7778.178 with V4L2 HW decode patches for Rockchip.
+  * Cross-compiled from x86_64 using chromium's bundled clang.
+  * Three fourier patches: enable-v4l2-decoder-default,
+    wayland-allow-direct-egl-gles2, nv12-external-oes-on-modifier-external-only.
+
+ -- Markus Fritsche <mfritsche@reauktion.de>  Sat, 24 May 2026 09:00:00 +0200
@@ -0,0 +1,32 @@
+Format: https://www.debian.org/doc/packaging-manuals/copyright-format/1.0/
+Upstream-Name: Chromium
+Upstream-Contact: chromium-dev@chromium.org
+Source: https://www.chromium.org/
+
+Files: *
+Copyright: The Chromium Authors
+License: BSD-3-Clause
+
+Files: debian/*
+Copyright: 2026 Markus Fritsche <mfritsche@reauktion.de>
+License: BSD-3-Clause
+
+License: BSD-3-Clause
+ Redistribution and use in source and binary forms, with or without
+ modification, are permitted provided that the following conditions are met:
+ .
+ 1. Redistributions of source code must retain the above copyright notice,
+    this list of conditions and the following disclaimer.
+ .
+ 2. Redistributions in binary form must reproduce the above copyright
+    notice, this list of conditions and the following disclaimer in the
+    documentation and/or other materials provided with the distribution.
+ .
+ 3. Neither the name of the copyright holder nor the names of its
+    contributors may be used to endorse or promote products derived from
+    this software without specific prior written permission.
+ .
+ THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ ARE DISCLAIMED.
@@ -0,0 +1,92 @@
+From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
+From: claude-noether <claude-noether@noreply.localhost>
+Date: Sun, 25 May 2026 12:00:00 +0200
+Subject: [PATCH] avcodec/aarch64/h264dsp: route H.264 luma-h deblock through daedalus-fourier
+
+Sibling of 0005 (which substituted v_loop_filter_luma).  Same
+NEON-to-NEON substitution: H264DSPContext.h_loop_filter_luma →
+daedalus_recipe_dispatch_h264_deblock_luma_h.  The H kernel landed
+in daedalus-fourier PR #9 (CPU NEON only — no QPU shader yet).
+
+libavcodec.so ctx is no-QPU per the existing 0003-0005 / 0007
+pattern; we cannot assume Vulkan in arbitrary host processes
+(firefox-fourier RDD, mpv-fourier, etc.).
+
+Intra (bS=4) h_loop_filter_luma_intra stays on the in-tree NEON .S
+code; daedalus_h264_deblock_meta only covers the non-intra path.
+An intra-h substitution can land once daedalus-fourier exposes a
+dispatch helper (the kernel already exists internally per PR #11).
+
+Refs reauktion/daedalus-v4l2#11 — substitution arc step 2 cycle 8 H.
+---
+diff --git a/libavcodec/aarch64/h264_idct_daedalus.c b/libavcodec/aarch64/h264_idct_daedalus.c
+--- a/libavcodec/aarch64/h264_idct_daedalus.c	2026-05-25 13:09:33.694760715 +0200
+++ libavcodec/aarch64/h264_idct_daedalus.c	2026-05-25 13:09:33.715603719 +0200
+@@ -1,9 +1,10 @@
+ /*
+- * H.264 4x4 / 8x8 IDCT + luma-v deblock — daedalus-fourier substitution shims.
+ * H.264 4x4 / 8x8 IDCT + luma v/h deblock — daedalus-fourier substitution shims.
+  *
+  * Routes H264DSPContext.idct_add           → daedalus_recipe_dispatch_h264_idct4
+  *        H264DSPContext.idct8_add          → daedalus_recipe_dispatch_h264_idct8
+  *        H264DSPContext.v_loop_filter_luma → daedalus_recipe_dispatch_h264_deblock_luma_v
+ *        H264DSPContext.h_loop_filter_luma → daedalus_recipe_dispatch_h264_deblock_luma_h
+  * instead of the in-tree ff_h264_*_neon assembly.  The recipe layer
+  * picks the substrate (CPU NEON for cycles 6 + 7 by default; cycle 8
+  * is CPU primary with QPU opportunistic — the ctx below is no-QPU,
+@@ -45,6 +46,8 @@
+ void ff_h264_idct8_add_daedalus(uint8_t *dst, int16_t *block, int stride);
+ void ff_h264_v_loop_filter_luma_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                          int alpha, int beta, int8_t *tc0);
+void ff_h264_h_loop_filter_luma_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                         int alpha, int beta, int8_t *tc0);
+ 
+ void ff_h264_idct_add_daedalus(uint8_t *dst, int16_t *block, int stride)
+ {
+@@ -84,3 +87,22 @@
+     daedalus_recipe_dispatch_h264_deblock_luma_v(g_dctx, pix, (size_t)stride,
+                                                  1, &meta);
+ }
+
+void ff_h264_h_loop_filter_luma_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                         int alpha, int beta, int8_t *tc0)
+{
+    daedalus_h264_deblock_meta meta = {
+        .dst_off = 0,
+        .alpha   = alpha,
+        .beta    = beta,
+    };
+    meta.tc0[0] = tc0[0];
+    meta.tc0[1] = tc0[1];
+    meta.tc0[2] = tc0[2];
+    meta.tc0[3] = tc0[3];
+
+    pthread_once(&g_dctx_once, daedalus_ctx_init_once);
+
+    daedalus_recipe_dispatch_h264_deblock_luma_h(g_dctx, pix, (size_t)stride,
+                                                 1, &meta);
+}
+diff --git a/libavcodec/aarch64/h264dsp_init_aarch64.c b/libavcodec/aarch64/h264dsp_init_aarch64.c
+--- a/libavcodec/aarch64/h264dsp_init_aarch64.c	2026-05-25 13:09:33.695937103 +0200
+++ libavcodec/aarch64/h264dsp_init_aarch64.c	2026-05-25 13:09:33.715541700 +0200
+@@ -31,6 +31,8 @@
+                                          int alpha, int beta, int8_t *tc0);
+ void ff_h264_h_loop_filter_luma_neon(uint8_t *pix, ptrdiff_t stride, int alpha,
+                                      int beta, int8_t *tc0);
+void ff_h264_h_loop_filter_luma_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                         int alpha, int beta, int8_t *tc0);
+ void ff_h264_v_loop_filter_luma_intra_neon(uint8_t *pix, ptrdiff_t stride, int alpha,
+                                            int beta);
+ void ff_h264_h_loop_filter_luma_intra_neon(uint8_t *pix, ptrdiff_t stride, int alpha,
+@@ -117,7 +119,7 @@
+ 
+     if (have_neon(cpu_flags) && bit_depth == 8) {
+         c->v_loop_filter_luma   = ff_h264_v_loop_filter_luma_daedalus;
+-        c->h_loop_filter_luma   = ff_h264_h_loop_filter_luma_neon;
+        c->h_loop_filter_luma   = ff_h264_h_loop_filter_luma_daedalus;
+         c->v_loop_filter_luma_intra= ff_h264_v_loop_filter_luma_intra_neon;
+         c->h_loop_filter_luma_intra= ff_h264_h_loop_filter_luma_intra_neon;
+ 
+--
+2.47.3
+
@@ -0,0 +1,127 @@
+From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
+From: claude-noether <claude-noether@noreply.localhost>
+Date: Sun, 25 May 2026 12:00:00 +0200
+Subject: [PATCH] avcodec/aarch64/h264dsp: route H.264 chroma v/h deblock through daedalus-fourier
+
+Chroma siblings of 0005 (luma_v) and 0008 (luma_h).  Same
+NEON-to-NEON pattern via the daedalus recipe layer:
+
+  H264DSPContext.v_loop_filter_chroma →
+    daedalus_recipe_dispatch_h264_deblock_chroma_v
+  H264DSPContext.h_loop_filter_chroma →
+    daedalus_recipe_dispatch_h264_deblock_chroma_h
+
+Both kernels landed in daedalus-fourier PR #10.  Recipe table
+routes AUTO to CPU NEON (no chroma QPU shaders yet), so this
+is plumbing-only and stays bit-exact against the in-tree NEON.
+
+Intra chroma (bS=4) loop filters remain on in-tree NEON;
+daedalus_h264_deblock_meta covers the non-intra (bS<4) path.
+
+Refs reauktion/daedalus-v4l2#11 — substitution arc step 2 cycle 8 chroma.
+---
+diff --git a/libavcodec/aarch64/h264_idct_daedalus.c b/libavcodec/aarch64/h264_idct_daedalus.c
+--- a/libavcodec/aarch64/h264_idct_daedalus.c	2026-05-25 13:15:45.995368233 +0200
+++ libavcodec/aarch64/h264_idct_daedalus.c	2026-05-25 13:15:46.015839177 +0200
+@@ -1,10 +1,12 @@
+ /*
+- * H.264 4x4 / 8x8 IDCT + luma v/h deblock — daedalus-fourier substitution shims.
+ * H.264 4x4 / 8x8 IDCT + luma v/h + chroma v/h deblock — daedalus-fourier substitution shims.
+  *
+  * Routes H264DSPContext.idct_add           → daedalus_recipe_dispatch_h264_idct4
+  *        H264DSPContext.idct8_add          → daedalus_recipe_dispatch_h264_idct8
+- *        H264DSPContext.v_loop_filter_luma → daedalus_recipe_dispatch_h264_deblock_luma_v
+- *        H264DSPContext.h_loop_filter_luma → daedalus_recipe_dispatch_h264_deblock_luma_h
+ *        H264DSPContext.v_loop_filter_luma   → daedalus_recipe_dispatch_h264_deblock_luma_v
+ *        H264DSPContext.h_loop_filter_luma   → daedalus_recipe_dispatch_h264_deblock_luma_h
+ *        H264DSPContext.v_loop_filter_chroma → daedalus_recipe_dispatch_h264_deblock_chroma_v
+ *        H264DSPContext.h_loop_filter_chroma → daedalus_recipe_dispatch_h264_deblock_chroma_h
+  * instead of the in-tree ff_h264_*_neon assembly.  The recipe layer
+  * picks the substrate (CPU NEON for cycles 6 + 7 by default; cycle 8
+  * is CPU primary with QPU opportunistic — the ctx below is no-QPU,
+@@ -48,6 +50,10 @@
+                                          int alpha, int beta, int8_t *tc0);
+ void ff_h264_h_loop_filter_luma_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                          int alpha, int beta, int8_t *tc0);
+void ff_h264_v_loop_filter_chroma_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                           int alpha, int beta, int8_t *tc0);
+void ff_h264_h_loop_filter_chroma_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                           int alpha, int beta, int8_t *tc0);
+ 
+ void ff_h264_idct_add_daedalus(uint8_t *dst, int16_t *block, int stride)
+ {
+@@ -106,3 +112,41 @@
+     daedalus_recipe_dispatch_h264_deblock_luma_h(g_dctx, pix, (size_t)stride,
+                                                  1, &meta);
+ }
+
+void ff_h264_v_loop_filter_chroma_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                           int alpha, int beta, int8_t *tc0)
+{
+    daedalus_h264_deblock_meta meta = {
+        .dst_off = 0,
+        .alpha   = alpha,
+        .beta    = beta,
+    };
+    meta.tc0[0] = tc0[0];
+    meta.tc0[1] = tc0[1];
+    meta.tc0[2] = tc0[2];
+    meta.tc0[3] = tc0[3];
+
+    pthread_once(&g_dctx_once, daedalus_ctx_init_once);
+
+    daedalus_recipe_dispatch_h264_deblock_chroma_v(g_dctx, pix, (size_t)stride,
+                                                   1, &meta);
+}
+
+void ff_h264_h_loop_filter_chroma_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                           int alpha, int beta, int8_t *tc0)
+{
+    daedalus_h264_deblock_meta meta = {
+        .dst_off = 0,
+        .alpha   = alpha,
+        .beta    = beta,
+    };
+    meta.tc0[0] = tc0[0];
+    meta.tc0[1] = tc0[1];
+    meta.tc0[2] = tc0[2];
+    meta.tc0[3] = tc0[3];
+
+    pthread_once(&g_dctx_once, daedalus_ctx_init_once);
+
+    daedalus_recipe_dispatch_h264_deblock_chroma_h(g_dctx, pix, (size_t)stride,
+                                                   1, &meta);
+}
+diff --git a/libavcodec/aarch64/h264dsp_init_aarch64.c b/libavcodec/aarch64/h264dsp_init_aarch64.c
+--- a/libavcodec/aarch64/h264dsp_init_aarch64.c	2026-05-25 13:15:45.996482360 +0200
+++ libavcodec/aarch64/h264dsp_init_aarch64.c	2026-05-25 13:15:46.025604910 +0200
+@@ -39,8 +39,12 @@
+                                            int beta);
+ void ff_h264_v_loop_filter_chroma_neon(uint8_t *pix, ptrdiff_t stride, int alpha,
+                                        int beta, int8_t *tc0);
+void ff_h264_v_loop_filter_chroma_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                           int alpha, int beta, int8_t *tc0);
+ void ff_h264_h_loop_filter_chroma_neon(uint8_t *pix, ptrdiff_t stride, int alpha,
+                                        int beta, int8_t *tc0);
+void ff_h264_h_loop_filter_chroma_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                           int alpha, int beta, int8_t *tc0);
+ void ff_h264_h_loop_filter_chroma422_neon(uint8_t *pix, ptrdiff_t stride, int alpha,
+                                           int beta, int8_t *tc0);
+ void ff_h264_v_loop_filter_chroma_intra_neon(uint8_t *pix, ptrdiff_t stride,
+@@ -123,11 +127,11 @@
+         c->v_loop_filter_luma_intra= ff_h264_v_loop_filter_luma_intra_neon;
+         c->h_loop_filter_luma_intra= ff_h264_h_loop_filter_luma_intra_neon;
+ 
+-        c->v_loop_filter_chroma = ff_h264_v_loop_filter_chroma_neon;
+        c->v_loop_filter_chroma = ff_h264_v_loop_filter_chroma_daedalus;
+         c->v_loop_filter_chroma_intra = ff_h264_v_loop_filter_chroma_intra_neon;
+ 
+         if (chroma_format_idc <= 1) {
+-            c->h_loop_filter_chroma = ff_h264_h_loop_filter_chroma_neon;
+            c->h_loop_filter_chroma = ff_h264_h_loop_filter_chroma_daedalus;
+             c->h_loop_filter_chroma_intra = ff_h264_h_loop_filter_chroma_intra_neon;
+             c->h_loop_filter_chroma_mbaff_intra = ff_h264_h_loop_filter_chroma_mbaff_intra_neon;
+         } else {
+--
+2.47.3
+
@@ -0,0 +1,126 @@
+From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
+From: claude-noether <claude-noether@noreply.localhost>
+Date: Sun, 25 May 2026 12:30:00 +0200
+Subject: [PATCH] avcodec/aarch64/h264dsp: route H.264 luma intra deblock through daedalus-fourier
+
+Adds the bS=4 intra-strength variants of the already-substituted
+luma_v / luma_h deblock (0005, 0008).  Intra MBs and certain
+inter-MB edges (4x4 transform boundaries inside an Intra_NxN
+neighbour) force boundary strength to 4 per H.264 §8.7.2.1.
+
+  H264DSPContext.v_loop_filter_luma_intra →
+    daedalus_recipe_dispatch_h264_deblock_luma_v_intra
+  H264DSPContext.h_loop_filter_luma_intra →
+    daedalus_recipe_dispatch_h264_deblock_luma_h_intra
+
+Both kernels landed in daedalus-fourier PR #11.  Recipe table
+routes AUTO to CPU NEON (no intra QPU shaders yet) — plumbing-
+only NEON-to-NEON via daedalus, bit-exact against the in-tree
+FFmpeg NEON path.
+
+Signature differs from bS<4: no tc0 argument.  The wrapper
+passes daedalus_h264_deblock_meta with alpha/beta set; tc0[] is
+ignored by the intra dispatch (bS=4 hardcodes the strength).
+
+Chroma intra variants are deferred to a follow-up PR because the
+chroma path has a 4:2:0 / 4:2:2 split (chroma_format_idc gating)
+that needs explicit conditional substitution to avoid running
+the 4:2:0-only daedalus dispatch on 4:2:2 chroma.
+
+Refs reauktion/daedalus-v4l2#11 — substitution arc step 2 cycle 8 intra.
+---
+diff --git a/libavcodec/aarch64/h264_idct_daedalus.c b/libavcodec/aarch64/h264_idct_daedalus.c
+--- a/libavcodec/aarch64/h264_idct_daedalus.c	2026-05-25 13:18:54.992244965 +0200
+++ libavcodec/aarch64/h264_idct_daedalus.c	2026-05-25 13:20:12.338122217 +0200
+@@ -1,5 +1,5 @@
+ /*
+- * H.264 4x4 / 8x8 IDCT + luma v/h + chroma v/h deblock — daedalus-fourier substitution shims.
+ * H.264 4x4 / 8x8 IDCT + luma v/h (inter + intra) + chroma v/h deblock — daedalus-fourier substitution shims.
+  *
+  * Routes H264DSPContext.idct_add           → daedalus_recipe_dispatch_h264_idct4
+  *        H264DSPContext.idct8_add          → daedalus_recipe_dispatch_h264_idct8
+@@ -7,6 +7,8 @@
+  *        H264DSPContext.h_loop_filter_luma   → daedalus_recipe_dispatch_h264_deblock_luma_h
+  *        H264DSPContext.v_loop_filter_chroma → daedalus_recipe_dispatch_h264_deblock_chroma_v
+  *        H264DSPContext.h_loop_filter_chroma → daedalus_recipe_dispatch_h264_deblock_chroma_h
+ *        H264DSPContext.v_loop_filter_luma_intra → daedalus_recipe_dispatch_h264_deblock_luma_v_intra
+ *        H264DSPContext.h_loop_filter_luma_intra → daedalus_recipe_dispatch_h264_deblock_luma_h_intra
+  * instead of the in-tree ff_h264_*_neon assembly.  The recipe layer
+  * picks the substrate (CPU NEON for cycles 6 + 7 by default; cycle 8
+  * is CPU primary with QPU opportunistic — the ctx below is no-QPU,
+@@ -54,6 +56,10 @@
+                                            int alpha, int beta, int8_t *tc0);
+ void ff_h264_h_loop_filter_chroma_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                            int alpha, int beta, int8_t *tc0);
+void ff_h264_v_loop_filter_luma_intra_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                                int alpha, int beta);
+void ff_h264_h_loop_filter_luma_intra_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                                int alpha, int beta);
+ 
+ void ff_h264_idct_add_daedalus(uint8_t *dst, int16_t *block, int stride)
+ {
+@@ -150,3 +156,34 @@
+     daedalus_recipe_dispatch_h264_deblock_chroma_h(g_dctx, pix, (size_t)stride,
+                                                    1, &meta);
+ }
+
+void ff_h264_v_loop_filter_luma_intra_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                                int alpha, int beta)
+{
+    daedalus_h264_deblock_meta meta = {
+        .dst_off = 0,
+        .alpha   = alpha,
+        .beta    = beta,
+    };
+    /* tc0[] is ignored by the intra-strength dispatch (bS=4 hardcodes the strength). */
+
+    pthread_once(&g_dctx_once, daedalus_ctx_init_once);
+
+    daedalus_recipe_dispatch_h264_deblock_luma_v_intra(g_dctx, pix, (size_t)stride,
+                                                        1, &meta);
+}
+
+void ff_h264_h_loop_filter_luma_intra_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                                int alpha, int beta)
+{
+    daedalus_h264_deblock_meta meta = {
+        .dst_off = 0,
+        .alpha   = alpha,
+        .beta    = beta,
+    };
+
+    pthread_once(&g_dctx_once, daedalus_ctx_init_once);
+
+    daedalus_recipe_dispatch_h264_deblock_luma_h_intra(g_dctx, pix, (size_t)stride,
+                                                        1, &meta);
+}
+diff --git a/libavcodec/aarch64/h264dsp_init_aarch64.c b/libavcodec/aarch64/h264dsp_init_aarch64.c
+--- a/libavcodec/aarch64/h264dsp_init_aarch64.c	2026-05-25 13:18:54.993349573 +0200
+++ libavcodec/aarch64/h264dsp_init_aarch64.c	2026-05-25 13:20:12.338265830 +0200
+@@ -35,8 +35,12 @@
+                                          int alpha, int beta, int8_t *tc0);
+ void ff_h264_v_loop_filter_luma_intra_neon(uint8_t *pix, ptrdiff_t stride, int alpha,
+                                            int beta);
+void ff_h264_v_loop_filter_luma_intra_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                                int alpha, int beta);
+ void ff_h264_h_loop_filter_luma_intra_neon(uint8_t *pix, ptrdiff_t stride, int alpha,
+                                            int beta);
+void ff_h264_h_loop_filter_luma_intra_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                                int alpha, int beta);
+ void ff_h264_v_loop_filter_chroma_neon(uint8_t *pix, ptrdiff_t stride, int alpha,
+                                        int beta, int8_t *tc0);
+ void ff_h264_v_loop_filter_chroma_daedalus(uint8_t *pix, ptrdiff_t stride,
+@@ -124,8 +128,8 @@
+     if (have_neon(cpu_flags) && bit_depth == 8) {
+         c->v_loop_filter_luma   = ff_h264_v_loop_filter_luma_daedalus;
+         c->h_loop_filter_luma   = ff_h264_h_loop_filter_luma_daedalus;
+-        c->v_loop_filter_luma_intra= ff_h264_v_loop_filter_luma_intra_neon;
+-        c->h_loop_filter_luma_intra= ff_h264_h_loop_filter_luma_intra_neon;
+        c->v_loop_filter_luma_intra= ff_h264_v_loop_filter_luma_intra_daedalus;
+        c->h_loop_filter_luma_intra= ff_h264_h_loop_filter_luma_intra_daedalus;
+ 
+         c->v_loop_filter_chroma = ff_h264_v_loop_filter_chroma_daedalus;
+         c->v_loop_filter_chroma_intra = ff_h264_v_loop_filter_chroma_intra_neon;
+--
+2.47.3
+
@@ -0,0 +1,101 @@
+From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
+From: claude-noether <claude-noether@noreply.localhost>
+Date: Sun, 25 May 2026 13:00:00 +0200
+Subject: [PATCH] avcodec/aarch64/h264dsp: route H.264 chroma DC Hadamard through daedalus-fourier
+
+Substitutes H264DSPContext.chroma_dc_dequant_idct in the
+4:2:0 / bit_depth=8 init path with a wrapper that composes
+the daedalus chroma DC Hadamard primitive (fourier PR #25)
+with qmul scaling FFmpeg does in one fused function.
+
+Bit-exact against ff_h264_chroma_dc_dequant_idct_8_c.
+Hadamard correctness gated by fourier PR #23 test suite.
+
+4:2:2 chroma stays on the in-tree 422 variant (same
+gating shape as 0009 chroma deblock substitution).
+
+Requires daedalus-fourier commit b9f9ff2 or later (PR #25
+exposing the public Hadamard symbol).  Pin bumps in PKGBUILD
+and build-deb.sh come in the same commit.
+---
+diff --git a/libavcodec/aarch64/h264_idct_daedalus.c b/libavcodec/aarch64/h264_idct_daedalus.c
+--- a/libavcodec/aarch64/h264_idct_daedalus.c	2026-05-25 13:38:32.019491484 +0200
+++ libavcodec/aarch64/h264_idct_daedalus.c	2026-05-25 13:38:32.033821507 +0200
+@@ -1,5 +1,5 @@
+ /*
+- * H.264 4x4 / 8x8 IDCT + luma v/h (inter + intra) + chroma v/h deblock — daedalus-fourier substitution shims.
+ * H.264 4x4 / 8x8 IDCT + luma v/h (inter+intra) + chroma v/h deblock + chroma DC Hadamard — daedalus-fourier substitution shims.
+  *
+  * Routes H264DSPContext.idct_add           → daedalus_recipe_dispatch_h264_idct4
+  *        H264DSPContext.idct8_add          → daedalus_recipe_dispatch_h264_idct8
+@@ -9,6 +9,7 @@
+  *        H264DSPContext.h_loop_filter_chroma → daedalus_recipe_dispatch_h264_deblock_chroma_h
+  *        H264DSPContext.v_loop_filter_luma_intra → daedalus_recipe_dispatch_h264_deblock_luma_v_intra
+  *        H264DSPContext.h_loop_filter_luma_intra → daedalus_recipe_dispatch_h264_deblock_luma_h_intra
+ *        H264DSPContext.chroma_dc_dequant_idct   → daedalus_h264_chroma_dc_hadamard_2x2 + caller-side qmul
+  * instead of the in-tree ff_h264_*_neon assembly.  The recipe layer
+  * picks the substrate (CPU NEON for cycles 6 + 7 by default; cycle 8
+  * is CPU primary with QPU opportunistic — the ctx below is no-QPU,
+@@ -60,6 +61,7 @@
+                                                 int alpha, int beta);
+ void ff_h264_h_loop_filter_luma_intra_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                                 int alpha, int beta);
+void ff_h264_chroma_dc_dequant_idct_daedalus(int16_t *block, int qmul);
+ 
+ void ff_h264_idct_add_daedalus(uint8_t *dst, int16_t *block, int stride)
+ {
+@@ -187,3 +189,32 @@
+     daedalus_recipe_dispatch_h264_deblock_luma_h_intra(g_dctx, pix, (size_t)stride,
+                                                         1, &meta);
+ }
+
+/* Composes daedalus_h264_chroma_dc_hadamard_2x2 with the qmul scaling
+ * that FFmpeg's reference does in one fused function (h264idct_template.c
+ * ff_h264_chroma_dc_dequant_idct).
+ *
+ * The 4 DC coefficients are scattered across the per-MB coefficient
+ * buffer at offsets [r*stride + c*xStride] (stride=32, xStride=16).
+ * Extract into a contiguous int16[4], run the Hadamard, then apply
+ * the qmul scale and write back to the original positions.
+ *
+ * No daedalus ctx needed; the Hadamard is a pure stateless primitive.
+ */
+void ff_h264_chroma_dc_dequant_idct_daedalus(int16_t *block, int qmul)
+{
+    enum { stride = 32, xStride = 16 };
+    int16_t dc[4];
+
+    dc[0] = block[stride*0 + xStride*0];
+    dc[1] = block[stride*0 + xStride*1];
+    dc[2] = block[stride*1 + xStride*0];
+    dc[3] = block[stride*1 + xStride*1];
+
+    daedalus_h264_chroma_dc_hadamard_2x2(dc);
+
+    block[stride*0 + xStride*0] = (int16_t)((int)dc[0] * qmul >> 7);
+    block[stride*0 + xStride*1] = (int16_t)((int)dc[1] * qmul >> 7);
+    block[stride*1 + xStride*0] = (int16_t)((int)dc[2] * qmul >> 7);
+    block[stride*1 + xStride*1] = (int16_t)((int)dc[3] * qmul >> 7);
+}
+diff --git a/libavcodec/aarch64/h264dsp_init_aarch64.c b/libavcodec/aarch64/h264dsp_init_aarch64.c
+--- a/libavcodec/aarch64/h264dsp_init_aarch64.c	2026-05-25 13:38:32.020346459 +0200
+++ libavcodec/aarch64/h264dsp_init_aarch64.c	2026-05-25 13:38:32.033909804 +0200
+@@ -41,6 +41,7 @@
+                                            int beta);
+ void ff_h264_h_loop_filter_luma_intra_daedalus(uint8_t *pix, ptrdiff_t stride,
+                                                 int alpha, int beta);
+void ff_h264_chroma_dc_dequant_idct_daedalus(int16_t *block, int qmul);
+ void ff_h264_v_loop_filter_chroma_neon(uint8_t *pix, ptrdiff_t stride, int alpha,
+                                        int beta, int8_t *tc0);
+ void ff_h264_v_loop_filter_chroma_daedalus(uint8_t *pix, ptrdiff_t stride,
+@@ -135,6 +136,7 @@
+         c->v_loop_filter_chroma_intra = ff_h264_v_loop_filter_chroma_intra_neon;
+ 
+         if (chroma_format_idc <= 1) {
+            c->chroma_dc_dequant_idct = ff_h264_chroma_dc_dequant_idct_daedalus;
+             c->h_loop_filter_chroma = ff_h264_h_loop_filter_chroma_daedalus;
+             c->h_loop_filter_chroma_intra = ff_h264_h_loop_filter_chroma_intra_neon;
+             c->h_loop_filter_chroma_mbaff_intra = ff_h264_h_loop_filter_chroma_mbaff_intra_neon;
+--
+2.47.3
+
@@ -45,7 +45,7 @@ PKGREL=10  # pkgrel=10 — H.264 luma qpel mc20 daedalus-fourier substitution
 # DAEDALUS_KERNEL_H264_QPEL_MC20.  Cycle 9 plumbs the last H.264 NEON
 # kernel through the recipe layer.  Daemon-side build (debian/daedalus-v4l2)
 # can bump in a follow-up; this PR only changes the libavcodec.so consumer.
-DAEDALUS_FOURIER_COMMIT=209a4218bcb98b91c04f07ad61513bb04adb13ad
+DAEDALUS_FOURIER_COMMIT=b9f9ff2a89c068aea54dcb52b543afddad28311e  # PR #25 — public chroma DC Hadamard

 HERE=$(dirname "$(readlink -f "$0")")

@@ -74,6 +74,10 @@ patch -Np1 -i "$HERE/0004-h264-idct8-daedalus-fourier.patch"
 patch -Np1 -i "$HERE/0005-h264-deblock-luma-v-daedalus-fourier.patch"
 patch -Np1 -i "$HERE/0006-h264-restore-low-delay.patch"
 patch -Np1 -i "$HERE/0007-h264-qpel-mc20-daedalus-fourier.patch"
+patch -Np1 -i "$HERE/0008-h264-deblock-luma-h-daedalus-fourier.patch"
+patch -Np1 -i "$HERE/0009-h264-deblock-chroma-daedalus-fourier.patch"
+patch -Np1 -i "$HERE/0010-h264-deblock-luma-intra-daedalus-fourier.patch"
+patch -Np1 -i "$HERE/0011-h264-chroma-dc-hadamard-daedalus-fourier.patch"

 # --- daedalus-fourier: fetch + build static .a with PIC, install to a
 # per-build prefix; libavcodec.so links it into the shared object so
Author	SHA1	Message	Date
claude-noether	d8aa3aae8d	ffmpeg-v4l2-request-fourier: route H.264 chroma DC Hadamard through daedalus-fourier (0011) Substitutes H264DSPContext.chroma_dc_dequant_idct in the 4:2:0 / bit_depth=8 init path with a wrapper that composes the daedalus chroma DC Hadamard primitive (daedalus-fourier PR #25) with the qmul scaling FFmpeg's reference does in one fused function (h264idct_template.c::ff_h264_chroma_dc_dequant_idct). Algorithm per H.264 §8.5.11.1 / §8.5.11.2: 1. Extract 4 DCs from the scattered positions in the per-MB coefficient buffer (stride=32, xStride=16) 2. 2x2 Hadamard transform (daedalus primitive) 3. qmul scale + >> 7, write back to original positions Bit-exact against ff_h264_chroma_dc_dequant_idct_8_c. The Hadamard itself is gated by the fourier PR #23 7-case test suite (including the H·H = 4·I algebraic invariant), and the public-API parity test added in PR #25 confirms the src/ symbol matches the test ref. 4:2:2 chroma stays on the in-tree ff_h264_chroma422_dc_dequant_idct_c path — same chroma_format_idc<=1 gating shape as 0009 chroma deblock. Pin bump: _daedalus_fourier_commit / DAEDALUS_FOURIER_COMMIT bumped to b9f9ff2a (post-PR #25) so the build picks up the public daedalus_h264_chroma_dc_hadamard_2x2 symbol. Verified the patch applies cleanly on top of 0001-0010 against the pinned upstream commit b57fbbe5 on hertz.	2026-05-25 13:39:54 +02:00
marfrit	1f58ff2b6b	Merge pull request 'ffmpeg-v4l2-request-fourier: route H.264 luma intra deblock through daedalus-fourier (0010)' (#99 ) from claude-noether/marfrit-packages:noether/h264-substitute-deblock-intra into main Reviewed-on: marfrit/marfrit-packages#99	2026-05-25 11:28:33 +00:00
claude-noether	45be17fbdf	ffmpeg-v4l2-request-fourier: route H.264 luma intra deblock through daedalus-fourier (0010) Adds the bS=4 intra-strength variants of the already-substituted luma_v / luma_h deblock (0005, 0008). Intra MBs and certain inter-MB edges (4x4 transform boundaries inside an Intra_NxN neighbour) force boundary strength to 4 per H.264 §8.7.2.1. H264DSPContext.v_loop_filter_luma_intra → daedalus_recipe_dispatch_h264_deblock_luma_v_intra H264DSPContext.h_loop_filter_luma_intra → daedalus_recipe_dispatch_h264_deblock_luma_h_intra Both kernels landed in daedalus-fourier PR #11. Recipe → CPU NEON (no intra QPU shaders yet); plumbing-only NEON-to-NEON via daedalus. Signature differs from bS<4: no tc0 argument. Wrapper passes daedalus_h264_deblock_meta with alpha/beta set; tc0[] is ignored by the intra dispatch (bS=4 hardcodes the strength). Chroma intra variants are deferred to a follow-up because the chroma init has a 4:2:0 / 4:2:2 split (chroma_format_idc gating) — the daedalus dispatch is 4:2:0-only and needs explicit conditional substitution to avoid running on 4:2:2 chroma. Verified the patch applies cleanly on top of 0001-0009 against the pinned upstream commit b57fbbe5 on hertz.	2026-05-25 13:21:00 +02:00
marfrit	7b9bb9b2d0	Merge pull request 'ffmpeg-v4l2-request-fourier: route H.264 chroma v/h deblock through daedalus-fourier (0009)' (#98 ) from claude-noether/marfrit-packages:noether/h264-substitute-deblock-chroma into main Reviewed-on: marfrit/marfrit-packages#98	2026-05-25 11:18:15 +00:00
claude-noether	babb280410	ffmpeg-v4l2-request-fourier: route H.264 chroma v/h deblock through daedalus-fourier (0009) Chroma siblings of 0005 (luma_v) and 0008 (luma_h). Same NEON-to-NEON pattern via the daedalus recipe layer: H264DSPContext.v_loop_filter_chroma → daedalus_recipe_dispatch_h264_deblock_chroma_v H264DSPContext.h_loop_filter_chroma → daedalus_recipe_dispatch_h264_deblock_chroma_h Both kernels landed in daedalus-fourier PR #10. Recipe table routes AUTO to CPU NEON (no chroma QPU shaders yet), so this is plumbing- only and stays bit-exact against the in-tree NEON. Intra chroma (bS=4) loop filters remain on in-tree NEON; daedalus_h264_deblock_meta covers the non-intra (bS<4) path. Verified the patch applies cleanly on top of 0001-0008 against the pinned upstream commit b57fbbe5 on hertz. Wires the new patch into both arch/PKGBUILD and debian/build-deb.sh.	2026-05-25 13:16:45 +02:00
marfrit	5b48d1c743	Merge pull request 'ffmpeg-v4l2-request-fourier: route H.264 luma-h deblock through daedalus-fourier (0008)' (#97 ) from claude-noether/marfrit-packages:noether/h264-substitute-deblock-luma-h into main Reviewed-on: marfrit/marfrit-packages#97	2026-05-25 11:14:26 +00:00
claude-noether	624f83e877	ffmpeg-v4l2-request-fourier: route H.264 luma-h deblock through daedalus-fourier (0008) Adds patch 0008 to the substitution arc, mirroring 0005's V variant for H.264 non-intra bS<4 horizontal luma deblock. H264DSPContext.h_loop_filter_luma → daedalus_recipe_dispatch_h264_deblock_luma_h The H kernel was added to daedalus-fourier in PR #9 (vendored ff_h264_h_loop_filter_luma_neon, wired through the same CPU-dispatch pattern as V). Recipe table routes AUTO to CPU NEON (no QPU shader for H yet), so this is a NEON-to-NEON substitution via the daedalus recipe layer — same shape as 0005. The libavcodec.so ctx remains no-QPU (daedalus_ctx_create_no_qpu), matching the existing 0003/0004/0005/0007 patches. Higher-cycle QPU init waits for a feature-flag gating change in a separate PR. Intra (bS=4) h_loop_filter_luma_intra stays on the in-tree NEON .S code; daedalus_h264_deblock_meta covers the non-intra path only. A follow-up can route intra once daedalus-fourier exposes the intra-h dispatch (the kernel already exists internally per fourier PR #11). Wires the new patch into both arch/PKGBUILD and debian/build-deb.sh sequences. Verified the patch applies cleanly on top of 0001-0007 against the pinned upstream commit b57fbbe5 on hertz.	2026-05-25 13:10:05 +02:00
marfrit	902de73a02	Merge pull request 'mesa-panvk-bifrost r7: fix XFB store channel-extract for packed varyings (iter19)' (#96 ) from claude-noether/marfrit-packages:mesa-panvk-bifrost-r7 into main Reviewed-on: marfrit/marfrit-packages#96	2026-05-25 09:22:47 +00:00
marfrit	c14c22f942	mesa-panvk-bifrost r7: fix XFB store channel-extract for packed varyings (iter19) Adds 0007-panvk-bifrost-xfb-component-base-fix.patch — eliminates a reliable SIGSEGV in vkCreateGraphicsPipeline whenever an XFB-bound vertex output is declared with non-zero `layout (component=N)`. Surfaced by dEQP-VK.transform_feedback.simple.holes_vert (Mali-G52 r1 MC1, PAN_ARCH 7). Backtrace lands 11 frames into libvulkan_panfrost.so called from vkt::TransformFeedback::TransformFeedbackHolesInstance:: iterate. Root cause: iter17's lower_xfb_output_iter17 (and upstream pan_nir_lower_xfb, which has the identical `// TODO`) computes the source-channel mask as `mask << channel_idx`, where channel_idx is the varying-location component (0..3) but src only contains channels starting at nir_intrinsic_component(intr). For a scalar declared component=2, the lowering computed `mask << 2` against a 1-component src — out-of-range; the malformed nir_def segfaulted in downstream NIR constant-folding (nir_constant_expressions.c::evaluate_). Fix translates channel_idx to source-channel space by subtracting nir_intrinsic_component(intr) before shifting the mask, and replaces the elided release-mode asserts with explicit release-mode guards (closes the same elision class as the original bug). Verified on PineTab2 (Mali-G52 r1 MC1, PAN_ARCH 7) against vulkan-cts 1.3.10.0: - holes_vert / holes_extra_draw_vert no longer SIGSEGV (now Fail on color-check; that is a separate iter20 finding). - basic_: 36/36 Pass. depth_clip_: 1 Pass + 4 NotSupported. lines_or_triangles: 16 NotSupported. 0 Fail across the set. Caveat (not regressions): max_output_components_64/_128/_256 were never reached on the r5 sweep — watchdog killed transform_feedback after the holes_vert crash. With this fix in place, they now run and surface their own pre-existing coredumps, confirmed on shipped r6 baseline too. iter20+ territory. Phase 5 (2nd-model) review: APPROVE WITH CHANGES (non-blocking). Changes applied: release-mode defensive guards on both preconditions plus a dispatcher-side comment clarifying the i2+j semantics. Cross-refs: - ~/src/panvk-bifrost/iter19/phase{0,1,2,3}_holes_vert.md Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>	2026-05-25 11:16:56 +02:00
marfrit	b113e053f0	Merge pull request 'aish: package v0.1.0 for arch + debian' (#95 ) from claude-noether/marfrit-packages:noether/aish-v0.1.0-package into main Reviewed-on: marfrit/marfrit-packages#95	2026-05-24 22:40:53 +00:00
marfrit	8ec4c57ad7	aish: package v0.1.0 for arch + debian aish is an AI-augmented conversational shell in LuaJIT 2.x with FFI bindings to libcurl, GNU readline, and libc — no C extensions, no build step. Source-of-truth: git.reauktion.de/marfrit/aish, tag v0.1.0 (tarball sha256 9ebc3939e028832e39391ae33efacb5ec9bcd99d123cbc8ca1cd6ca9a640b5b5). The arch and debian recipes mirror the lmcp pattern (pure-Lua any-arch package, no makefile, install copies modules directly): arch/aish/PKGBUILD — depends=(luajit readline curl) debian/aish/build-deb.sh — pure dpkg-deb, SOURCE_DATE_EPOCH pinned debian/aish/debian/{control,changelog,copyright} Install layout, matching what main.lua's script-dir-relative package.path expects after the wrapper execs `luajit /usr/share/lua/5.1/aish/main.lua`: /usr/bin/aish ← bin/aish wrapper /usr/share/lua/5.1/aish/{main,broker,context,executor,history, mcp,renderer,repl,router,safety,secrets}.lua /usr/share/lua/5.1/aish/ffi/{curl,libc,pty,readline}.lua /usr/share/lua/5.1/aish/vendor/dkjson.lua /usr/share/doc/aish/{README.md,LICENSE,examples/config.lua} CI: two new jobs in .gitea/workflows/build.yml at the end of file. aish-any chains needs:lmcp-debian (parallel-DAG with claude-his-any, serialized via the shared arch-aarch64 runner — avoids needless wait through the unrelated fourier stack). aish-debian chains needs:aish-any. Both invoke the standard check-already-published.sh fast-skip on no- change pushes. Sonnet review (per feedback_reviews_use_sonnet.md + bugfix-process step 4): no blockers. Folded in two findings before commit: switched needs: from mpv-fourier-aarch64 to lmcp-debian (cleaner DAG, faster cold-build wall clock), removed the dead Build-Depends: debhelper- compat line from debian/aish/debian/control (build-deb.sh doesn't use debhelper). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>	2026-05-25 00:35:01 +02:00
marfrit	beb09c4863	Merge pull request 'mesa-panvk-bifrost: r5 -> r6 — advertise VK_EXT_legacy_dithering on Bifrost' (#94 ) from claude-noether/marfrit-packages:noether/mesa-panvk-bifrost-r6-legacy-dithering into main Reviewed-on: marfrit/marfrit-packages#94	2026-05-24 22:28:58 +00:00
marfrit	7708744eb2	mesa-panvk-bifrost: r5 -> r6 — advertise VK_EXT_legacy_dithering on Bifrost Backports Mesa main's unconditional flip of VK_EXT_legacy_dithering. Pure-software composition; no new HW path. vk_render_pass already gates on enabled_features.legacyDithering and panvk_vX_blend + pan_format already plumb the dithered BLEND descriptor (BFMT2 table has MALI_BLEND_AU encodings for RGB565/RGB5A1/RGBA4/RGB10A2 on PAN_ARCH 7). Our r5 base just hadn't picked up the cherry-pick. Phase 7 verify on ohm (PineTab2 / RK3566 / Mali-G52 r1 MC1) with a locally-built r6 lib at /tmp/r6_test_lib/: Feature advertisement: r5: VK_EXT_legacy_dithering not in extension list, legacyDithering=false r6: VK_EXT_legacy_dithering rev 2 advertised, legacyDithering=true dEQP subsets (delta): dEQP-VK.api..legacy_dithering r5/r6 both: 2 P / 0 F (identical) dEQP-VK.renderpass.dithering.* r5/r6 both: 0 P / 0 F / 94 NS (identical) dEQP-VK.renderpass2.dithering.* r5/r6 both: 0 P / 0 F / 94 NS (identical) Net: zero regressions, advertisement-only delta as expected. Second-model review (per bugfix-process step 4) traced the full code path through vk_render_pass + panvk_vX_cmd_draw + panvk_vX_blend + pan_format BFMT2. No interaction with our r1 nullDescriptor (disjoint paths). Mesa upstream marks ext DONE for panvk in docs/features.txt. ARM's own libmali r51p0 driver (BXODROIDN2PL, 2024-08) lists VK_EXT_legacy_dithering in its Vulkan extension string table, confirming the feature is shipped by ARM for Mali-G52-class hardware. Follow-up out of scope: the 94 renderpass-dithering tests show as NotSupported on both r5 and r6 — there's a separate panvk-side prereq the dEQP harness checks (likely a specific format-feature combination). Worth investigating in a future iteration. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>	2026-05-25 00:23:40 +02:00
marfrit	5d97cf15d6	Merge pull request 'chromium-fourier: bump to 148.0.7778.178, add Debian recipe' (#93 ) from claude-noether/marfrit-packages:noether/chromium-fourier-148 into main Reviewed-on: marfrit/marfrit-packages#93	2026-05-24 19:46:13 +00:00
marfrit	58f67d4b2c	chromium-fourier: bump to 148.0.7778.178, add Debian recipe Arch: version 147→148, drop enable_nacl (removed upstream), fix nv12-external-oes patch context for 148 (base/numerics/safe_conversions.h include removed upstream). Header comment updated: native build fiction → cross-compile reality. Debian: new build-deb.sh that assembles .deb from pre-built artifacts on CT 220 (data). Same binary artifacts as the Arch package, launcher at /usr/bin/chromium-fourier (no Conflicts with stock chromium on Debian). Both packages published to packages.reauktion.de: - Arch: marfrit/aarch64/chromium-fourier 1:148.0.7778.178-1 - Debian: trixie+bookworm/main/arm64 chromium-fourier 1:148.0.7778.178-1 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-05-24 09:45:30 +02:00
marfrit	685f85c22e	Merge pull request 'mesa-panvk-bifrost: r4 -> r5 — advertise fragmentStoresAndAtomics on Bifrost (closes panvk-bifrost#2)' (#92 ) from claude-noether/marfrit-packages:noether/mesa-panvk-bifrost-r5 into main Reviewed-on: marfrit/marfrit-packages#92	2026-05-23 12:19:10 +00:00
marfrit	6896853544	mesa-panvk-bifrost: r4 -> r5 — advertise fragmentStoresAndAtomics on Bifrost Backports Mesa main's unconditional flip of .fragmentStoresAndAtomics to true in src/panfrost/vulkan/panvk_vX_physical_device.c. Closes the Dawn-WebGPU adapter rejection at PhysicalDeviceVk.cpp:250 that caused brave-vulkan to fall back to the SwiftShader CPU adapter on PineTab2/Mali-G52, per marfrit/panvk-bifrost#2. Phase 7 verify on ohm (PineTab2, RK3566, Mali-G52 r1 MC1) with a locally-built r5 lib installed to /tmp/r5_test_lib/: dEQP-VK.glsl.atomic_operations.: r4: 48 pass / 0 fail / 992 NotSupported (1040 total) r5: 80 pass / 0 fail / 960 NotSupported (1040 total) delta: +32 newly-passing, zero new failures dEQP-VK.image.store.: r4: 2772 pass / 0 fail / 238 NotSupported (3010 total) r5: 2772 pass / 0 fail / 238 NotSupported (3010 total) delta: identical (image.store is independent of the flag) The disjunction with instance->force_enable_shader_atomics is kept as a documented kill-switch even though the compiler folds it away — it leaves the DRI option pan_force_enable_shader_atomics semantically wired for future rebases or downstream debugging. Patch reviewed via 2nd-model pass (per bugfix-process step 4): recommended keeping the disjunction (applied), Bifrost-only-vs-unconditional left unconditional to match upstream (applied), pre-ship CTS subset (applied with results above). Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>	2026-05-23 14:04:44 +02:00
marfrit	fd56eca3cb	Merge pull request 'mesa-panvk-bifrost{,-video}: fix url= to real Gitea repo' (#91 ) from claude-noether/marfrit-packages:noether/panvk-bifrost-url-fix into main Reviewed-on: marfrit/marfrit-packages#91	2026-05-23 03:18:37 +00:00