marfrit/marfrit-packages
0
0
Fork 1
Code Issues 2 Pull Requests Actions Packages Projects Releases Wiki Activity

ffmpeg-v4l2-request-fourier: route remaining H.264 qpel 8x8 positions through daedalus-fourier (0012) #101

Merged
marfrit merged 1 commits from claude-noether/marfrit-packages:noether/h264-substitute-qpel-rest into main 2026-05-25 12:19:56 +00:00
Conversation 0 Commits 1 Files Changed 4
+495 -2
4 changed files with 495 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
arch/ffmpeg-v4l2-request-fourier/0012-h264-qpel-rest-daedalus-fourier.patch
+245
View File
@@ -0,0 +1,245 @@
From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
From: claude-noether <claude-noether@noreply.localhost>
Date: Sun, 25 May 2026 14:00:00 +0200
Subject: [PATCH] avcodec/aarch64/h264qpel: route remaining qpel 8x8 positions through daedalus-fourier
Closes the H.264 qpel substitution. Extends 0007 (which routed only
mc20 put_) to ALL 15 useful positions in BOTH the put_ and avg_
tables, skipping mc00 (integer copy / pointer-only fast path).
29 substitutions total: 14 new put_ + 15 avg_. Each is a uniform
wrapper around daedalus_recipe_dispatch_h264_qpel_{avg_,}mcXY exposed
by daedalus-fourier PRs #15-#20.
All recipe-table entries route AUTO to CPU NEON (no QPU shaders
for any qpel position other than mc20 yet), so this is plumbing-only
NEON-to-NEON — bit-exact against the in-tree ff_*_h264_qpel8_*_neon
path.
16x16 qpel tables ([0][...]) stay on the in-tree NEON. daedalus
only exposes 8x8 today; 16x16 substitution can land once fourier
provides those variants (likely just dispatching the 8x8 path four
times with shifted dst/src offsets).
Refs reauktion/daedalus-v4l2#11 — substitution arc qpel buildout.
---
diff --git a/libavcodec/aarch64/h264_qpel_daedalus.c b/libavcodec/aarch64/h264_qpel_daedalus.c
--- a/libavcodec/aarch64/h264_qpel_daedalus.c 2026-05-25 14:05:05.789298250 +0200
+++ libavcodec/aarch64/h264_qpel_daedalus.c 2026-05-25 14:05:05.818358374 +0200
@@ -1,10 +1,13 @@
/*
- * H.264 luma qpel mc20 (8x8, horizontal half-pel, 6-tap "put")
- * — daedalus-fourier substitution shim.
+ * H.264 luma qpel 8x8 — daedalus-fourier substitution shims (put_ + avg_).
*
- * Routes H264QpelContext.put_h264_qpel_pixels_tab[1][2] through
- * daedalus_recipe_dispatch_h264_qpel_mc20 instead of
- * ff_put_h264_qpel8_mc20_neon. The recipe layer picks the substrate
+ * Routes ALL 15 useful positions in H264QpelContext's 8x8 put_ and
+ * avg_ tables through daedalus_recipe_dispatch_h264_qpel_mc{XY}
+ * (skipping mc00 which is integer copy / FFmpeg's pointer-only fast
+ * path). Plumbing-only NEON-by-recipe — daedalus-fourier PRs #15-#20
+ * exposed each variant via the same dispatch signature, so the
+ * substitution is a uniform macro across put_/avg_ and across all
+ * 15 mc positions. The recipe layer picks the substrate
* (CPU NEON for cycle 9; QPU not viable — per-block 7.6 ns vs
* ~250 ns QPU dispatch floor, see docs/k9_h264qpel_mc20.md).
*
@@ -48,3 +51,53 @@
daedalus_recipe_dispatch_h264_qpel_mc20(g_dctx, dst, src, (size_t)stride,
1, &meta);
}
+
+
+/* All other 8x8 qpel positions follow the same dispatch shape as mc20
+ * above. The macro collapses ~600 LOC of one-wrapper-per-variant
+ * boilerplate (29 variants total: 14 put_ + 15 avg_). */
+#define DEFINE_QPEL_WRAPPER(type, suffix, dispatch_fn) \
+void ff_ ## type ## _h264_qpel8_ ## suffix ## _daedalus(uint8_t *dst, \
+ const uint8_t *src, ptrdiff_t stride); \
+void ff_ ## type ## _h264_qpel8_ ## suffix ## _daedalus(uint8_t *dst, \
+ const uint8_t *src, ptrdiff_t stride) \
+{ \
+ static const daedalus_h264_qpel_meta meta = { .dst_off = 0, .src_off = 0 }; \
+ pthread_once(&g_dctx_once, daedalus_ctx_init_once); \
+ dispatch_fn(g_dctx, dst, src, (size_t)stride, 1, &meta); \
+}
+
+/* put_ variants (mc20 stays on the explicit definition above). */
+DEFINE_QPEL_WRAPPER(put, mc10, daedalus_recipe_dispatch_h264_qpel_mc10)
+DEFINE_QPEL_WRAPPER(put, mc30, daedalus_recipe_dispatch_h264_qpel_mc30)
+DEFINE_QPEL_WRAPPER(put, mc01, daedalus_recipe_dispatch_h264_qpel_mc01)
+DEFINE_QPEL_WRAPPER(put, mc11, daedalus_recipe_dispatch_h264_qpel_mc11)
+DEFINE_QPEL_WRAPPER(put, mc21, daedalus_recipe_dispatch_h264_qpel_mc21)
+DEFINE_QPEL_WRAPPER(put, mc31, daedalus_recipe_dispatch_h264_qpel_mc31)
+DEFINE_QPEL_WRAPPER(put, mc02, daedalus_recipe_dispatch_h264_qpel_mc02)
+DEFINE_QPEL_WRAPPER(put, mc12, daedalus_recipe_dispatch_h264_qpel_mc12)
+DEFINE_QPEL_WRAPPER(put, mc22, daedalus_recipe_dispatch_h264_qpel_mc22)
+DEFINE_QPEL_WRAPPER(put, mc32, daedalus_recipe_dispatch_h264_qpel_mc32)
+DEFINE_QPEL_WRAPPER(put, mc03, daedalus_recipe_dispatch_h264_qpel_mc03)
+DEFINE_QPEL_WRAPPER(put, mc13, daedalus_recipe_dispatch_h264_qpel_mc13)
+DEFINE_QPEL_WRAPPER(put, mc23, daedalus_recipe_dispatch_h264_qpel_mc23)
+DEFINE_QPEL_WRAPPER(put, mc33, daedalus_recipe_dispatch_h264_qpel_mc33)
+
+/* avg_ variants — all 15 useful positions. */
+DEFINE_QPEL_WRAPPER(avg, mc10, daedalus_recipe_dispatch_h264_qpel_avg_mc10)
+DEFINE_QPEL_WRAPPER(avg, mc20, daedalus_recipe_dispatch_h264_qpel_avg_mc20)
+DEFINE_QPEL_WRAPPER(avg, mc30, daedalus_recipe_dispatch_h264_qpel_avg_mc30)
+DEFINE_QPEL_WRAPPER(avg, mc01, daedalus_recipe_dispatch_h264_qpel_avg_mc01)
+DEFINE_QPEL_WRAPPER(avg, mc11, daedalus_recipe_dispatch_h264_qpel_avg_mc11)
+DEFINE_QPEL_WRAPPER(avg, mc21, daedalus_recipe_dispatch_h264_qpel_avg_mc21)
+DEFINE_QPEL_WRAPPER(avg, mc31, daedalus_recipe_dispatch_h264_qpel_avg_mc31)
+DEFINE_QPEL_WRAPPER(avg, mc02, daedalus_recipe_dispatch_h264_qpel_avg_mc02)
+DEFINE_QPEL_WRAPPER(avg, mc12, daedalus_recipe_dispatch_h264_qpel_avg_mc12)
+DEFINE_QPEL_WRAPPER(avg, mc22, daedalus_recipe_dispatch_h264_qpel_avg_mc22)
+DEFINE_QPEL_WRAPPER(avg, mc32, daedalus_recipe_dispatch_h264_qpel_avg_mc32)
+DEFINE_QPEL_WRAPPER(avg, mc03, daedalus_recipe_dispatch_h264_qpel_avg_mc03)
+DEFINE_QPEL_WRAPPER(avg, mc13, daedalus_recipe_dispatch_h264_qpel_avg_mc13)
+DEFINE_QPEL_WRAPPER(avg, mc23, daedalus_recipe_dispatch_h264_qpel_avg_mc23)
+DEFINE_QPEL_WRAPPER(avg, mc33, daedalus_recipe_dispatch_h264_qpel_avg_mc33)
+
+#undef DEFINE_QPEL_WRAPPER
diff --git a/libavcodec/aarch64/h264qpel_init_aarch64.c b/libavcodec/aarch64/h264qpel_init_aarch64.c
--- a/libavcodec/aarch64/h264qpel_init_aarch64.c 2026-05-25 14:05:05.790403989 +0200
+++ libavcodec/aarch64/h264qpel_init_aarch64.c 2026-05-25 14:05:05.819136071 +0200
@@ -50,6 +50,64 @@
void ff_put_h264_qpel8_mc30_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
void ff_put_h264_qpel8_mc20_daedalus(uint8_t *dst, const uint8_t *src,
ptrdiff_t stride);
+void ff_put_h264_qpel8_mc10_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_put_h264_qpel8_mc30_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_put_h264_qpel8_mc01_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_put_h264_qpel8_mc11_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_put_h264_qpel8_mc21_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_put_h264_qpel8_mc31_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_put_h264_qpel8_mc02_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_put_h264_qpel8_mc12_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_put_h264_qpel8_mc22_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_put_h264_qpel8_mc32_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_put_h264_qpel8_mc03_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_put_h264_qpel8_mc13_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_put_h264_qpel8_mc23_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_put_h264_qpel8_mc33_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc10_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc20_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc30_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc01_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc11_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc21_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc31_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc02_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc12_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc22_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc32_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc03_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc13_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc23_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc33_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
void ff_put_h264_qpel8_mc01_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
void ff_put_h264_qpel8_mc11_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
void ff_put_h264_qpel8_mc21_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
@@ -164,21 +222,21 @@
c->put_h264_qpel_pixels_tab[0][15] = ff_put_h264_qpel16_mc33_neon;
c->put_h264_qpel_pixels_tab[1][ 0] = ff_put_h264_qpel8_mc00_neon;
- c->put_h264_qpel_pixels_tab[1][ 1] = ff_put_h264_qpel8_mc10_neon;
+ c->put_h264_qpel_pixels_tab[1][ 1] = ff_put_h264_qpel8_mc10_daedalus;
c->put_h264_qpel_pixels_tab[1][ 2] = ff_put_h264_qpel8_mc20_daedalus;
- c->put_h264_qpel_pixels_tab[1][ 3] = ff_put_h264_qpel8_mc30_neon;
- c->put_h264_qpel_pixels_tab[1][ 4] = ff_put_h264_qpel8_mc01_neon;
- c->put_h264_qpel_pixels_tab[1][ 5] = ff_put_h264_qpel8_mc11_neon;
- c->put_h264_qpel_pixels_tab[1][ 6] = ff_put_h264_qpel8_mc21_neon;
- c->put_h264_qpel_pixels_tab[1][ 7] = ff_put_h264_qpel8_mc31_neon;
- c->put_h264_qpel_pixels_tab[1][ 8] = ff_put_h264_qpel8_mc02_neon;
- c->put_h264_qpel_pixels_tab[1][ 9] = ff_put_h264_qpel8_mc12_neon;
- c->put_h264_qpel_pixels_tab[1][10] = ff_put_h264_qpel8_mc22_neon;
- c->put_h264_qpel_pixels_tab[1][11] = ff_put_h264_qpel8_mc32_neon;
- c->put_h264_qpel_pixels_tab[1][12] = ff_put_h264_qpel8_mc03_neon;
- c->put_h264_qpel_pixels_tab[1][13] = ff_put_h264_qpel8_mc13_neon;
- c->put_h264_qpel_pixels_tab[1][14] = ff_put_h264_qpel8_mc23_neon;
- c->put_h264_qpel_pixels_tab[1][15] = ff_put_h264_qpel8_mc33_neon;
+ c->put_h264_qpel_pixels_tab[1][ 3] = ff_put_h264_qpel8_mc30_daedalus;
+ c->put_h264_qpel_pixels_tab[1][ 4] = ff_put_h264_qpel8_mc01_daedalus;
+ c->put_h264_qpel_pixels_tab[1][ 5] = ff_put_h264_qpel8_mc11_daedalus;
+ c->put_h264_qpel_pixels_tab[1][ 6] = ff_put_h264_qpel8_mc21_daedalus;
+ c->put_h264_qpel_pixels_tab[1][ 7] = ff_put_h264_qpel8_mc31_daedalus;
+ c->put_h264_qpel_pixels_tab[1][ 8] = ff_put_h264_qpel8_mc02_daedalus;
+ c->put_h264_qpel_pixels_tab[1][ 9] = ff_put_h264_qpel8_mc12_daedalus;
+ c->put_h264_qpel_pixels_tab[1][10] = ff_put_h264_qpel8_mc22_daedalus;
+ c->put_h264_qpel_pixels_tab[1][11] = ff_put_h264_qpel8_mc32_daedalus;
+ c->put_h264_qpel_pixels_tab[1][12] = ff_put_h264_qpel8_mc03_daedalus;
+ c->put_h264_qpel_pixels_tab[1][13] = ff_put_h264_qpel8_mc13_daedalus;
+ c->put_h264_qpel_pixels_tab[1][14] = ff_put_h264_qpel8_mc23_daedalus;
+ c->put_h264_qpel_pixels_tab[1][15] = ff_put_h264_qpel8_mc33_daedalus;
c->avg_h264_qpel_pixels_tab[0][ 0] = ff_avg_h264_qpel16_mc00_neon;
c->avg_h264_qpel_pixels_tab[0][ 1] = ff_avg_h264_qpel16_mc10_neon;
@@ -198,21 +256,21 @@
c->avg_h264_qpel_pixels_tab[0][15] = ff_avg_h264_qpel16_mc33_neon;
c->avg_h264_qpel_pixels_tab[1][ 0] = ff_avg_h264_qpel8_mc00_neon;
- c->avg_h264_qpel_pixels_tab[1][ 1] = ff_avg_h264_qpel8_mc10_neon;
- c->avg_h264_qpel_pixels_tab[1][ 2] = ff_avg_h264_qpel8_mc20_neon;
- c->avg_h264_qpel_pixels_tab[1][ 3] = ff_avg_h264_qpel8_mc30_neon;
- c->avg_h264_qpel_pixels_tab[1][ 4] = ff_avg_h264_qpel8_mc01_neon;
- c->avg_h264_qpel_pixels_tab[1][ 5] = ff_avg_h264_qpel8_mc11_neon;
- c->avg_h264_qpel_pixels_tab[1][ 6] = ff_avg_h264_qpel8_mc21_neon;
- c->avg_h264_qpel_pixels_tab[1][ 7] = ff_avg_h264_qpel8_mc31_neon;
- c->avg_h264_qpel_pixels_tab[1][ 8] = ff_avg_h264_qpel8_mc02_neon;
- c->avg_h264_qpel_pixels_tab[1][ 9] = ff_avg_h264_qpel8_mc12_neon;
- c->avg_h264_qpel_pixels_tab[1][10] = ff_avg_h264_qpel8_mc22_neon;
- c->avg_h264_qpel_pixels_tab[1][11] = ff_avg_h264_qpel8_mc32_neon;
- c->avg_h264_qpel_pixels_tab[1][12] = ff_avg_h264_qpel8_mc03_neon;
- c->avg_h264_qpel_pixels_tab[1][13] = ff_avg_h264_qpel8_mc13_neon;
- c->avg_h264_qpel_pixels_tab[1][14] = ff_avg_h264_qpel8_mc23_neon;
- c->avg_h264_qpel_pixels_tab[1][15] = ff_avg_h264_qpel8_mc33_neon;
+ c->avg_h264_qpel_pixels_tab[1][ 1] = ff_avg_h264_qpel8_mc10_daedalus;
+ c->avg_h264_qpel_pixels_tab[1][ 2] = ff_avg_h264_qpel8_mc20_daedalus;
+ c->avg_h264_qpel_pixels_tab[1][ 3] = ff_avg_h264_qpel8_mc30_daedalus;
+ c->avg_h264_qpel_pixels_tab[1][ 4] = ff_avg_h264_qpel8_mc01_daedalus;
+ c->avg_h264_qpel_pixels_tab[1][ 5] = ff_avg_h264_qpel8_mc11_daedalus;
+ c->avg_h264_qpel_pixels_tab[1][ 6] = ff_avg_h264_qpel8_mc21_daedalus;
+ c->avg_h264_qpel_pixels_tab[1][ 7] = ff_avg_h264_qpel8_mc31_daedalus;
+ c->avg_h264_qpel_pixels_tab[1][ 8] = ff_avg_h264_qpel8_mc02_daedalus;
+ c->avg_h264_qpel_pixels_tab[1][ 9] = ff_avg_h264_qpel8_mc12_daedalus;
+ c->avg_h264_qpel_pixels_tab[1][10] = ff_avg_h264_qpel8_mc22_daedalus;
+ c->avg_h264_qpel_pixels_tab[1][11] = ff_avg_h264_qpel8_mc32_daedalus;
+ c->avg_h264_qpel_pixels_tab[1][12] = ff_avg_h264_qpel8_mc03_daedalus;
+ c->avg_h264_qpel_pixels_tab[1][13] = ff_avg_h264_qpel8_mc13_daedalus;
+ c->avg_h264_qpel_pixels_tab[1][14] = ff_avg_h264_qpel8_mc23_daedalus;
+ c->avg_h264_qpel_pixels_tab[1][15] = ff_avg_h264_qpel8_mc33_daedalus;
} else if (have_neon(cpu_flags) && bit_depth == 10) {
c->put_h264_qpel_pixels_tab[0][ 1] = ff_put_h264_qpel16_mc10_neon_10;
c->put_h264_qpel_pixels_tab[0][ 2] = ff_put_h264_qpel16_mc20_neon_10;
--
2.47.3
arch/ffmpeg-v4l2-request-fourier/PKGBUILD
+4 -2
View File
@@ -98,8 +98,9 @@ source=("git+https://github.com/Kwiboo/FFmpeg.git#commit=${_commit}"
'0008-h264-deblock-luma-h-daedalus-fourier.patch'
'0009-h264-deblock-chroma-daedalus-fourier.patch'
'0010-h264-deblock-luma-intra-daedalus-fourier.patch'
'0011-h264-chroma-dc-hadamard-daedalus-fourier.patch')
sha256sums=('SKIP' 'SKIP' 'SKIP' 'SKIP' 'SKIP' 'SKIP' 'SKIP' 'SKIP' 'SKIP' 'SKIP' 'SKIP' 'SKIP' 'SKIP')
'0011-h264-chroma-dc-hadamard-daedalus-fourier.patch'
'0012-h264-qpel-rest-daedalus-fourier.patch')
sha256sums=('SKIP' 'SKIP' 'SKIP' 'SKIP' 'SKIP' 'SKIP' 'SKIP' 'SKIP' 'SKIP' 'SKIP' 'SKIP' 'SKIP' 'SKIP' 'SKIP')
pkgver() {
cd "${_srcname}"
@@ -121,6 +122,7 @@ prepare() {
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 "${srcdir}/0011-h264-chroma-dc-hadamard-daedalus-fourier.patch"
patch -Np1 -i "${srcdir}/0012-h264-qpel-rest-daedalus-fourier.patch"
}
build() {
debian/ffmpeg-v4l2-request-fourier/0012-h264-qpel-rest-daedalus-fourier.patch
Vendored
+245
View File
@@ -0,0 +1,245 @@
From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
From: claude-noether <claude-noether@noreply.localhost>
Date: Sun, 25 May 2026 14:00:00 +0200
Subject: [PATCH] avcodec/aarch64/h264qpel: route remaining qpel 8x8 positions through daedalus-fourier
Closes the H.264 qpel substitution. Extends 0007 (which routed only
mc20 put_) to ALL 15 useful positions in BOTH the put_ and avg_
tables, skipping mc00 (integer copy / pointer-only fast path).
29 substitutions total: 14 new put_ + 15 avg_. Each is a uniform
wrapper around daedalus_recipe_dispatch_h264_qpel_{avg_,}mcXY exposed
by daedalus-fourier PRs #15-#20.
All recipe-table entries route AUTO to CPU NEON (no QPU shaders
for any qpel position other than mc20 yet), so this is plumbing-only
NEON-to-NEON — bit-exact against the in-tree ff_*_h264_qpel8_*_neon
path.
16x16 qpel tables ([0][...]) stay on the in-tree NEON. daedalus
only exposes 8x8 today; 16x16 substitution can land once fourier
provides those variants (likely just dispatching the 8x8 path four
times with shifted dst/src offsets).
Refs reauktion/daedalus-v4l2#11 — substitution arc qpel buildout.
---
diff --git a/libavcodec/aarch64/h264_qpel_daedalus.c b/libavcodec/aarch64/h264_qpel_daedalus.c
--- a/libavcodec/aarch64/h264_qpel_daedalus.c 2026-05-25 14:05:05.789298250 +0200
+++ libavcodec/aarch64/h264_qpel_daedalus.c 2026-05-25 14:05:05.818358374 +0200
@@ -1,10 +1,13 @@
/*
- * H.264 luma qpel mc20 (8x8, horizontal half-pel, 6-tap "put")
- * — daedalus-fourier substitution shim.
+ * H.264 luma qpel 8x8 — daedalus-fourier substitution shims (put_ + avg_).
*
- * Routes H264QpelContext.put_h264_qpel_pixels_tab[1][2] through
- * daedalus_recipe_dispatch_h264_qpel_mc20 instead of
- * ff_put_h264_qpel8_mc20_neon. The recipe layer picks the substrate
+ * Routes ALL 15 useful positions in H264QpelContext's 8x8 put_ and
+ * avg_ tables through daedalus_recipe_dispatch_h264_qpel_mc{XY}
+ * (skipping mc00 which is integer copy / FFmpeg's pointer-only fast
+ * path). Plumbing-only NEON-by-recipe — daedalus-fourier PRs #15-#20
+ * exposed each variant via the same dispatch signature, so the
+ * substitution is a uniform macro across put_/avg_ and across all
+ * 15 mc positions. The recipe layer picks the substrate
* (CPU NEON for cycle 9; QPU not viable — per-block 7.6 ns vs
* ~250 ns QPU dispatch floor, see docs/k9_h264qpel_mc20.md).
*
@@ -48,3 +51,53 @@
daedalus_recipe_dispatch_h264_qpel_mc20(g_dctx, dst, src, (size_t)stride,
1, &meta);
}
+
+
+/* All other 8x8 qpel positions follow the same dispatch shape as mc20
+ * above. The macro collapses ~600 LOC of one-wrapper-per-variant
+ * boilerplate (29 variants total: 14 put_ + 15 avg_). */
+#define DEFINE_QPEL_WRAPPER(type, suffix, dispatch_fn) \
+void ff_ ## type ## _h264_qpel8_ ## suffix ## _daedalus(uint8_t *dst, \
+ const uint8_t *src, ptrdiff_t stride); \
+void ff_ ## type ## _h264_qpel8_ ## suffix ## _daedalus(uint8_t *dst, \
+ const uint8_t *src, ptrdiff_t stride) \
+{ \
+ static const daedalus_h264_qpel_meta meta = { .dst_off = 0, .src_off = 0 }; \
+ pthread_once(&g_dctx_once, daedalus_ctx_init_once); \
+ dispatch_fn(g_dctx, dst, src, (size_t)stride, 1, &meta); \
+}
+
+/* put_ variants (mc20 stays on the explicit definition above). */
+DEFINE_QPEL_WRAPPER(put, mc10, daedalus_recipe_dispatch_h264_qpel_mc10)
+DEFINE_QPEL_WRAPPER(put, mc30, daedalus_recipe_dispatch_h264_qpel_mc30)
+DEFINE_QPEL_WRAPPER(put, mc01, daedalus_recipe_dispatch_h264_qpel_mc01)
+DEFINE_QPEL_WRAPPER(put, mc11, daedalus_recipe_dispatch_h264_qpel_mc11)
+DEFINE_QPEL_WRAPPER(put, mc21, daedalus_recipe_dispatch_h264_qpel_mc21)
+DEFINE_QPEL_WRAPPER(put, mc31, daedalus_recipe_dispatch_h264_qpel_mc31)
+DEFINE_QPEL_WRAPPER(put, mc02, daedalus_recipe_dispatch_h264_qpel_mc02)
+DEFINE_QPEL_WRAPPER(put, mc12, daedalus_recipe_dispatch_h264_qpel_mc12)
+DEFINE_QPEL_WRAPPER(put, mc22, daedalus_recipe_dispatch_h264_qpel_mc22)
+DEFINE_QPEL_WRAPPER(put, mc32, daedalus_recipe_dispatch_h264_qpel_mc32)
+DEFINE_QPEL_WRAPPER(put, mc03, daedalus_recipe_dispatch_h264_qpel_mc03)
+DEFINE_QPEL_WRAPPER(put, mc13, daedalus_recipe_dispatch_h264_qpel_mc13)
+DEFINE_QPEL_WRAPPER(put, mc23, daedalus_recipe_dispatch_h264_qpel_mc23)
+DEFINE_QPEL_WRAPPER(put, mc33, daedalus_recipe_dispatch_h264_qpel_mc33)
+
+/* avg_ variants — all 15 useful positions. */
+DEFINE_QPEL_WRAPPER(avg, mc10, daedalus_recipe_dispatch_h264_qpel_avg_mc10)
+DEFINE_QPEL_WRAPPER(avg, mc20, daedalus_recipe_dispatch_h264_qpel_avg_mc20)
+DEFINE_QPEL_WRAPPER(avg, mc30, daedalus_recipe_dispatch_h264_qpel_avg_mc30)
+DEFINE_QPEL_WRAPPER(avg, mc01, daedalus_recipe_dispatch_h264_qpel_avg_mc01)
+DEFINE_QPEL_WRAPPER(avg, mc11, daedalus_recipe_dispatch_h264_qpel_avg_mc11)
+DEFINE_QPEL_WRAPPER(avg, mc21, daedalus_recipe_dispatch_h264_qpel_avg_mc21)
+DEFINE_QPEL_WRAPPER(avg, mc31, daedalus_recipe_dispatch_h264_qpel_avg_mc31)
+DEFINE_QPEL_WRAPPER(avg, mc02, daedalus_recipe_dispatch_h264_qpel_avg_mc02)
+DEFINE_QPEL_WRAPPER(avg, mc12, daedalus_recipe_dispatch_h264_qpel_avg_mc12)
+DEFINE_QPEL_WRAPPER(avg, mc22, daedalus_recipe_dispatch_h264_qpel_avg_mc22)
+DEFINE_QPEL_WRAPPER(avg, mc32, daedalus_recipe_dispatch_h264_qpel_avg_mc32)
+DEFINE_QPEL_WRAPPER(avg, mc03, daedalus_recipe_dispatch_h264_qpel_avg_mc03)
+DEFINE_QPEL_WRAPPER(avg, mc13, daedalus_recipe_dispatch_h264_qpel_avg_mc13)
+DEFINE_QPEL_WRAPPER(avg, mc23, daedalus_recipe_dispatch_h264_qpel_avg_mc23)
+DEFINE_QPEL_WRAPPER(avg, mc33, daedalus_recipe_dispatch_h264_qpel_avg_mc33)
+
+#undef DEFINE_QPEL_WRAPPER
diff --git a/libavcodec/aarch64/h264qpel_init_aarch64.c b/libavcodec/aarch64/h264qpel_init_aarch64.c
--- a/libavcodec/aarch64/h264qpel_init_aarch64.c 2026-05-25 14:05:05.790403989 +0200
+++ libavcodec/aarch64/h264qpel_init_aarch64.c 2026-05-25 14:05:05.819136071 +0200
@@ -50,6 +50,64 @@
void ff_put_h264_qpel8_mc30_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
void ff_put_h264_qpel8_mc20_daedalus(uint8_t *dst, const uint8_t *src,
ptrdiff_t stride);
+void ff_put_h264_qpel8_mc10_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_put_h264_qpel8_mc30_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_put_h264_qpel8_mc01_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_put_h264_qpel8_mc11_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_put_h264_qpel8_mc21_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_put_h264_qpel8_mc31_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_put_h264_qpel8_mc02_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_put_h264_qpel8_mc12_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_put_h264_qpel8_mc22_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_put_h264_qpel8_mc32_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_put_h264_qpel8_mc03_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_put_h264_qpel8_mc13_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_put_h264_qpel8_mc23_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_put_h264_qpel8_mc33_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc10_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc20_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc30_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc01_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc11_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc21_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc31_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc02_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc12_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc22_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc32_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc03_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc13_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc23_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc33_daedalus(uint8_t *dst, const uint8_t *src,
+ ptrdiff_t stride);
void ff_put_h264_qpel8_mc01_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
void ff_put_h264_qpel8_mc11_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
void ff_put_h264_qpel8_mc21_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
@@ -164,21 +222,21 @@
c->put_h264_qpel_pixels_tab[0][15] = ff_put_h264_qpel16_mc33_neon;
c->put_h264_qpel_pixels_tab[1][ 0] = ff_put_h264_qpel8_mc00_neon;
- c->put_h264_qpel_pixels_tab[1][ 1] = ff_put_h264_qpel8_mc10_neon;
+ c->put_h264_qpel_pixels_tab[1][ 1] = ff_put_h264_qpel8_mc10_daedalus;
c->put_h264_qpel_pixels_tab[1][ 2] = ff_put_h264_qpel8_mc20_daedalus;
- c->put_h264_qpel_pixels_tab[1][ 3] = ff_put_h264_qpel8_mc30_neon;
- c->put_h264_qpel_pixels_tab[1][ 4] = ff_put_h264_qpel8_mc01_neon;
- c->put_h264_qpel_pixels_tab[1][ 5] = ff_put_h264_qpel8_mc11_neon;
- c->put_h264_qpel_pixels_tab[1][ 6] = ff_put_h264_qpel8_mc21_neon;
- c->put_h264_qpel_pixels_tab[1][ 7] = ff_put_h264_qpel8_mc31_neon;
- c->put_h264_qpel_pixels_tab[1][ 8] = ff_put_h264_qpel8_mc02_neon;
- c->put_h264_qpel_pixels_tab[1][ 9] = ff_put_h264_qpel8_mc12_neon;
- c->put_h264_qpel_pixels_tab[1][10] = ff_put_h264_qpel8_mc22_neon;
- c->put_h264_qpel_pixels_tab[1][11] = ff_put_h264_qpel8_mc32_neon;
- c->put_h264_qpel_pixels_tab[1][12] = ff_put_h264_qpel8_mc03_neon;
- c->put_h264_qpel_pixels_tab[1][13] = ff_put_h264_qpel8_mc13_neon;
- c->put_h264_qpel_pixels_tab[1][14] = ff_put_h264_qpel8_mc23_neon;
- c->put_h264_qpel_pixels_tab[1][15] = ff_put_h264_qpel8_mc33_neon;
+ c->put_h264_qpel_pixels_tab[1][ 3] = ff_put_h264_qpel8_mc30_daedalus;
+ c->put_h264_qpel_pixels_tab[1][ 4] = ff_put_h264_qpel8_mc01_daedalus;
+ c->put_h264_qpel_pixels_tab[1][ 5] = ff_put_h264_qpel8_mc11_daedalus;
+ c->put_h264_qpel_pixels_tab[1][ 6] = ff_put_h264_qpel8_mc21_daedalus;
+ c->put_h264_qpel_pixels_tab[1][ 7] = ff_put_h264_qpel8_mc31_daedalus;
+ c->put_h264_qpel_pixels_tab[1][ 8] = ff_put_h264_qpel8_mc02_daedalus;
+ c->put_h264_qpel_pixels_tab[1][ 9] = ff_put_h264_qpel8_mc12_daedalus;
+ c->put_h264_qpel_pixels_tab[1][10] = ff_put_h264_qpel8_mc22_daedalus;
+ c->put_h264_qpel_pixels_tab[1][11] = ff_put_h264_qpel8_mc32_daedalus;
+ c->put_h264_qpel_pixels_tab[1][12] = ff_put_h264_qpel8_mc03_daedalus;
+ c->put_h264_qpel_pixels_tab[1][13] = ff_put_h264_qpel8_mc13_daedalus;
+ c->put_h264_qpel_pixels_tab[1][14] = ff_put_h264_qpel8_mc23_daedalus;
+ c->put_h264_qpel_pixels_tab[1][15] = ff_put_h264_qpel8_mc33_daedalus;
c->avg_h264_qpel_pixels_tab[0][ 0] = ff_avg_h264_qpel16_mc00_neon;
c->avg_h264_qpel_pixels_tab[0][ 1] = ff_avg_h264_qpel16_mc10_neon;
@@ -198,21 +256,21 @@
c->avg_h264_qpel_pixels_tab[0][15] = ff_avg_h264_qpel16_mc33_neon;
c->avg_h264_qpel_pixels_tab[1][ 0] = ff_avg_h264_qpel8_mc00_neon;
- c->avg_h264_qpel_pixels_tab[1][ 1] = ff_avg_h264_qpel8_mc10_neon;
- c->avg_h264_qpel_pixels_tab[1][ 2] = ff_avg_h264_qpel8_mc20_neon;
- c->avg_h264_qpel_pixels_tab[1][ 3] = ff_avg_h264_qpel8_mc30_neon;
- c->avg_h264_qpel_pixels_tab[1][ 4] = ff_avg_h264_qpel8_mc01_neon;
- c->avg_h264_qpel_pixels_tab[1][ 5] = ff_avg_h264_qpel8_mc11_neon;
- c->avg_h264_qpel_pixels_tab[1][ 6] = ff_avg_h264_qpel8_mc21_neon;
- c->avg_h264_qpel_pixels_tab[1][ 7] = ff_avg_h264_qpel8_mc31_neon;
- c->avg_h264_qpel_pixels_tab[1][ 8] = ff_avg_h264_qpel8_mc02_neon;
- c->avg_h264_qpel_pixels_tab[1][ 9] = ff_avg_h264_qpel8_mc12_neon;
- c->avg_h264_qpel_pixels_tab[1][10] = ff_avg_h264_qpel8_mc22_neon;
- c->avg_h264_qpel_pixels_tab[1][11] = ff_avg_h264_qpel8_mc32_neon;
- c->avg_h264_qpel_pixels_tab[1][12] = ff_avg_h264_qpel8_mc03_neon;
- c->avg_h264_qpel_pixels_tab[1][13] = ff_avg_h264_qpel8_mc13_neon;
- c->avg_h264_qpel_pixels_tab[1][14] = ff_avg_h264_qpel8_mc23_neon;
- c->avg_h264_qpel_pixels_tab[1][15] = ff_avg_h264_qpel8_mc33_neon;
+ c->avg_h264_qpel_pixels_tab[1][ 1] = ff_avg_h264_qpel8_mc10_daedalus;
+ c->avg_h264_qpel_pixels_tab[1][ 2] = ff_avg_h264_qpel8_mc20_daedalus;
+ c->avg_h264_qpel_pixels_tab[1][ 3] = ff_avg_h264_qpel8_mc30_daedalus;
+ c->avg_h264_qpel_pixels_tab[1][ 4] = ff_avg_h264_qpel8_mc01_daedalus;
+ c->avg_h264_qpel_pixels_tab[1][ 5] = ff_avg_h264_qpel8_mc11_daedalus;
+ c->avg_h264_qpel_pixels_tab[1][ 6] = ff_avg_h264_qpel8_mc21_daedalus;
+ c->avg_h264_qpel_pixels_tab[1][ 7] = ff_avg_h264_qpel8_mc31_daedalus;
+ c->avg_h264_qpel_pixels_tab[1][ 8] = ff_avg_h264_qpel8_mc02_daedalus;
+ c->avg_h264_qpel_pixels_tab[1][ 9] = ff_avg_h264_qpel8_mc12_daedalus;
+ c->avg_h264_qpel_pixels_tab[1][10] = ff_avg_h264_qpel8_mc22_daedalus;
+ c->avg_h264_qpel_pixels_tab[1][11] = ff_avg_h264_qpel8_mc32_daedalus;
+ c->avg_h264_qpel_pixels_tab[1][12] = ff_avg_h264_qpel8_mc03_daedalus;
+ c->avg_h264_qpel_pixels_tab[1][13] = ff_avg_h264_qpel8_mc13_daedalus;
+ c->avg_h264_qpel_pixels_tab[1][14] = ff_avg_h264_qpel8_mc23_daedalus;
+ c->avg_h264_qpel_pixels_tab[1][15] = ff_avg_h264_qpel8_mc33_daedalus;
} else if (have_neon(cpu_flags) && bit_depth == 10) {
c->put_h264_qpel_pixels_tab[0][ 1] = ff_put_h264_qpel16_mc10_neon_10;
c->put_h264_qpel_pixels_tab[0][ 2] = ff_put_h264_qpel16_mc20_neon_10;
--
2.47.3
debian/ffmpeg-v4l2-request-fourier/build-deb.sh
Vendored
+1
View File
@@ -78,6 +78,7 @@ 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"
patch -Np1 -i "$HERE/0012-h264-qpel-rest-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