claude-noether
f4047f3145
Reverts the no_qpu → qpu-capable ctx flip that landed via patch 0014 (marfrit-packages PR #104). PR #104 was justified by daedalus-fourier PR #36's "QPU 4.30x faster than CPU NEON" bench result. That number was a measurement artifact: v3d_runner.read_spv() did a bare cwd-relative fopen() with no path search, so when the bench was run from the source dir (as in PR #36), the SPVs at $builddir/v3d_*.spv were not found, every QPU dispatch returned -1 fast, and the bench loop timed the failure path. daedalus-fourier PR #37 fixes the SPV search + bench preflight. Corrected numbers on hertz (Pi 5 V3D 7.1): kernel CPU ns/op QPU ns/op winner IDCT 4x4 luma 10.75 217.63 CPU 20.24x IDCT 8x8 luma 29.69 785.94 CPU 26.47x Deblock luma_v 17.63 467.42 CPU 26.51x Deblock luma_h 38.30 498.53 CPU 13.02x qpel mc20 (8x8) 30.17 1300.44 CPU 43.10x qpel mc02 (8x8) 17.69 1363.40 CPU 77.08x qpel mc22 (8x8) 71.60 1948.37 CPU 27.21x 1080p sum: CPU 5.57 ms vs QPU 123.54 ms — QPU 22x SLOWER. Until daedalus QPU dispatch overhead is actually competitive (separate multi-task effort tracked on the daedalus-fourier side), libavcodec.so substitution must stay on daedalus_ctx_create_no_qpu() so the host processes (firefox-fourier RDD, mpv-fourier, daedalus_v4l2_daemon) don't pessimize their H.264 decode path. Adds 0015-h264-ctx-revert-to-no-qpu.patch (2-line revert of patch 0014) to both arch PKGBUILD and debian build-deb.sh. Both pkgrel bumped 11 → 12. Refs reauktion/daedalus-fourier!37.
74 lines
3.1 KiB
Diff
74 lines
3.1 KiB
Diff
From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
|
|
From: Markus Fritsche <mfritsche@reauktion.de>
|
|
Date: Mon, 25 May 2026 22:00:00 +0200
|
|
Subject: [PATCH] avcodec/aarch64/h264: revert ctx flip — daedalus-fourier PR
|
|
#36 was a measurement artifact
|
|
MIME-Version: 1.0
|
|
Content-Type: text/plain; charset=UTF-8
|
|
Content-Transfer-Encoding: 8bit
|
|
|
|
Reverts the daedalus_ctx_create_no_qpu() → daedalus_ctx_create() flip
|
|
that landed in 0014-h264-ctx-qpu-capable.patch (marfrit-packages PR
|
|
#104). The flip was justified by daedalus-fourier PR #36 which
|
|
reported a 4.30x QPU-over-CPU win on the 1080p H.264 hot-path sum.
|
|
|
|
That number was a measurement artifact. The bench tool's
|
|
v3d_runner.read_spv() did a bare fopen() that resolved relative to
|
|
cwd; when run from the source directory (as in PR #36), the SPVs at
|
|
$builddir/v3d_*.spv were not found, every QPU dispatch returned -1
|
|
fast, and the loop timed the failure path. Daedalus-fourier PR #37
|
|
fixes the SPV search + bench preflight; corrected numbers from hertz
|
|
(Pi 5 V3D 7.1) show QPU is 12-77x SLOWER than CPU NEON at every
|
|
H.264 hot-path kernel:
|
|
|
|
kernel CPU ns/op QPU ns/op winner
|
|
IDCT 4x4 luma 10.75 217.63 CPU 20.24x
|
|
IDCT 8x8 luma 29.69 785.94 CPU 26.47x
|
|
Deblock luma_v 17.63 467.42 CPU 26.51x
|
|
Deblock luma_h 38.30 498.53 CPU 13.02x
|
|
qpel mc20 (8x8) 30.17 1300.44 CPU 43.10x
|
|
qpel mc02 (8x8) 17.69 1363.40 CPU 77.08x
|
|
qpel mc22 (8x8) 71.60 1948.37 CPU 27.21x
|
|
|
|
1080p sum: CPU 5.57 ms vs QPU 123.54 ms — QPU 22x slower.
|
|
|
|
Until the daedalus QPU dispatch overhead is actually competitive (a
|
|
multi-task effort tracked on the daedalus-fourier side), the
|
|
libavcodec.so substitution must stay on daedalus_ctx_create_no_qpu()
|
|
to avoid pessimizing every host process that loads it
|
|
(firefox-fourier RDD, mpv-fourier, daedalus_v4l2_daemon).
|
|
|
|
Both H.264 TUs (h264_idct_daedalus.c, h264_qpel_daedalus.c) are
|
|
reverted; the change is a 2-line revert of patch 0014.
|
|
|
|
Refs reauktion/daedalus-fourier!37 (the retraction PR).
|
|
---
|
|
libavcodec/aarch64/h264_idct_daedalus.c | 2 +-
|
|
libavcodec/aarch64/h264_qpel_daedalus.c | 2 +-
|
|
2 files changed, 2 insertions(+), 2 deletions(-)
|
|
|
|
diff --git a/libavcodec/aarch64/h264_idct_daedalus.c b/libavcodec/aarch64/h264_idct_daedalus.c
|
|
--- a/libavcodec/aarch64/h264_idct_daedalus.c
|
|
+++ b/libavcodec/aarch64/h264_idct_daedalus.c
|
|
@@ -32,7 +32,7 @@ static pthread_once_t g_dctx_once = PTHREAD_ONCE_INIT;
|
|
|
|
static void daedalus_ctx_init_once(void)
|
|
{
|
|
- g_dctx = daedalus_ctx_create();
|
|
+ g_dctx = daedalus_ctx_create_no_qpu();
|
|
}
|
|
|
|
void ff_h264_idct_add_daedalus(uint8_t *dst, int16_t *block, int stride);
|
|
diff --git a/libavcodec/aarch64/h264_qpel_daedalus.c b/libavcodec/aarch64/h264_qpel_daedalus.c
|
|
--- a/libavcodec/aarch64/h264_qpel_daedalus.c
|
|
+++ b/libavcodec/aarch64/h264_qpel_daedalus.c
|
|
@@ -38,7 +38,7 @@ static pthread_once_t g_dctx_once = PTHREAD_ONCE_INIT;
|
|
|
|
static void daedalus_ctx_init_once(void)
|
|
{
|
|
- g_dctx = daedalus_ctx_create();
|
|
+ g_dctx = daedalus_ctx_create_no_qpu();
|
|
}
|
|
|
|
void ff_put_h264_qpel8_mc20_daedalus(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
|