[FFmpeg-devel] [PATCH] h264: integrate clear_blocks calls with IDCT.
Michael Niedermayer
michaelni at gmx.at
Tue Feb 19 03:33:55 CET 2013
On Mon, Feb 18, 2013 at 04:50:14PM -0800, Ronald S. Bultje wrote:
> From: "Ronald S. Bultje" <rsbultje at gmail.com>
>
> The non-intra-pcm branch in hl_decode_mb (simple, 8bpp) goes from 700
> to 672 cycles, and the complete loop of decode_mb_cabac and hl_decode_mb
> (in the decode_slice loop) goes from 1759 to 1733 cycles on the clip
> tested (cathedral), i.e. almost 30 cycles per mb faster.
> ---
> libavcodec/arm/h264idct_neon.S | 25 ++++++---
> libavcodec/h264.c | 14 ++---
> libavcodec/h264_mb_template.c | 11 +---
> libavcodec/h264addpx_template.c | 4 ++
> libavcodec/h264dsp.c | 4 +-
> libavcodec/h264dsp.h | 4 +-
> libavcodec/h264idct_template.c | 16 ++++--
> libavcodec/h264pred.h | 8 +--
> libavcodec/h264pred_template.c | 28 ++++++----
> libavcodec/ppc/h264_altivec.c | 3 ++
> libavcodec/svq3.c | 4 +-
> libavcodec/x86/h264_idct.asm | 108 ++++++++++++++++++++++++++++---------
> libavcodec/x86/h264_idct_10bit.asm | 53 ++++++++++++++++--
> 13 files changed, 205 insertions(+), 77 deletions(-)
on x86, svq3_watermark.mov still shows artifacts
on arm many h264 tests fail, random example below:
(i didnt spot any segfaults, but the tests didnt finish yet)
--- ffmpeg/tests/ref/fate/h264-conformance-frext-frext_mmco4_sony_b 2013-02-19 01:03:50.269711778 +0100
+++ tests/data/fate/h264-conformance-frext-frext_mmco4_sony_b 2013-02-19 03:25:34.873890947 +0100
@@ -1,61 +1,61 @@
#tb 0: 1/25
-0, 0, 0, 1, 152064, 0x24846654
-0, 1, 1, 1, 152064, 0xd33945be
-0, 2, 2, 1, 152064, 0x961cbbfb
-0, 3, 3, 1, 152064, 0x91eb854d
-0, 4, 4, 1, 152064, 0x998351b4
-0, 5, 5, 1, 152064, 0xe7515c67
-0, 6, 6, 1, 152064, 0xd32058ae
-0, 7, 7, 1, 152064, 0x99cb5695
-0, 8, 8, 1, 152064, 0x281de175
-0, 9, 9, 1, 152064, 0xf4146f55
-0, 10, 10, 1, 152064, 0x63abbb61
-0, 11, 11, 1, 152064, 0xf44550c5
-0, 12, 12, 1, 152064, 0x5797a5c2
-0, 13, 13, 1, 152064, 0x13b15c60
-0, 14, 14, 1, 152064, 0x6521e682
-0, 15, 15, 1, 152064, 0x780c32a4
-0, 16, 16, 1, 152064, 0x6289f37c
-0, 17, 17, 1, 152064, 0x96d64d81
-0, 18, 18, 1, 152064, 0x12ac80ef
-0, 19, 19, 1, 152064, 0x470956e4
-0, 20, 20, 1, 152064, 0x3954f944
-0, 21, 21, 1, 152064, 0x2147f0db
-0, 22, 22, 1, 152064, 0xca7c9a39
-0, 23, 23, 1, 152064, 0x727e7772
-0, 24, 24, 1, 152064, 0xb3b0f8c8
-0, 25, 25, 1, 152064, 0x6d6eb7ba
-0, 26, 26, 1, 152064, 0x1eaf5734
-0, 27, 27, 1, 152064, 0xd7938056
-0, 28, 28, 1, 152064, 0xcae1ead3
-0, 29, 29, 1, 152064, 0x6411b82c
-0, 30, 30, 1, 152064, 0x6f001718
-0, 31, 31, 1, 152064, 0x3724d9b8
-0, 32, 32, 1, 152064, 0x04501fec
-0, 33, 33, 1, 152064, 0x3f4d13bb
-0, 34, 34, 1, 152064, 0x38f2510a
-0, 35, 35, 1, 152064, 0x5bb71106
-0, 36, 36, 1, 152064, 0xe888b09d
-0, 37, 37, 1, 152064, 0xca6f5f6a
-0, 38, 38, 1, 152064, 0x01c7d329
-0, 39, 39, 1, 152064, 0x4e670932
-0, 40, 40, 1, 152064, 0x651b4984
-0, 41, 41, 1, 152064, 0x76371fa3
-0, 42, 42, 1, 152064, 0x60574af0
-0, 43, 43, 1, 152064, 0xcc1f2080
-0, 44, 44, 1, 152064, 0xdbf29eb6
-0, 45, 45, 1, 152064, 0x8ff870be
-0, 46, 46, 1, 152064, 0x5d39ea56
-0, 47, 47, 1, 152064, 0xabb7c0a5
-0, 48, 48, 1, 152064, 0xecb15771
-0, 49, 49, 1, 152064, 0x571ab1ae
-0, 50, 50, 1, 152064, 0xc0f22bad
-0, 51, 51, 1, 152064, 0x22e7593f
-0, 52, 52, 1, 152064, 0x3c3dee75
-0, 53, 53, 1, 152064, 0x1c80b995
-0, 54, 54, 1, 152064, 0x5ba442fa
-0, 55, 55, 1, 152064, 0xe2291c05
-0, 56, 56, 1, 152064, 0xa780b11e
-0, 57, 57, 1, 152064, 0x15bf83fa
-0, 58, 58, 1, 152064, 0x60fb7c11
-0, 59, 59, 1, 152064, 0x3fb8cee5
+0, 0, 0, 1, 152064, 0x07ba97a3
+0, 1, 1, 1, 152064, 0x36d54b72
+0, 2, 2, 1, 152064, 0x71e68289
+0, 3, 3, 1, 152064, 0x4eddc5bc
+0, 4, 4, 1, 152064, 0x60510fa7
+0, 5, 5, 1, 152064, 0xb9b00e87
+0, 6, 6, 1, 152064, 0x93856465
+0, 7, 7, 1, 152064, 0xee2bfd9c
+0, 8, 8, 1, 152064, 0x359c4171
+0, 9, 9, 1, 152064, 0xab1d7f8e
+0, 10, 10, 1, 152064, 0x5ba19a95
+0, 11, 11, 1, 152064, 0xf9602415
+0, 12, 12, 1, 152064, 0x50bfb92e
+0, 13, 13, 1, 152064, 0xeab46e77
+0, 14, 14, 1, 152064, 0xcad19286
+0, 15, 15, 1, 152064, 0xf6977ca8
+0, 16, 16, 1, 152064, 0x7dd6dfb4
+0, 17, 17, 1, 152064, 0xc42d634d
+0, 18, 18, 1, 152064, 0x9b4a5e7c
+0, 19, 19, 1, 152064, 0xab8ae045
+0, 20, 20, 1, 152064, 0x2880c285
+0, 21, 21, 1, 152064, 0xb2799c00
+0, 22, 22, 1, 152064, 0x2bd360b3
+0, 23, 23, 1, 152064, 0xfbc3de30
+0, 24, 24, 1, 152064, 0x83651fa1
+0, 25, 25, 1, 152064, 0x235a2b0e
+0, 26, 26, 1, 152064, 0x46dd59de
+0, 27, 27, 1, 152064, 0xa65776ec
+0, 28, 28, 1, 152064, 0x4c2664e6
+0, 29, 29, 1, 152064, 0x61a08fdf
+0, 30, 30, 1, 152064, 0x039fd408
+0, 31, 31, 1, 152064, 0xeaf266d2
+0, 32, 32, 1, 152064, 0xd0064f4b
+0, 33, 33, 1, 152064, 0x585d3176
+0, 34, 34, 1, 152064, 0xd951e9e4
+0, 35, 35, 1, 152064, 0xf9fc8295
+0, 36, 36, 1, 152064, 0xddb31ca2
+0, 37, 37, 1, 152064, 0x3317b6e1
+0, 38, 38, 1, 152064, 0xfb353e5c
+0, 39, 39, 1, 152064, 0xdd1d47f3
+0, 40, 40, 1, 152064, 0x2826c3bf
+0, 41, 41, 1, 152064, 0xccde4ca0
+0, 42, 42, 1, 152064, 0x06430283
+0, 43, 43, 1, 152064, 0xd5579a9b
+0, 44, 44, 1, 152064, 0x0f99ad7b
+0, 45, 45, 1, 152064, 0xa21b8419
+0, 46, 46, 1, 152064, 0xd37b7cec
+0, 47, 47, 1, 152064, 0x22bbfd3f
+0, 48, 48, 1, 152064, 0x6df32b91
+0, 49, 49, 1, 152064, 0xd0753615
+0, 50, 50, 1, 152064, 0x5e26507e
+0, 51, 51, 1, 152064, 0x797aa7e6
+0, 52, 52, 1, 152064, 0xb924b517
+0, 53, 53, 1, 152064, 0xa77d0e9c
+0, 54, 54, 1, 152064, 0x3c8358a7
+0, 55, 55, 1, 152064, 0x2f9ce8d2
+0, 56, 56, 1, 152064, 0x612f3283
+0, 57, 57, 1, 152064, 0xf4985ba3
+0, 58, 58, 1, 152064, 0x1cd81827
+0, 59, 59, 1, 152064, 0x9b4a7f7b
Test h264-conformance-frext-frext_mmco4_sony_b failed. Look at tests/data/fate/h264-conformance-frext-frext_mmco4_sony_b.err for details.
[...]
--
Michael GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB
DNS cache poisoning attacks, popular search engine, Google internet authority
dont be evil, please
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: Digital signature
URL: <http://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20130219/51359e35/attachment.asc>
More information about the ffmpeg-devel
mailing list