From: Michael Niedermayer <michael@niedermayer.cc>
To: FFmpeg development discussions and patches <ffmpeg-devel@ffmpeg.org>
Subject: Re: [FFmpeg-devel] [PATCH 1/2] avutil/random_seed: Speed up fate test
Date: Sun, 30 Jan 2022 19:23:35 +0100
Message-ID: <20220130182335.GG2829255@pb2> (raw)
In-Reply-To: <164337232528.23111.7444999659506493047@lain.red.khirnov.net>
[-- Attachment #1.1: Type: text/plain, Size: 5262 bytes --]
On Fri, Jan 28, 2022 at 01:18:45PM +0100, Anton Khirnov wrote:
> Quoting Michael Niedermayer (2022-01-25 11:21:08)
> > This decreases the quality of the seeds during the test, it does not affect
> > the seeds outside the test.
> > There is a small chance that this causes test failures, if that happens
> > the threshold needs adjusting
> >
> > Testing on an idle x86_64 system shows that this passes even with the test strength
> > increased to 32768 from 256 with 0 retries
> > So test failures are not anticipated
> > Lowering the threshold from 3 to 2 causes failure at test strength of 32768
> >
> > Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
> > ---
> > libavutil/random_seed.c | 2 +-
> > 1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/libavutil/random_seed.c b/libavutil/random_seed.c
> > index 70dc509d2f..509b74936c 100644
> > --- a/libavutil/random_seed.c
> > +++ b/libavutil/random_seed.c
> > @@ -95,7 +95,7 @@ static uint32_t get_generic_seed(void)
> > last_td = t - last_t;
> > buffer[++i & 511] += last_td % 3294638521U;
> > if ((t - init_t) >= CLOCKS_PER_SEC>>5)
> > - if (last_i && i - last_i > 4 || i - last_i > 64 || TEST && i - last_i > 8)
> > + if (last_i && i - last_i > 4 || i - last_i > 64 || TEST && i - last_i > 3)
>
> On my Ryzen 5950x, this brings the test runtime from > 60s to ~30s,
> which is a good improvement, but still quite a lot (the entire FATE run
> with -j32 and without the random-seed test is ~40s).
>
> How about something like this instead (takes about 8s on my machines, observed
> no test failures):
this works, though it may be better to estimate the collected entropy
and stop once enough is collected
i played around with the code a bit and got the test to pass with 32768 in 5min
down from i think 17min
and the unmodified 256 test from 8 sec to 3 sec
the 8 is with my 2 and your patch applied
but the resulting threshold in the code doesnt feel correct
also a completely different approuch would be to predict/compress the
timestamp deltas to estimate the entropy
i need to do something else now or i loose my sanity, maybe ill look at
it again in a few days, but feel free to continue working on the WIP below
if you like (if something makes no sense in it it likely doesnt)
thx
diff --git a/libavutil/random_seed.c b/libavutil/random_seed.c
index b1bbbd1b69..0255de7302 100644
--- a/libavutil/random_seed.c
+++ b/libavutil/random_seed.c
@@ -62,18 +62,21 @@ static int read_random(uint32_t *dst, const char *file)
#endif
}
+#define N_COUNTS 5
+
static uint32_t get_generic_seed(void)
{
uint64_t tmp[120/8];
struct AVSHA *sha = (void*)tmp;
clock_t last_t = 0;
clock_t last_td = 0;
- clock_t init_t = 0;
+ clock_t last_td2 = 0;
static uint64_t i = 0;
static uint32_t buffer[512] = { 0 };
unsigned char digest[20];
uint64_t last_i = i;
- uint64_t cnt = 0;
+ uint32_t last_td_count[N_COUNTS] = {0};
+ int entropy = 0;
av_assert0(sizeof(tmp) >= av_sha_size);
@@ -89,21 +92,33 @@ static uint32_t get_generic_seed(void)
for (;;) {
clock_t t = clock();
- if (last_t + 2*last_td + (CLOCKS_PER_SEC > 1000) >= t &&
- !(TEST && cnt > (1 << 15))) {
- last_td = t - last_t;
- buffer[i & 511] = 1664525*buffer[i & 511] + 1013904223 + (last_td % 3294638521U);
- cnt++;
+ clock_t tmp = t - last_t;
+ if (last_td == tmp || last_td2 == tmp) {
+ buffer[i & 511] = 1664525*buffer[i & 511] + 1013904223 + (tmp % 3294638521U);
+ if (last_td == tmp) {
+ last_td_count[0] ++;
+ } else {
+ int add_entropy = INT_MAX;
+ for (int j = N_COUNTS-1; j; j--) {
+ add_entropy = FFMIN(add_entropy, abs(last_td_count[j] - last_td_count[0]));
+ last_td_count[j] = last_td_count[j-1];
+ }
+ last_td_count[0] = 0;
+ if (last_td2 != last_td)
+ entropy += av_log2(add_entropy);
+ last_td2 = last_td;
+ }
} else {
- last_td = t - last_t;
- buffer[++i & 511] += last_td % 3294638521U;
- if ((t - init_t) >= CLOCKS_PER_SEC>>5)
- if (last_i && i - last_i > 4 || i - last_i > 64 || TEST && i - last_i > 3)
- break;
+ if (last_td2 != last_td)
+ entropy += av_log2(FFMIN(abs(last_td - tmp), abs(last_td2 - tmp)) + 1) + av_log2(last_td_count[0]);
+ buffer[++i & 511] += tmp % 3294638521U;
+ last_td2 = last_td;
+ last_td_count[0] = 0;
}
+ if (last_i && entropy > 32 || entropy > 128 || TEST && entropy > 8)
+ break;
+ last_td = tmp;
last_t = t;
- if (!init_t)
- init_t = t;
}
if(TEST) {
[...]
--
Michael GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB
No snowflake in an avalanche ever feels responsible. -- Voltaire
[-- Attachment #1.2: signature.asc --]
[-- Type: application/pgp-signature, Size: 195 bytes --]
[-- Attachment #2: Type: text/plain, Size: 251 bytes --]
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
prev parent reply other threads:[~2022-01-30 18:23 UTC|newest]
Thread overview: 4+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-01-25 10:21 Michael Niedermayer
2022-01-25 10:21 ` [FFmpeg-devel] [PATCH 2/2] libavutil/random_seed: Skip testing generic seed twice Michael Niedermayer
2022-01-28 12:18 ` [FFmpeg-devel] [PATCH 1/2] avutil/random_seed: Speed up fate test Anton Khirnov
2022-01-30 18:23 ` Michael Niedermayer [this message]
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20220130182335.GG2829255@pb2 \
--to=michael@niedermayer.cc \
--cc=ffmpeg-devel@ffmpeg.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Git Inbox Mirror of the ffmpeg-devel mailing list - see https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
This inbox may be cloned and mirrored by anyone:
git clone --mirror https://master.gitmailbox.com/ffmpegdev/0 ffmpegdev/git/0.git
# If you have public-inbox 1.1+ installed, you may
# initialize and index your mirror using the following commands:
public-inbox-init -V2 ffmpegdev ffmpegdev/ https://master.gitmailbox.com/ffmpegdev \
ffmpegdev@gitmailbox.com
public-inbox-index ffmpegdev
Example config snippet for mirrors.
AGPL code for this site: git clone https://public-inbox.org/public-inbox.git