Skip to contentSkip to conversation

pcm:file: delegate htimestamping to slave instead of always getting real_htimestamp

PATCH In review [email protected] 3 participants 5 messages Mar 23, 2017 – Mar 29, 2017

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 pcm:file: delegate htimestamping to slave instead of always getting real_htimestamp

From: Andreas Pape <[email protected]>

purpose of this fix, is to read most accurate timestamps.

From documentation of /src/pcm/pcm.c, we can see:
"""" \par Timestamp mode

The timestamp mode specifies, if timestamps are activated. Currently, only #SND_PCM_TSTAMP_NONE and #SND_PCM_TSTAMP_MMAP modes are known.
The mmap mode means that timestamp is taken on every period time boundary. Corresponding position in the ring buffer assigned to timestamp can be obtained using #snd_pcm_htimestamp() function. """"

As snd_pcm_generic_htimestamp() internally calls snd_pcm_htimestamp() to read time, so accurate timestamp can be read from snd_pcm_generic_htimestamp().

Also, in case of pcm_file, if the underlying slave is hardware, then we would wish to read elapsed hardware time, as it will be the most accurate, as opposed to the elapsed wall time. This will provide pcm_file with the most accurate timestamps.

Following are the timesamps read with timestamp enabled, for with fix and without fix scenarios:

1> With fix:
:~#time aplay --enable-tstamp -Dhtstamp_test --period-time=5000 -v -fdat /dev/urandom Playing raw data '/dev/urandom' : Signed 16 bit Little Endian, Rate 48000 Hz, Stereo File PCM (file=/tmp/swarate_out.wav)
Final file PCM (file=/tmp/swarate_out.wav)
..

Slave: Hardware PCM card 0 'imx6q-sabresd-wm8962' device 0 subdevice 0 Its setup is:

  stream       : PLAYBACK
  access       : RW_INTERLEAVED
  format       : S16_LE
  subformat    : STD
  channels     : 2
  rate         : 48000
  exact rate   : 48000 (48000/1)
  msbits       : 16
  buffer_size  : 24000
  period_size  : 240
  period_time  : 5000
  tstamp_mode  : ENABLE

.
.
Before sleep = 142:409.807623
After sleep = 142:409.807623
Before sleep = 142:414.806016 (calling snd_pcm_htimestamp(handle, &avail, &tstamp_before)) sleep of 2 milisec
After sleep = 142:414.806016 (calling snd_pcm_htimestamp(handle, &avail, &tstamp_after)

Quoted 1 line

From the above timestamps, we can see that slave has returned the same timestamps, as --period-time choosen is 5msec.

2> Without this fix:
The timestamps are returned with realtime value.

:~# time aplay --enable-tstamp -Dhtstamp_test --period-time=5000 -v -fdat /dev/urandom
Playing raw data '/dev/urandom' : Signed 16 bit Little Endian, Rate 48000 Hz, Stereo
File PCM (file=/tmp/swarate_out.wav)
Final file PCM (file=/tmp/swarate_out.wav)
.
.
Slave: Hardware PCM card 0 'imx6q-sabresd-wm8962' device 0 subdevice 0
Its setup is:
  stream       : PLAYBACK
  access       : RW_INTERLEAVED
  format       : S16_LE
  subformat    : STD
  channels     : 2
  rate         : 48000
  exact rate   : 48000 (48000/1)
  msbits       : 16
  buffer_size  : 24000
  period_size  : 240
  period_time  : 5000
  tstamp_mode  : ENABLE
.
.
Before sleep = 241:136.875845 (calling snd_pcm_htimestamp(handle, &avail, &tstamp_before))
sleep of 2 milisec
After sleep = 241:139.076376 (calling snd_pcm_htimestamp(handle, &avail, &tstamp_after)

We can observe here, the timestamps shows time diff of ~2ms, which is the time gap of sleep duration. Before sleep = 241:139.617588
After sleep = 241:141.746845
Before sleep = 241:142.291618
After sleep = 241:144.406406
Before sleep = 241:144.951421
After sleep = 241:147.066118
Before sleep = 241:147.623421
After sleep = 241:149.740573

src/pcm/pcm_file.c +1 -1
index 8012251..fe36a1c 100644
@@ -728,7 +728,7 @@ static const snd_pcm_fast_ops_t snd_pcm_file_fast_ops = {
 	.poll_descriptors_count = snd_pcm_generic_poll_descriptors_count,
 	.poll_descriptors = snd_pcm_generic_poll_descriptors,
 	.poll_revents = snd_pcm_generic_poll_revents,
-	.htimestamp = snd_pcm_generic_real_htimestamp,
+	.htimestamp = snd_pcm_generic_htimestamp,
 };
 
 /**
Signature · 1 line
2.7.4
#1 lore raw

ping

-----Original Message-----
From: [email protected] [mailto:[email protected]]
Sent: 23 March 2017 17:11
To: [email protected]
Cc: [email protected]; Sutar, Mounesh <[email protected]>; Andreas Pape <[email protected]>; Mounesh Sutar <[email protected]>
Subject: [PATCH 1/1] pcm:file: delegate htimestamping to slave instead of always getting real_htimestamp

From: Andreas Pape <[email protected]>

purpose of this fix, is to read most accurate timestamps.

From documentation of /src/pcm/pcm.c, we can see:
"""" \par Timestamp mode

The timestamp mode specifies, if timestamps are activated. Currently, only #SND_PCM_TSTAMP_NONE and #SND_PCM_TSTAMP_MMAP modes are known.
The mmap mode means that timestamp is taken on every period time boundary. Corresponding position in the ring buffer assigned to timestamp can be obtained using #snd_pcm_htimestamp() function. """"

As snd_pcm_generic_htimestamp() internally calls snd_pcm_htimestamp() to read time, so accurate timestamp can be read from snd_pcm_generic_htimestamp().

Also, in case of pcm_file, if the underlying slave is hardware, then we would wish to read elapsed hardware time, as it will be the most accurate, as opposed to the elapsed wall time. This will provide pcm_file with the most accurate timestamps.

Following are the timesamps read with timestamp enabled, for with fix and without fix scenarios:

1> With fix:
:~#time aplay --enable-tstamp -Dhtstamp_test --period-time=5000 -v -fdat /dev/urandom Playing raw data '/dev/urandom' : Signed 16 bit Little Endian, Rate 48000 Hz, Stereo File PCM (file=/tmp/swarate_out.wav) Final file PCM (file=/tmp/swarate_out.wav) ..

Slave: Hardware PCM card 0 'imx6q-sabresd-wm8962' device 0 subdevice 0 Its setup is:

  stream       : PLAYBACK
  access       : RW_INTERLEAVED
  format       : S16_LE
  subformat    : STD
  channels     : 2
  rate         : 48000
  exact rate   : 48000 (48000/1)
  msbits       : 16
  buffer_size  : 24000
  period_size  : 240
  period_time  : 5000
  tstamp_mode  : ENABLE

.
.
Before sleep = 142:409.807623
After sleep = 142:409.807623
Before sleep = 142:414.806016 (calling snd_pcm_htimestamp(handle, &avail, &tstamp_before)) sleep of 2 milisec After sleep = 142:414.806016 (calling snd_pcm_htimestamp(handle, &avail, &tstamp_after) From the above timestamps, we can see that slave has returned the same timestamps, as --period-time choosen is 5msec.

2> Without this fix:
The timestamps are returned with realtime value.

:~# time aplay --enable-tstamp -Dhtstamp_test --period-time=5000 -v -fdat /dev/urandom Playing raw data '/dev/urandom' : Signed 16 bit Little Endian, Rate 48000 Hz, Stereo File PCM (file=/tmp/swarate_out.wav) Final file PCM (file=/tmp/swarate_out.wav) .
.
Slave: Hardware PCM card 0 'imx6q-sabresd-wm8962' device 0 subdevice 0 Its setup is:
  stream       : PLAYBACK
  access       : RW_INTERLEAVED
  format       : S16_LE
  subformat    : STD
  channels     : 2
  rate         : 48000
  exact rate   : 48000 (48000/1)
  msbits       : 16
  buffer_size  : 24000
  period_size  : 240
  period_time  : 5000
  tstamp_mode  : ENABLE
.
.
Before sleep = 241:136.875845 (calling snd_pcm_htimestamp(handle, &avail, &tstamp_before)) sleep of 2 milisec After sleep = 241:139.076376 (calling snd_pcm_htimestamp(handle, &avail, &tstamp_after)

We can observe here, the timestamps shows time diff of ~2ms, which is the time gap of sleep duration. Before sleep = 241:139.617588
After sleep = 241:141.746845
Before sleep = 241:142.291618
After sleep = 241:144.406406
Before sleep = 241:144.951421
After sleep = 241:147.066118
Before sleep = 241:147.623421
After sleep = 241:149.740573

src/pcm/pcm_file.c +1 -1
@@ -728,7 +728,7 @@ static const snd_pcm_fast_ops_t snd_pcm_file_fast_ops = {
 	.poll_descriptors_count = snd_pcm_generic_poll_descriptors_count,
 	.poll_descriptors = snd_pcm_generic_poll_descriptors,
 	.poll_revents = snd_pcm_generic_poll_revents,
-	.htimestamp = snd_pcm_generic_real_htimestamp,
+	.htimestamp = snd_pcm_generic_htimestamp,
 };
 
 /**

--
2.7.4

#2 lore raw
Quoted 2 lines · Mounesh

ping

No need for ping. Please give some time for review.

Takashi

Quoted 97 lines

-----Original Message-----
From: [email protected] [mailto:[email protected]]
Sent: 23 March 2017 17:11
To: [email protected]
Cc: [email protected]; Sutar, Mounesh <[email protected]>; Andreas Pape <[email protected]>; Mounesh Sutar <[email protected]>
Subject: [PATCH 1/1] pcm:file: delegate htimestamping to slave instead of always getting real_htimestamp

From: Andreas Pape <[email protected]>

purpose of this fix, is to read most accurate timestamps.

From documentation of /src/pcm/pcm.c, we can see:
"""" \par Timestamp mode

The timestamp mode specifies, if timestamps are activated. Currently, only #SND_PCM_TSTAMP_NONE and #SND_PCM_TSTAMP_MMAP modes are known.
The mmap mode means that timestamp is taken on every period time boundary. Corresponding position in the ring buffer assigned to timestamp can be obtained using #snd_pcm_htimestamp() function. """"

As snd_pcm_generic_htimestamp() internally calls snd_pcm_htimestamp() to read time, so accurate timestamp can be read from snd_pcm_generic_htimestamp().

Also, in case of pcm_file, if the underlying slave is hardware, then we would wish to read elapsed hardware time, as it will be the most accurate, as opposed to the elapsed wall time. This will provide pcm_file with the most accurate timestamps.

Following are the timesamps read with timestamp enabled, for with fix and without fix scenarios:

1> With fix:
:~#time aplay --enable-tstamp -Dhtstamp_test --period-time=5000 -v -fdat /dev/urandom Playing raw data '/dev/urandom' : Signed 16 bit Little Endian, Rate 48000 Hz, Stereo File PCM (file=/tmp/swarate_out.wav) Final file PCM (file=/tmp/swarate_out.wav) ..

Slave: Hardware PCM card 0 'imx6q-sabresd-wm8962' device 0 subdevice 0 Its setup is:

  stream       : PLAYBACK
  access       : RW_INTERLEAVED
  format       : S16_LE
  subformat    : STD
  channels     : 2
  rate         : 48000
  exact rate   : 48000 (48000/1)
  msbits       : 16
  buffer_size  : 24000
  period_size  : 240
  period_time  : 5000
  tstamp_mode  : ENABLE

.
.
Before sleep = 142:409.807623
After sleep = 142:409.807623
Before sleep = 142:414.806016 (calling snd_pcm_htimestamp(handle, &avail, &tstamp_before)) sleep of 2 milisec After sleep = 142:414.806016 (calling snd_pcm_htimestamp(handle, &avail, &tstamp_after) From the above timestamps, we can see that slave has returned the same timestamps, as --period-time choosen is 5msec.

2> Without this fix:
The timestamps are returned with realtime value.

:~# time aplay --enable-tstamp -Dhtstamp_test --period-time=5000 -v -fdat /dev/urandom Playing raw data '/dev/urandom' : Signed 16 bit Little Endian, Rate 48000 Hz, Stereo File PCM (file=/tmp/swarate_out.wav) Final file PCM (file=/tmp/swarate_out.wav) .
.
Slave: Hardware PCM card 0 'imx6q-sabresd-wm8962' device 0 subdevice 0 Its setup is:
  stream       : PLAYBACK
  access       : RW_INTERLEAVED
  format       : S16_LE
  subformat    : STD
  channels     : 2
  rate         : 48000
  exact rate   : 48000 (48000/1)
  msbits       : 16
  buffer_size  : 24000
  period_size  : 240
  period_time  : 5000
  tstamp_mode  : ENABLE
.
.
Before sleep = 241:136.875845 (calling snd_pcm_htimestamp(handle, &avail, &tstamp_before)) sleep of 2 milisec After sleep = 241:139.076376 (calling snd_pcm_htimestamp(handle, &avail, &tstamp_after)

We can observe here, the timestamps shows time diff of ~2ms, which is the time gap of sleep duration. Before sleep = 241:139.617588
After sleep = 241:141.746845
Before sleep = 241:142.291618
After sleep = 241:144.406406
Before sleep = 241:144.951421
After sleep = 241:147.066118
Before sleep = 241:147.623421
After sleep = 241:149.740573

src/pcm/pcm_file.c +1 -2
@@ -728,7 +728,7 @@ static const snd_pcm_fast_ops_t snd_pcm_file_fast_ops = {
 	.poll_descriptors_count = snd_pcm_generic_poll_descriptors_count,
 	.poll_descriptors = snd_pcm_generic_poll_descriptors,
 	.poll_revents = snd_pcm_generic_poll_revents,
-	.htimestamp = snd_pcm_generic_real_htimestamp,
+	.htimestamp = snd_pcm_generic_htimestamp,
 };
 
 /**
--

2.7.4

#3 lore raw

sure, Thanks.

Quoted 136 lines · Takashi Iwai
Quoted 2 lines · Mounesh

ping

No need for ping. Please give some time for review.

Takashi

Quoted 6 lines

-----Original Message-----
From: [email protected] [mailto:[email protected]]
Sent: 23 March 2017 17:11
To: [email protected]
Cc: [email protected]; Sutar, Mounesh <

[email protected]>; Andreas Pape <[email protected]>; Mounesh Sutar <[email protected]>

Quoted 1 line

Subject: [PATCH 1/1] pcm:file: delegate htimestamping to slave instead

of always getting real_htimestamp

Quoted 9 lines

From: Andreas Pape <[email protected]>

purpose of this fix, is to read most accurate timestamps.

From documentation of /src/pcm/pcm.c, we can see:
"""" \par Timestamp mode

The timestamp mode specifies, if timestamps are activated. Currently,

only #SND_PCM_TSTAMP_NONE and #SND_PCM_TSTAMP_MMAP modes are known.

Quoted 1 line

The mmap mode means that timestamp is taken on every period time

boundary. Corresponding position in the ring buffer assigned to timestamp can be obtained using #snd_pcm_htimestamp() function. """"

Quoted 2 lines

As snd_pcm_generic_htimestamp() internally calls snd_pcm_htimestamp()

to read time, so accurate timestamp can be read from snd_pcm_generic_htimestamp().

Quoted 2 lines

Also, in case of pcm_file, if the underlying slave is hardware, then we

would wish to read elapsed hardware time, as it will be the most accurate, as opposed to the elapsed wall time.

Quoted 3 lines

This will provide pcm_file with the most accurate timestamps.

Following are the timesamps read with timestamp enabled, for with fix

and without fix scenarios:

Quoted 3 lines

1> With fix:
:~#time aplay --enable-tstamp -Dhtstamp_test --period-time=5000 -v -fdat

/dev/urandom Playing raw data '/dev/urandom' : Signed 16 bit Little Endian, Rate 48000 Hz, Stereo File PCM (file=/tmp/swarate_out.wav) Final file PCM (file=/tmp/swarate_out.wav) ..

Quoted 2 lines

Slave: Hardware PCM card 0 'imx6q-sabresd-wm8962' device 0 subdevice 0

Its setup is:

Quoted 12 lines
  stream       : PLAYBACK
  access       : RW_INTERLEAVED
  format       : S16_LE
  subformat    : STD
  channels     : 2
  rate         : 48000
  exact rate   : 48000 (48000/1)
  msbits       : 16
  buffer_size  : 24000
  period_size  : 240
  period_time  : 5000
  tstamp_mode  : ENABLE
Quoted 5 lines

.
.
Before sleep = 142:409.807623
After sleep = 142:409.807623
Before sleep = 142:414.806016 (calling snd_pcm_htimestamp(handle,

&avail, &tstamp_before)) sleep of 2 milisec After sleep = 142:414.806016 (calling snd_pcm_htimestamp(handle, &avail, &tstamp_after) From the above timestamps, we can see that slave has returned the same timestamps, as --period-time choosen is 5msec.

Quoted 5 lines

2> Without this fix:
The timestamps are returned with realtime value.

:~# time aplay --enable-tstamp -Dhtstamp_test --period-time=5000 -v

-fdat /dev/urandom Playing raw data '/dev/urandom' : Signed 16 bit Little Endian, Rate 48000 Hz, Stereo File PCM (file=/tmp/swarate_out.wav) Final file PCM (file=/tmp/swarate_out.wav) .

Quoted 2 lines

.
Slave: Hardware PCM card 0 'imx6q-sabresd-wm8962' device 0 subdevice 0

Its setup is:

Quoted 12 lines
  stream       : PLAYBACK
  access       : RW_INTERLEAVED
  format       : S16_LE
  subformat    : STD
  channels     : 2
  rate         : 48000
  exact rate   : 48000 (48000/1)
  msbits       : 16
  buffer_size  : 24000
  period_size  : 240
  period_time  : 5000
  tstamp_mode  : ENABLE
Quoted 3 lines

.
.
Before sleep = 241:136.875845 (calling snd_pcm_htimestamp(handle,

&avail, &tstamp_before)) sleep of 2 milisec After sleep = 241:139.076376 (calling snd_pcm_htimestamp(handle, &avail, &tstamp_after)

Quoted 2 lines

We can observe here, the timestamps shows time diff of ~2ms, which is

the time gap of sleep duration.

src/pcm/pcm_file.c index

Quoted 13 lines

Before sleep = 241:139.617588
After sleep = 241:141.746845
Before sleep = 241:142.291618
After sleep = 241:144.406406
Before sleep = 241:144.951421
After sleep = 241:147.066118
Before sleep = 241:147.623421
After sleep = 241:149.740573

src/pcm/pcm_file.c index renamed from src/pcm/pcm_file.c +0 -0

8012251..fe36a1c 100644

src/pcm/pcm_file.c static const snd_pcm_fast_ops_t

Quoted 3 lines
src/pcm/pcm_file.c +0 -0
@@ -728,7 +728,7 @@ static const snd_pcm_fast_ops_t

snd_pcm_file_fast_ops = {

Quoted 11 lines
src/pcm/pcm_file.c +1 -2
      .poll_descriptors_count = snd_pcm_generic_poll_descriptors_count,
      .poll_descriptors = snd_pcm_generic_poll_descriptors,
      .poll_revents = snd_pcm_generic_poll_revents,
-     .htimestamp = snd_pcm_generic_real_htimestamp,
+     .htimestamp = snd_pcm_generic_htimestamp,
 };
 
 /**
--

2.7.4

#4 lore raw
In reply to [email protected]
Quoted 87 lines · On Thu, 23 Mar 2017 12:40:45 +0100, [email protected] wrote:

From: Andreas Pape <[email protected]>

purpose of this fix, is to read most accurate timestamps.

From documentation of /src/pcm/pcm.c, we can see:
"""" \par Timestamp mode

The timestamp mode specifies, if timestamps are activated. Currently, only #SND_PCM_TSTAMP_NONE and #SND_PCM_TSTAMP_MMAP modes are known.
The mmap mode means that timestamp is taken on every period time boundary. Corresponding position in the ring buffer assigned to timestamp can be obtained using #snd_pcm_htimestamp() function. """"

As snd_pcm_generic_htimestamp() internally calls snd_pcm_htimestamp() to read time, so accurate timestamp can be read from snd_pcm_generic_htimestamp().

Also, in case of pcm_file, if the underlying slave is hardware, then we would wish to read elapsed hardware time, as it will be the most accurate, as opposed to the elapsed wall time. This will provide pcm_file with the most accurate timestamps.

Following are the timesamps read with timestamp enabled, for with fix and without fix scenarios:

1> With fix:
:~#time aplay --enable-tstamp -Dhtstamp_test --period-time=5000 -v -fdat /dev/urandom Playing raw data '/dev/urandom' : Signed 16 bit Little Endian, Rate 48000 Hz, Stereo File PCM (file=/tmp/swarate_out.wav)
Final file PCM (file=/tmp/swarate_out.wav)
..

Slave: Hardware PCM card 0 'imx6q-sabresd-wm8962' device 0 subdevice 0 Its setup is:

  stream       : PLAYBACK
  access       : RW_INTERLEAVED
  format       : S16_LE
  subformat    : STD
  channels     : 2
  rate         : 48000
  exact rate   : 48000 (48000/1)
  msbits       : 16
  buffer_size  : 24000
  period_size  : 240
  period_time  : 5000
  tstamp_mode  : ENABLE

.
.
Before sleep = 142:409.807623
After sleep = 142:409.807623
Before sleep = 142:414.806016 (calling snd_pcm_htimestamp(handle, &avail, &tstamp_before)) sleep of 2 milisec
After sleep = 142:414.806016 (calling snd_pcm_htimestamp(handle, &avail, &tstamp_after)

Quoted 1 line

From the above timestamps, we can see that slave has returned the same timestamps, as --period-time choosen is 5msec.

2> Without this fix:
The timestamps are returned with realtime value.

:~# time aplay --enable-tstamp -Dhtstamp_test --period-time=5000 -v -fdat /dev/urandom
Playing raw data '/dev/urandom' : Signed 16 bit Little Endian, Rate 48000 Hz, Stereo
File PCM (file=/tmp/swarate_out.wav)
Final file PCM (file=/tmp/swarate_out.wav)
.
.
Slave: Hardware PCM card 0 'imx6q-sabresd-wm8962' device 0 subdevice 0
Its setup is:
  stream       : PLAYBACK
  access       : RW_INTERLEAVED
  format       : S16_LE
  subformat    : STD
  channels     : 2
  rate         : 48000
  exact rate   : 48000 (48000/1)
  msbits       : 16
  buffer_size  : 24000
  period_size  : 240
  period_time  : 5000
  tstamp_mode  : ENABLE
.
.
Before sleep = 241:136.875845 (calling snd_pcm_htimestamp(handle, &avail, &tstamp_before))
sleep of 2 milisec
After sleep = 241:139.076376 (calling snd_pcm_htimestamp(handle, &avail, &tstamp_after)

We can observe here, the timestamps shows time diff of ~2ms, which is the time gap of sleep duration. Before sleep = 241:139.617588
After sleep = 241:141.746845
Before sleep = 241:142.291618
After sleep = 241:144.406406
Before sleep = 241:144.951421
After sleep = 241:147.066118
Before sleep = 241:147.623421
After sleep = 241:149.740573

Quoted 2 lines

Applied now, thanks.

Takashi