Skip to contentSkip to conversation

SBC encoder conformance test for version 4.25 passed!

PATCH In review Christian Hoene 10 participants 24 messages Jan 1, 2009 – Jul 15, 2009 1 message not in the archive Bluetooth drivers

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

#1 lore raw

Hello Siarhei,

sorry for the delay - these days my family comes first ;-)

Good news! All tests passed!
http://net.cs.uni-tuebingen.de/html/nexgenvoip/

Congratulations and a happy new year!

Christian

Quoted 12 lines

-----Original Message-----
From: Siarhei Siamashka [mailto:[email protected]]
Sent: Tuesday, December 30, 2008 11:46 AM
To: ext Christian Hoene
Subject: SBC encoder conformance test for version 4.25

Hello Christian,

Could you please rebuild SBC encoder test results on
http://net.cs.uni-tuebingen.de/html/nexgenvoip/html/
for the just released version 4.25?

Quoted 6 lines

I'm going to submit SIMD optimizations for the analysis filter which are almost done now (using ARM NEON, but Intel MMX/SSE2 requires only a minor modification). In any case I would prefer to make sure that the current C code is in a good shape first.

Thanks a lot for your participation. I hope that with our combined

efforts,

Quoted 5 lines

we can make SBC really much better and faster.

--
Best regards,
Siarhei Siamashka

#2 lore raw
Quoted 6 lines · On Thursday 01 January 2009 16:24:21 ext Christian Hoene wrote:

Hello Siarhei,

sorry for the delay - these days my family comes first ;-)

Good news! All tests passed!
http://net.cs.uni-tuebingen.de/html/nexgenvoip/

Thank you very much. Sorry for the delay with reply too.

I wonder if we need to do something about test 08?

Also could you try running a special test of version 4.25 with SBC_HIGH_PRECISION enabled? The point is that it uses the best possible precision for the analysis filter and minimizes effects of rounding errors (I would estimate that the precision should be even better than single precision floating point). Having the test run with this precision settings can expose some of the other minor bugs in the code (if they exist), which can now be hidden under the slight precision loss because of using 16-bit fixed point calculations.

Best regards,
Siarhei Siamashka

#3 lore raw

PATCH 1/1 SBC Encoder program

Quoted 2 lines

I wonder if we need to do something about test 08?

Yes, there was a bug in the encoder program. Attached the patch.

Greetings
Christian

Quoted 1 line

From e302dcac067486b83d1ba824227fd14878321cb2 Mon Sep 17 00:00:00 2001

From: Christian Hoene <[email protected]>
Date: Mon, 5 Jan 2009 12:38:52 +0100
Subject: [PATCH 3/3] Bug fix in SBC encoder program to correctly handle file sizes.

---
 sbc/sbcenc.c |    4 +++-
 1 files changed, 3 insertions(+), 1 deletions(-)
sbc/sbcenc.c +3 -1
index 180c348..9a3df3f 100644
@@ -47,7 +47,7 @@ static ssize_t __read(int fd, void *buf, size_t count)
 	while (count > 0) {
 		len = read(fd, buf + pos, count);
 		if (len <= 0)
-			return len;
+			return pos>len?pos:len;
 
 		count -= len;
 		pos   += len;
@@ -188,6 +188,8 @@ static void encode(char *filename, int subbands, int bitpool, int joint,
 
 		len = sbc_encode(&sbc, input, size,
 					output, sizeof(output), &encoded);
+		if (len <= 0)
+			break;
 		if (len < size)
 			memmove(input, input + len, size - len);
 
Signature · 1 line
1.6.1.rc2
#4 lore raw

Re: [PATCH] SBC Encoder program

Hi Christian

Quoted 6 lines

Yes, there was a bug in the encoder program. Attached the patch.

Greetings
Christian

Coding style:

return pos > len ? pos : len;

Signature · 2 lines
Luiz Augusto von Dentz
Engenheiro de Computação
#5 lore raw

Re: [PATCH] SBC Encoder program

Quoted 4 lines · Luiz Augusto von Dentz

Hi Christian

Quoted 2 lines

Yes, there was a bug in the encoder program. Attached the patch.

Quoted 8 lines
Quoted 4 lines

Greetings
Christian

Coding style:

return pos > len ? pos : len;

Thanks. I corrected it and even found another bug.

Attached the new and corrected patch.

Greetings
Christian

Quoted 1 line

From 9a914d4b282f4c885ff074263c4da03e482fd99d Mon Sep 17 00:00:00 2001

From: Christian Hoene <[email protected]>
Date: Mon, 5 Jan 2009 13:26:08 +0100
Subject: [PATCH 3/3] Fixed correct handling of frame sizes in the encoder

---
 sbc/sbc.c    |    6 +++---
 sbc/sbc.h    |    2 +-
 sbc/sbcenc.c |    4 +++-
 3 files changed, 7 insertions(+), 5 deletions(-)
sbc/sbc.c +3 -3
index ce52e1e..959510f 100644
@@ -69,7 +69,7 @@ struct sbc_frame {
 	uint8_t subband_mode;
 	uint8_t subbands;
 	uint8_t bitpool;
-	uint8_t codesize;
+	uint16_t codesize;
 	uint8_t length;
 
 	/* bit number x set means joint stereo has been used in subband x */
@@ -1293,9 +1293,9 @@ int sbc_get_frame_duration(sbc_t *sbc)
 	return (1000000 * blocks * subbands) / frequency;
 }
 
-int sbc_get_codesize(sbc_t *sbc)
+uint16_t sbc_get_codesize(sbc_t *sbc)
 {
-	uint8_t subbands, channels, blocks;
+	uint16_t subbands, channels, blocks;
 	struct sbc_priv *priv;
 
 	priv = sbc->priv;
sbc/sbc.h +1 -1
index ab47e32..73f3058 100644
@@ -87,7 +87,7 @@ int sbc_encode(sbc_t *sbc, void *input, int input_len, void *output,
 		int output_len, int *written);
 int sbc_get_frame_length(sbc_t *sbc);
 int sbc_get_frame_duration(sbc_t *sbc);
-int sbc_get_codesize(sbc_t *sbc);
+uint16_t sbc_get_codesize(sbc_t *sbc);
 void sbc_finish(sbc_t *sbc);
 
 #ifdef __cplusplus
sbc/sbcenc.c +3 -1
index 180c348..124a9e2 100644
@@ -47,7 +47,7 @@ static ssize_t __read(int fd, void *buf, size_t count)
 	while (count > 0) {
 		len = read(fd, buf + pos, count);
 		if (len <= 0)
-			return len;
+			return pos > len ? pos : len;
 
 		count -= len;
 		pos   += len;
@@ -188,6 +188,8 @@ static void encode(char *filename, int subbands, int bitpool, int joint,
 
 		len = sbc_encode(&sbc, input, size,
 					output, sizeof(output), &encoded);
+		if (len <= 0)
+			break;
 		if (len < size)
 			memmove(input, input + len, size - len);
 
Signature · 1 line
1.6.1.rc2
#6 lore raw

Re: [PATCH] SBC Encoder program

Hi Christian,

Just the codesize is 16 bit, right? So I guess the following changes are not really necessary:

+ uint16_t subbands, channels, blocks;

Signature · 2 lines
Luiz Augusto von Dentz
Engenheiro de Computação
#7 lore raw

RE: [PATCH] SBC Encoder program

sbc/sbc.c int sbc_get_frame_duration(sbc_t *sbc) · line 1298

Quoted 5 lines

Just the codesize is 16 bit, right? So I guess the following changes are not really necessary:

sbc/sbc.c +1 -0
+	uint16_t subbands, channels, blocks;

No, but then we have to change the last line of the function to
return ((uint16_t)subbands) * blocks * channels * 2;

CH

#8 lore raw

Re: [PATCH] SBC Encoder program

Quoted 8 lines · On Monday 05 January 2009 16:42:39 ext Christian Hoene wrote:

sbc/sbc.c int sbc_get_frame_duration(sbc_t *sbc) · line 1298

Quoted 5 lines

Just the codesize is 16 bit, right? So I guess the following changes are not really necessary:

sbc/sbc.c +1 -0
+	uint16_t subbands, channels, blocks;

No, but then we have to change the last line of the function to
return ((uint16_t)subbands) * blocks * channels * 2;

Not quite so. There is such thing as "integer promotion" defined in C standard. And integer type is guaranteed to be at least 16 bit. But I agree that explicitly having uint16_t is much more likely not to raise any doubts about its correctness :)

Signature · 2 lines
Best regards,
Siarhei Siamashka
#9 lore raw

Re: [PATCH] SBC Encoder program

In reply to Christian Hoene
Quoted 8 lines · On Monday 05 January 2009 14:28:36 ext Christian Hoene wrote:
Quoted 7 lines · Luiz Augusto von Dentz

Hi Christian

Quoted 4 lines

Yes, there was a bug in the encoder program. Attached the patch.

Greetings
Christian

Quoted 7 lines
Quoted 3 lines

Coding style:

return pos > len ? pos : len;

Thanks. I corrected it and even found another bug.

Attached the new and corrected patch.

There are some parts that look a bit redundant/suspicious:

sbc/sbcenc.c static ssize_t __read(int fd, void *buf, size_t count) · line 53

Quoted 9 lines
sbc/sbcenc.c +1 -1
@@ -47,7 +47,7 @@ static ssize_t __read(int fd, void *buf, size_t count)
        while (count > 0) {
                len = read(fd, buf + pos, count);
                if (len <= 0)
-                       return len;
+                       return pos > len ? pos : len;
 
                count -= len;
                pos   += len;

Is the ternary '?' operator really needed here? In this part of code we know for sure that 'len' is less or equal to zero, also 'pos ' is a positive number or zero. Having just 'return pos' should be enough.

sbc/sbcenc.c static void encode(char *filename, int subbands, int · line 195

Quoted 9 lines
sbc/sbcenc.c +2 -0
@@ -188,6 +188,8 @@ static void encode(char *filename, int subbands, int
 bitpool, int joint,
 
                len = sbc_encode(&sbc, input, size,
                                        output, sizeof(output), &encoded);
+               if (len <= 0)
+                       break;
                if (len < size)
                        memmove(input, input + len, size - len);

If the return value is negative, there was some error. Probably showing some kind of error message is appropriate here instead of just silently bailing out.

Signature · 2 lines
Best regards,
Siarhei Siamashka
#10 lore raw

RE: [PATCH] SBC Encoder program

Quoted 9 lines

There are some parts that look a bit redundant/suspicious:

Quoted 7 lines
sbc/sbcenc.c +1 -1
@@ -47,7 +47,7 @@ static ssize_t __read(int fd, void *buf, size_t count)
        while (count > 0) {
                len = read(fd, buf + pos, count);
                if (len <= 0)
-                       return len;
+                       return pos > len ? pos : len;
 
Quoted 6 lines

sbc/sbcenc.c static ssize_t __read(int fd, void *buf, size_t count) · line 53

Quoted 2 lines
sbc/sbcenc.c +0 -0
                count -= len;
                pos   += len;

Is the ternary '?' operator really needed here? In this part of code we know for sure that 'len' is less or equal to zero, also 'pos ' is a positive number or zero. Having just 'return pos' should be enough.

No. If read returns an error, the this error is passed to __read, too.

Quoted 13 lines

sbc/sbcenc.c static void encode(char *filename, int subbands, int · line 195

Quoted 9 lines
sbc/sbcenc.c +2 -0
@@ -188,6 +188,8 @@ static void encode(char *filename, int subbands, int
 bitpool, int joint,
 
                len = sbc_encode(&sbc, input, size,
                                        output, sizeof(output), &encoded);
+               if (len <= 0)
+                       break;
                if (len < size)
                        memmove(input, input + len, size - len);

If the return value is negative, there was some error. Probably showing some kind of error message is appropriate here instead of just silently bailing out.

Yes, if len is -1 then errno depended message shall be printed and the program should quit.

CH

#11 lore raw

Re: [PATCH] SBC Encoder program

Quoted 8 lines · On Monday 05 January 2009 16:42:39 ext Christian Hoene wrote:
Quoted 8 lines

There are some parts that look a bit redundant/suspicious:

Quoted 7 lines
sbc/sbcenc.c +1 -1
@@ -47,7 +47,7 @@ static ssize_t __read(int fd, void *buf, size_t
 count) while (count > 0) {
                len = read(fd, buf + pos, count);
                if (len <= 0)
-                       return len;
+                       return pos > len ? pos : len;
 
Quoted 8 lines
Quoted 6 lines

sbc/sbcenc.c static ssize_t __read(int fd, void *buf, size_t count) · line 53

Quoted 2 lines
sbc/sbcenc.c +0 -0
                count -= len;
                pos   += len;

Is the ternary '?' operator really needed here? In this part of code we know for sure that 'len' is less or equal to zero, also 'pos ' is a positive number or zero. Having just 'return pos' should be enough.

No. If read returns an error, the this error is passed to __read, too.

I mean that 'return pos > len ? pos : len' and 'return pos' expressions are completely interexchangeable in this context.

Regarding this particular change itself, was there a reproducible problem in your testing that strictly required this fix? Was it involving sockets or pipes?

But the bug in 'sbc_get_codesize' is the unquestionable one for sure. Good job finding and fixing it.

Signature · 2 lines
Best regards,
Siarhei Siamashka
#12 lore raw

Re: [PATCH] SBC Encoder program

Quoted 1 line · On Monday 05 January 2009 17:18:44 ext Siarhei Siamashka wrote:

On Monday 05 January 2009 16:42:39 ext Christian Hoene wrote:

[...]

Quoted 2 lines

But the bug in 'sbc_get_codesize' is the unquestionable one for sure. Good job finding and fixing it.

Sorry, I meant the need to promote 'codesize' to uint16_t is unquestionable, not changing the 'sbc_get_codesize' function itself :)

Signature · 2 lines
Best regards,
Siarhei Siamashka
#13 lore raw

RE: [PATCH] SBC Encoder program

In reply to Siarhei Siamashka
Quoted 2 lines

I mean that 'return pos > len ? pos : len' and 'return pos' expressions are completely interexchangeable in this context.

They are slightly different but really, I do not care. Both fixes will work. At least one must be applied.
BTW: Why not use fread instead? ;-)

Greetings

Christian

#14 lore raw

Re: [PATCH] SBC Encoder program

Quoted 4 lines · On Monday 05 January 2009 17:33:10 ext Christian Hoene wrote:
Quoted 2 lines

I mean that 'return pos > len ? pos : len' and 'return pos' expressions are completely interexchangeable in this context.

They are slightly different but really, I do not care. Both fixes will work.

Yes, because they provide exactly identical results :-) In this sense, your fix is a bit broken, because it will never allow to pass through and return a negative error code.

Quoted 1 line

At least one must be applied. BTW: Why not use fread instead? ;-)

Agreed here. What's the point of having double underscored versions (such identifiers are reserved and not allowed to be used by programs) of read and write functions? Maybe they can be purged and just normal read/write calls could be used instead? Or alternatively fread/fwrite indeed?

Also there seems to be a performance issue: memove calls are redundant and can be avoided in SBC encoder program.

I can confirm that getting rid of the __read and __write functions (or fixing them with your patch or its modification) is the only change, needed to pass encoder test 08.

All the other changes are not directly related to this particular testcase, though a fix for 'codesize' is also nice to have. It would be required for a faster, memmove-free version of the encoder.

You have found and fixed two definite bugs (as I can see it now), that's very good. Now it would be nice to apply the fixes to git with the descriptive commit messages and move forward ;-)

Signature · 2 lines
Best regards,
Siarhei Siamashka
#15 lore raw

Re: [PATCH] SBC Encoder program

In reply to Siarhei Siamashka

2009/1/5 Siarhei Siamashka <[email protected]>:

Quoted 16 lines · On Monday 05 January 2009 16:42:39 ext Christian Hoene wrote:
Quoted 15 lines
Quoted 14 lines

There are some parts that look a bit redundant/suspicious:

sbc/sbcenc.c static ssize_t __read(int fd, void *buf, size_t · line 53

Quoted 9 lines
sbc/sbcenc.c +1 -1
@@ -47,7 +47,7 @@ static ssize_t __read(int fd, void *buf, size_t
 count) while (count > 0) {
                len = read(fd, buf + pos, count);
                if (len <= 0)
-                       return len;
+                       return pos > len ? pos : len;
 
                count -= len;
                pos   += len;

Is the ternary '?' operator really needed here? In this part of code we know for sure that 'len' is less or equal to zero, also 'pos ' is a positive number or zero. Having just 'return pos' should be enough.

Quoted 5 lines
Quoted 1 line

No. If read returns an error, the this error is passed to __read, too.

I mean that 'return pos > len ? pos : len' and 'return pos' expressions are completely interexchangeable in this context.

I wonder if you
shouldn't check before you read if position is not exceeding the length?

+ if (pos > len) // if we are beyond end then why
bother to read more
+ return pos;

Quoted 2 lines
                len = read(fd, buf + pos, count);
                if (len <= 0)

+ return len; //to report an error
-> - return len;
-> + return pos > len ? pos : len;
and as position is always >0 and len < 0 in this place it would always be reduced to pos

Quoted 3 lines
                count -= len;
                pos   += len;

Best wishes
Marcin Tolysz

#16 lore raw

Re: [PATCH] SBC Encoder program

In reply to Siarhei Siamashka
Quoted 18 lines · Siarhei Siamashka
Quoted 16 lines · On Monday 05 January 2009 16:42:39 ext Christian Hoene wrote:
Quoted 16 lines

There are some parts that look a bit redundant/suspicious:

sbc/sbcenc.c static ssize_t __read(int fd, void *buf, size_t · line 53

Quoted 10 lines
sbc/sbcenc.c +1 -1
@@ -47,7 +47,7 @@ static ssize_t __read(int fd, void *buf, size_t
 count) while (count > 0) {
                len = read(fd, buf + pos, count);
                if (len <= 0)
-                       return len;
+                       return pos > len ? pos : len;
 
                count -= len;
                pos   += len;

Is the ternary '?' operator really needed here? In this part of code we know for sure that 'len' is less or equal to zero, also 'pos ' is a positive number or zero. Having just 'return pos' should be enough.

Quoted 6 lines
Quoted 2 lines

No. If read returns an error, the this error is passed to __read, too.

I mean that 'return pos > len ? pos : len' and 'return pos' expressions are completely interexchangeable in this context.

Is it trying to say: return pos > 0 ? pos : len;

That's slightly functionally different, but makes it make sense to have a ternary there :) Dodgy to ignore the error though (when pos > 0).

#17 lore raw

Re: [PATCH] SBC Encoder program

In reply to Siarhei Siamashka

Hi all,

I want to apologize for my previous post :)

If what you want is to know is if there was an error & how much you red? that it might not work as you cann't return two value at once (not this way).

But if you want to now if there was an error on first reading?

sbc/sbcenc.c static ssize_t __read(int fd, void *buf, size_t count) · line 49

Quoted 4 lines
sbc/sbcenc.c +0 -0
@@ -47,7 +47,7 @@ static ssize_t __read(int fd, void *buf, size_t count)
        while (count > 0) {
                len = read(fd, buf + pos, count);
                if (len <= 0)

-> - return len;
-> + return pos > len ? pos : len;
+ return pos > 0 ? pos : len;

sbc/sbcenc.c static ssize_t __read(int fd, void *buf, size_t count) · line 53

Quoted 3 lines
sbc/sbcenc.c +0 -0
                count -= len;
                pos   += len;

this will return error code on first reading and the position successfully red (i.e. before the error) on sequent readings. (all assuming that pos==on entry)

You probably do not want to know the variant with position(=0) on the first reading and error code on subsequent(before your patch)

Before your patch it was an error code all the time After your patch it was the position all the time

Best Wishes
Marcin Tolysz

maintainer
#18 lore raw

Re: [PATCH] SBC Encoder program

In reply to Christian Hoene

Hi Christian,

Quoted 10 lines
Quoted 6 lines
Quoted 1 line

Yes, there was a bug in the encoder program. Attached the patch.

Coding style:

return pos > len ? pos : len;

Thanks. I corrected it and even found another bug.

Attached the new and corrected patch.

patch has been applied. Thanks.

Regards

Marcel

#20 lore raw

Is it possible that user change a stream path between Speaker and Bluetooth like normal phone?

Hi,

If you know or possible,
Could you explain me how to change stream path between Speaker or Bluetooth with BlueZ gstreamer plug-in like normal phone?

In my case, I could see it works well with a gstreamer output-selector element except some case.

1. User listens to Bluetooth Stereo Audio.
2. Switch off the headset when stream position is about 5 second from the start of the song.
3. Gstreamer Plug-in gst_avdtp_sink_render() we could see g_io_channel_write()works well even though headset is already off.
4. When stream position is about 9 second, g_io_channel_write() function is blocked....-.-;
5. When it is 14 second , g_io_chhanel_write() returns with timeout error.

It's problem that at 14 second , gstreamer could not do anything even if they know the error...

Related to this scenario , Do you have a good idea?

As a matter of fact, I try to avoid blocking , I change g_io_channel_write() as non-blocking call.
Because Non blocking call makes the situation to the gstreamer like normal condition. So we can change it easily as usual.

But the side effects occurs that music stream is mixed..when wireless communication doesn't work well-.-;

Regards.
Chanyeol

#21 lore raw

Re: Is it possible that user change a stream path between Speaker and Bluetooth like normal phone?

Quoted 8 lines · Chan-Yeol Park

Hi,

If you know or possible,
Could you explain me how to change stream path between Speaker or Bluetooth with BlueZ gstreamer plug-in like normal phone?

In my case, I could see it works well with a gstreamer output-selector element except some case.

Use PulseAudio's Bluetooth module, that's what it's made for.

#22 lore raw
In reply to Siarhei Siamashka
Quoted 6 lines
Quoted 2 lines

Good news! All tests passed!
http://net.cs.uni-tuebingen.de/html/nexgenvoip/

Thank you very much. Sorry for the delay with reply too.

I wonder if we need to do something about test 08?

Done.

Quoted 5 lines

Also could you try running a special test of version 4.25 with SBC_HIGH_PRECISION enabled? The point is that it uses the best possible precision for the analysis filter and minimizes effects of rounding errors (I would estimate that the precision should be even better than single precision floating point). Having the test run with

this

Quoted 3 lines

precision settings can expose some of the other minor bugs in the code (if they exist), which can now be hidden under the slight precision loss because of using 16-bit fixed point calculations.

I have uploaded the results to
http://net.cs.uni-tuebingen.de/html/nexgenvoip/4.25_high_precision

The encoding results did not change.

Best regards,

Christian

#23 lore raw
Quoted 17 lines · On Monday 05 January 2009 16:55:27 ext Christian Hoene wrote:
Quoted 6 lines
Quoted 2 lines

Good news! All tests passed!
http://net.cs.uni-tuebingen.de/html/nexgenvoip/

Thank you very much. Sorry for the delay with reply too.

I wonder if we need to do something about test 08?

Done.

Quoted 5 lines

Also could you try running a special test of version 4.25 with SBC_HIGH_PRECISION enabled? The point is that it uses the best possible precision for the analysis filter and minimizes effects of rounding errors (I would estimate that the precision should be even better than single precision floating point). Having the test run with

this

Quoted 8 lines
Quoted 3 lines

precision settings can expose some of the other minor bugs in the code (if they exist), which can now be hidden under the slight precision loss because of using 16-bit fixed point calculations.

I have uploaded the results to
http://net.cs.uni-tuebingen.de/html/nexgenvoip/4.25_high_precision

The encoding results did not change.

Great, it means that sbc encoder is mostly bug free (at least none of them shows up on this test set with all the available bugfixes applied).

Now a good improvement would be to tweak the default 16-bit version of analysis filter to improve its precision. It is a lot faster than the high precision variant and already provides sufficient quality :)

Signature · 2 lines
Best regards,
Siarhei Siamashka