Performance optimizations for sbcenc utility
PATCH 1/1 Performance optimizations for sbcenc utility
Hello all,
The attached patch improves performance of sbcenc utility. This is not very useful in general and does not improve the SBC codec itself. But when using sbcenc utility for benchmarking, an extra unneeded overhead skews the results a bit and this might be worth fixing.
Before patch:
real 0m14.984s user 0m12.981s sys 0m1.924s
After patch:
real 0m12.279s user 0m11.865s sys 0m0.360s
Christian, you fixed some bugs in sbcenc a bit earlier. Could you please also check if all your testcases also work fine with this new patch applied and it does not break anything?
Signature · 2 lines
Best regards, Siarhei Siamashka
Quoted 1 line
From
b6ffabd4f7755ed6c968ecec0c8a0931eb6d7a66Mon Sep 17 00:00:00 2001
From: Siarhei Siamashka <[email protected]>
Date: Fri, 16 Jan 2009 17:23:54 +0200
Subject: [PATCH] Performance optimizations for sbcenc utility
Read and write buffers sizes increased, memmove overhead eliminated. Nonportable cast from 'unsigned char *' to 'struct au_header *' is now also resolved as part of the changes.
--- sbc/sbcenc.c | 122 ++++++++++++++++++++++++---------------------------------- 1 files changed, 50 insertions(+), 72 deletions(-)
sbc/sbcenc.c +50 -72
@@ -40,45 +40,15 @@static int verbose =0 ; -static ssize_t __read (int fd,void *buf,size_t count) -{ -ssize_t len, pos =0 ; - -while (count >0 ) { - len =read (fd, buf + pos, count); -if (len <=0 ) -return pos > len ? pos : len; - - count -= len; - pos += len; - } - -return pos; -} - -static ssize_t __write (int fd,const void *buf,size_t count) -{ -ssize_t len, pos =0 ; - -while (count >0 ) { - len =write (fd, buf + pos, count); -if (len <=0 ) -return len; - - count -= len; - pos += len; - } - -return pos; -} +#define BUF_SIZE 32768 +static unsigned char input[BUF_SIZE ], output[BUF_SIZE +BUF_SIZE /4 ];static void encode (char *filename,int subbands,int bitpool,int joint,int dualchannel,int snr,int blocks) { -struct au_header *au_hdr; -unsigned char input[2048 ], output[2048 ]; +struct au_header au_hdr;sbc_t sbc; -int fd, len, size, count, encoded, srate; +int fd, len, size, count, encoded, srate, codesize, nframes;if (strcmp (filename,"-" )) { fd =open (filename,O_RDONLY );
@@ -90,8 +60,8 @@ static void encode(char *filename, int subbands, int bitpool, int joint, }else fd =fileno (stdin); - len =__read (fd, input,sizeof (input)); -if (len <sizeof (*au_hdr)) { + len =read (fd, &au_hdr,sizeof (au_hdr)); +if (len <sizeof (au_hdr)) {if (fd >fileno (stderr))fprintf (stderr,"Can't read header from file %s: %s\n" , filename,strerror (errno));
@@ -100,19 +70,17 @@ static void encode(char *filename, int subbands, int bitpool, int joint, goto done; } - au_hdr = (struct au_header *) input; - -if (au_hdr->magic !=AU_MAGIC || -BE_INT (au_hdr->hdr_size) >128 || -BE_INT (au_hdr->hdr_size) <24 || -BE_INT (au_hdr->encoding) !=AU_FMT_LIN16 ) { +if (au_hdr.magic !=AU_MAGIC || +BE_INT (au_hdr.hdr_size) >128 || +BE_INT (au_hdr.hdr_size) <24 || +BE_INT (au_hdr.encoding) !=AU_FMT_LIN16 ) {fprintf (stderr,"Not in Sun/NeXT audio S16_BE format\n" ); goto done; }sbc_init (&sbc,0L ); -switch (BE_INT (au_hdr->sample_rate)) { +switch (BE_INT (au_hdr.sample_rate)) {case 16000 : sbc.frequency =SBC_FREQ_16000 ;break ;
@@ -127,11 +95,11 @@ static void encode(char *filename, int subbands, int bitpool, int joint,break ; } - srate =BE_INT (au_hdr->sample_rate); + srate =BE_INT (au_hdr.sample_rate); sbc.subbands = subbands ==4 ?SBC_SB_4 :SBC_SB_8 ; -if (BE_INT (au_hdr->channels) ==1 ) { +if (BE_INT (au_hdr.channels) ==1 ) { sbc.mode =SBC_MODE_MONO ;if (joint || dualchannel) { fprintf(stderr, "Audio is mono but joint or "
@@ -151,9 +119,9 @@ static void encode(char *filename, int subbands, int bitpool, int joint, } sbc.endian =SBC_BE ; - count =BE_INT (au_hdr->data_size); - size = len -BE_INT (au_hdr->hdr_size); -memmove (input, input +BE_INT (au_hdr->hdr_size), size); + count =BE_INT (au_hdr.data_size); + size = len -BE_INT (au_hdr.hdr_size); +memmove (input, input +BE_INT (au_hdr.hdr_size), size); sbc.bitpool = bitpool; sbc.allocation = snr ?SBC_AM_SNR :SBC_AM_LOUDNESS ;
@@ -172,37 +140,47 @@ static void encode(char *filename, int subbands, int bitpool, int joint, "STEREO" : "JOINTSTEREO"); } + codesize =sbc_get_codesize (&sbc); + nframes =sizeof (input) / codesize;while (1 ) { -if (size <sizeof (input)) { - len =__read (fd, input + size,sizeof (input) - size); -if (len ==0 && size ==0 ) -break ; - -if (len <0 ) { -perror ("Can't read audio data" ); +unsigned char *inp, *outp; +/* read data for up to 'nframes' frames of input data */ + size =read (fd, input, codesize * nframes); +if (size <0 ) { +/* Something really bad happened */ +perror ("Can't read audio data" ); +break ; + } +if (size < codesize) { +/* Not enough data for encoding even a single frame */ +break ; + } +/* encode all the data from the input buffer in a loop */ + inp = input; + outp = output; +while (size >= codesize) { + len =sbc_encode (&sbc, inp, codesize, + outp,sizeof (output) - (outp - output), + &encoded); +if (len != codesize || encoded <=0 ) { +fprintf (stderr, +"sbc_encode fail, len=%d, encoded=%d\n" , + len, encoded);break ; } - - size += len; + size -= len; + inp += len; + outp += encoded; } - - len =sbc_encode (&sbc, input, size, - output,sizeof (output), &encoded); -if (len <=0 ) -break ; -if (len < size) -memmove (input, input + len, size - len); - - size -= len; - - len =__write (fileno (stdout), output, encoded); -if (len ==0 ) -break ; - -if (len <0 || len != encoded) { + len =write (fileno (stdout), output, outp - output); +if (len != outp - output) {perror ("Can't write SBC output" );break ; } +if (size >= codesize) { +/* sbc_encode failure has been detected earlier */ +break ; + } }sbc_finish (&sbc);
Signature · 1 line
1.5.6.5
Marcel Holtmann
maintainerHi Siarhei,
Quoted 13 lines
The attached patch improves performance of sbcenc utility. This is not very useful in general and does not improve the SBC codec itself. But when using sbcenc utility for benchmarking, an extra unneeded overhead skews the results a bit and this might be worth fixing.
Before patch:
real 0m14.984s user 0m12.981s sys 0m1.924sAfter patch:
Quoted 8 lines
real 0m12.279s user 0m11.865s sys 0m0.360sChristian, you fixed some bugs in sbcenc a bit earlier. Could you please also check if all your testcases also work fine with this new patch applied and it does not break anything?
patch has been applied to make testing easier. Send fixes if it breaks something.
Regards
Marcel
Quoted 20 lines · On Sunday 18 January 2009 17:15:44 ext Marcel Holtmann wrote:
Hi Siarhei,
Quoted 18 lines
The attached patch improves performance of sbcenc utility. This is not very useful in general and does not improve the SBC codec itself. But when using sbcenc utility for benchmarking, an extra unneeded overhead skews the results a bit and this might be worth fixing.
Before patch:
real 0m14.984s user 0m12.981s sys 0m1.924sAfter patch:
real 0m12.279s user 0m11.865s sys 0m0.360s
Quoted 6 lines
Quoted 3 lines
Christian, you fixed some bugs in sbcenc a bit earlier. Could you please also check if all your testcases also work fine with this new patch applied and it does not break anything?
patch has been applied to make testing easier. Send fixes if it breaks something.
Well, appears that it really broke something. I forgot to update the part of code which handled non 24 byte headers. A fix is attached. Now it should work fine in all cases.
Best regards,
Siarhei Siamashka
Quoted 1 line
From
3ad8e7024ad599572673ba0aee883cab12c0262cMon Sep 17 00:00:00 2001
From: Siarhei Siamashka <[email protected]>
Date: Sun, 18 Jan 2009 23:10:00 +0200
Subject: [PATCH] Fix for sbcenc breakage when au file header size is larger than 24 bytes
--- sbc/sbcenc.c | 24 +++++++++++++++++------- 1 files changed, 17 insertions(+), 7 deletions(-)
sbc/sbcenc.c +17 -7
@@ -48,7 +48,13 @@ static void encode(char *filename, int subbands, int bitpool, int joint, {struct au_header au_hdr;sbc_t sbc; -int fd, len, size, count, encoded, srate, codesize, nframes; +int fd, len, size, encoded, srate, codesize, nframes; + +if (sizeof (au_hdr) !=24 ) { +/* Sanity check just in case */ +fprintf (stderr,"FIXME: sizeof(au_hdr) != 24\n" ); +return ; + }if (strcmp (filename,"-" )) { fd =open (filename,O_RDONLY );
@@ -72,7 +78,7 @@ static void encode(char *filename, int subbands, int bitpool, int joint,if (au_hdr.magic !=AU_MAGIC ||BE_INT (au_hdr.hdr_size) >128 || -BE_INT (au_hdr.hdr_size) <24 || +BE_INT (au_hdr.hdr_size) <sizeof (au_hdr) ||BE_INT (au_hdr.encoding) !=AU_FMT_LIN16 ) {fprintf (stderr,"Not in Sun/NeXT audio S16_BE format\n" ); goto done;
@@ -119,9 +125,9 @@ static void encode(char *filename, int subbands, int bitpool, int joint, } sbc.endian =SBC_BE ; - count =BE_INT (au_hdr.data_size); - size = len -BE_INT (au_hdr.hdr_size); -memmove (input, input +BE_INT (au_hdr.hdr_size), size); +/* Skip extra bytes of the header if any */ +if (read (fd, input,BE_INT (au_hdr.hdr_size) - len) <0 ) + goto done; sbc.bitpool = bitpool; sbc.allocation = snr ?SBC_AM_SNR :SBC_AM_LOUDNESS ;
@@ -177,8 +183,12 @@ static void encode(char *filename, int subbands, int bitpool, int joint,perror ("Can't write SBC output" );break ; } -if (size >= codesize) { -/* sbc_encode failure has been detected earlier */ +if (size !=0 ) { +/* +* sbc_encode failure has been detected earlier or end +* of file reached (have trailing partial data which is +* insufficient to encode SBC frame) +*/ break ; } }
Signature · 1 line
1.5.6.5
Marcel Holtmann
maintainerHi Siarhei,
Quoted 22 lines
Quoted 22 lines
Quoted 21 lines
The attached patch improves performance of sbcenc utility. This is not very useful in general and does not improve the SBC codec itself. But when using sbcenc utility for benchmarking, an extra unneeded overhead skews the results a bit and this might be worth fixing.
Before patch:
real 0m14.984s user 0m12.981s sys 0m1.924sAfter patch:
real 0m12.279s user 0m11.865s sys 0m0.360sChristian, you fixed some bugs in sbcenc a bit earlier. Could you please also check if all your testcases also work fine with this new patch applied and it does not break anything?
Quoted 6 lines
Quoted 2 lines
patch has been applied to make testing easier. Send fixes if it breaks something.
Well, appears that it really broke something. I forgot to update the part of code which handled non 24 byte headers. A fix is attached. Now it should work fine in all cases.
patch has been applied. Thanks.
Regards
Marcel