Return-Path: From: Siarhei Siamashka To: "ext Marcel Holtmann" Subject: Re: [PATCH] Performance optimizations for sbcenc utility Date: Mon, 19 Jan 2009 13:05:36 +0200 Cc: linux-bluetooth@vger.kernel.org References: <200901161812.45754.siarhei.siamashka@nokia.com> <1232291744.5095.14.camel@californication> In-Reply-To: <1232291744.5095.14.camel@californication> MIME-Version: 1.0 Content-Type: Multipart/Mixed; boundary="Boundary-00=_A6FdJDf8l9D/fX0" Message-Id: <200901191305.36942.siarhei.siamashka@nokia.com> List-ID: --Boundary-00=_A6FdJDf8l9D/fX0 Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: 7bit Content-Disposition: inline On Sunday 18 January 2009 17:15:44 ext Marcel Holtmann wrote: > Hi Siarhei, > > > 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? > > 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 --Boundary-00=_A6FdJDf8l9D/fX0 Content-Type: text/x-diff; charset="utf-8"; name="0001-Fix-for-sbcenc-breakage-when-au-file-header-size-is.patch" Content-Transfer-Encoding: 8bit Content-Disposition: inline; filename="0001-Fix-for-sbcenc-breakage-when-au-file-header-size-is.patch" >From 3ad8e7024ad599572673ba0aee883cab12c0262c Mon Sep 17 00:00:00 2001 From: Siarhei Siamashka 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(-) diff --git a/sbc/sbcenc.c b/sbc/sbcenc.c index 388be2a..fba9be3 100644 --- a/sbc/sbcenc.c +++ b/sbc/sbcenc.c @@ -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; } } -- 1.5.6.5 --Boundary-00=_A6FdJDf8l9D/fX0--