From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([140.186.70.92]:50288) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1QN2Nb-0002VV-H2 for qemu-devel@nongnu.org; Thu, 19 May 2011 08:31:06 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1QN2Na-0003v3-9T for qemu-devel@nongnu.org; Thu, 19 May 2011 08:31:03 -0400 Received: from mx1.redhat.com ([209.132.183.28]:2799) by eggs.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1QN2Na-0003ul-12 for qemu-devel@nongnu.org; Thu, 19 May 2011 08:31:02 -0400 From: Kevin Wolf Date: Thu, 19 May 2011 14:33:24 +0200 Message-Id: <1305808412-16994-11-git-send-email-kwolf@redhat.com> In-Reply-To: <1305808412-16994-1-git-send-email-kwolf@redhat.com> References: <1305808412-16994-1-git-send-email-kwolf@redhat.com> Subject: [Qemu-devel] [PATCH 10/18] qemu_img: is_not_zero() optimization List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: anthony@codemonkey.ws Cc: kwolf@redhat.com, qemu-devel@nongnu.org From: Dmitry Konishchev I run qemu-img under profiler and realized, that most of CPU time is consumed by is_not_zero() function. I had made a couple of optimizations on it and got the following output for `time qemu-img convert -O qcow2 volume.qcow2 snapshot.qcow2`: Original qemu-img: real 0m56.159s user 0m34.670s sys 0m12.079s Patched qemu-img: real 0m34.805s user 0m18.445s sys 0m12.552s Signed-off-by: Dmitry Konishchev Signed-off-by: Kevin Wolf --- qemu-img.c | 29 ++++++++++++++++++++++++++--- 1 files changed, 26 insertions(+), 3 deletions(-) diff --git a/qemu-img.c b/qemu-img.c index 1da5484..4f162d1 100644 --- a/qemu-img.c +++ b/qemu-img.c @@ -496,14 +496,37 @@ static int img_commit(int argc, char **argv) return 0; } +/* + * Checks whether the sector is not a zero sector. + * + * Attention! The len must be a multiple of 4 * sizeof(long) due to + * restriction of optimizations in this function. + */ static int is_not_zero(const uint8_t *sector, int len) { + /* + * Use long as the biggest available internal data type that fits into the + * CPU register and unroll the loop to smooth out the effect of memory + * latency. + */ + int i; - len >>= 2; - for(i = 0;i < len; i++) { - if (((uint32_t *)sector)[i] != 0) + long d0, d1, d2, d3; + const long * const data = (const long *) sector; + + len /= sizeof(long); + + for(i = 0; i < len; i += 4) { + d0 = data[i + 0]; + d1 = data[i + 1]; + d2 = data[i + 2]; + d3 = data[i + 3]; + + if (d0 || d1 || d2 || d3) { return 1; + } } + return 0; } -- 1.7.2.3