From: Christophe LEROY Subject: Re: [PATCH] Performance Improvement in CRC16 Calculations. Date: Thu, 16 Aug 2018 17:47:25 +0200 Message-ID: References: <1533928331-21303-1-git-send-email-jeff.lien@wdc.com> <20180810201601.GA80850@gmail.com> Mime-Version: 1.0 Content-Type: text/plain; charset=utf-8; format=flowed Content-Transfer-Encoding: 8bit Cc: "linux-kernel@vger.kernel.org" , "linux-crypto@vger.kernel.org" , "linux-block@vger.kernel.org" , "linux-scsi@vger.kernel.org" , "herbert@gondor.apana.org.au" , "tim.c.chen@linux.intel.com" , "martin.petersen@oracle.com" , David Darrington , Jeff Furlong To: Jeffrey Lien , Eric Biggers Return-path: In-Reply-To: Content-Language: fr Sender: linux-kernel-owner@vger.kernel.org List-Id: linux-crypto.vger.kernel.org Hi, Don't be so sure that slice by 16 provides the best performance. Some time ago, I did the comparison for CRC32 between slice by 4 and slice by 8 (as included in the kernel), and the result is the following on an mpc8xx (powerpc/32) With CONFIG_CRC32_SLICEBY8: [ 1.109204] crc32: CRC_LE_BITS = 64, CRC_BE BITS = 64 [ 1.114401] crc32: self tests passed, processed 225944 bytes in 15118910 nsec [ 1.130655] crc32c: CRC_LE_BITS = 64 [ 1.134235] crc32c: self tests passed, processed 225944 bytes in 4479879 nsec With CONFIG_CRC32_SLICEBY4: [ 1.097129] crc32: CRC_LE_BITS = 32, CRC_BE BITS = 32 [ 1.101878] crc32: self tests passed, processed 225944 bytes in 8616242 nsec [ 1.116298] crc32c: CRC_LE_BITS = 32 [ 1.119607] crc32c: self tests passed, processed 225944 bytes in 3289576 nsec As you can see, slice by 4 is better than slice by 8 on that CPU. So I'm sure it is worth doing the test for CRC16 as well. Christophe Le 16/08/2018 à 16:02, Jeffrey Lien a écrit : > Eric, > We did not test the slice by 4 or 8 tables. I'm not sure of the value of doing that since the slice by 16 will provide the best performance gain. If I'm missing anything here, please let me know. > > I'm working on a new version of the patch based on the feedback from others and will also change the pointer variables to start with p and fix the indenting you mentioned below in the new version of the patch. > > Thanks > > Jeff Lien > > -----Original Message----- > From: Eric Biggers [mailto:ebiggers@kernel.org] > Sent: Friday, August 10, 2018 3:16 PM > To: Jeffrey Lien > Cc: linux-kernel@vger.kernel.org; linux-crypto@vger.kernel.org; linux-block@vger.kernel.org; linux-scsi@vger.kernel.org; herbert@gondor.apana.org.au; tim.c.chen@linux.intel.com; martin.petersen@oracle.com; David Darrington ; Jeff Furlong > Subject: Re: [PATCH] Performance Improvement in CRC16 Calculations. > > On Fri, Aug 10, 2018 at 02:12:11PM -0500, Jeff Lien wrote: >> This patch provides a performance improvement for the CRC16 >> calculations done in read/write workloads using the T10 Type 1/2/3 >> guard field. For example, today with sequential write workloads (one >> thread/CPU of IO) we consume 100% of the CPU because of the CRC16 >> computation bottleneck. Today's block devices are considerably >> faster, but the CRC16 calculation prevents folks from utilizing the >> throughput of such devices. To speed up this calculation and expose >> the block device throughput, we slice the old single byte for loop into a 16 byte for loop, with a larger CRC table to match. The result has shown 5x performance improvements on various big endian and little endian systems running the 4.18.0 kernel version. >> >> FIO Sequential Write, 64K Block Size, Queue Depth 64 >> BE Base Kernel: bw=201.5 MiB/s >> BE Modified CRC Calc: bw=968.1 MiB/s >> 4.80x performance improvement >> >> LE Base Kernel: bw=357 MiB/s >> LE Modified CRC Calc: bw=1964 MiB/s >> 5.51x performance improvement >> >> FIO Sequential Read, 64K Block Size, Queue Depth 64 >> BE Base Kernel: bw=611.2 MiB/s >> BE Modified CRC calc: bw=684.9 MiB/s >> 1.12x performance improvement >> >> LE Base Kernel: bw=797 MiB/s >> LE Modified CRC Calc: bw=2730 MiB/s >> 3.42x performance improvement > > Did you also test the slice-by-4 (requires 2048-byte table) and slice-by-8 (requires 4096-byte table) methods? Your proposal is slice-by-16 (requires 8192-byte table); the original was slice-by-1 (requires 512-byte table). > >> __u16 crc_t10dif_generic(__u16 crc, const unsigned char *buffer, >> size_t len) { >> - unsigned int i; >> + const __u8 *i = (const __u8 *)buffer; >> + const __u8 *i_end = i + len; >> + const __u8 *i_last16 = i + (len / 16 * 16); > > 'i' is normally a loop counter, not a pointer. > Use 'p', 'p_end', and 'p_last16'. > >> >> - for (i = 0 ; i < len ; i++) >> - crc = (crc << 8) ^ t10_dif_crc_table[((crc >> 8) ^ buffer[i]) & 0xff]; >> + for (; i < i_last16; i += 16) { >> + crc = t10_dif_crc_table[15][i[0] ^ (__u8)(crc >> 8)] ^ >> + t10_dif_crc_table[14][i[1] ^ (__u8)(crc >> 0)] ^ >> + t10_dif_crc_table[13][i[2]] ^ >> + t10_dif_crc_table[12][i[3]] ^ >> + t10_dif_crc_table[11][i[4]] ^ >> + t10_dif_crc_table[10][i[5]] ^ >> + t10_dif_crc_table[9][i[6]] ^ >> + t10_dif_crc_table[8][i[7]] ^ >> + t10_dif_crc_table[7][i[8]] ^ >> + t10_dif_crc_table[6][i[9]] ^ >> + t10_dif_crc_table[5][i[10]] ^ >> + t10_dif_crc_table[4][i[11]] ^ >> + t10_dif_crc_table[3][i[12]] ^ >> + t10_dif_crc_table[2][i[13]] ^ >> + t10_dif_crc_table[1][i[14]] ^ >> + t10_dif_crc_table[0][i[15]]; >> + } > > Please indent this properly. > > crc = t10_dif_crc_table[15][i[0] ^ (__u8)(crc >> 8)] ^ > t10_dif_crc_table[14][i[1] ^ (__u8)(crc >> 0)] ^ > t10_dif_crc_table[13][i[2]] ^ > t10_dif_crc_table[12][i[3]] ^ > t10_dif_crc_table[11][i[4]] ^ > ... > > - Eric >