xref: /linux/arch/x86/crypto/crc32c-pcl-intel-asm_64.S (revision 918731fa2808352387a872799d96f1e293e5f658)
16a8ce1efSTim Chen/*
26a8ce1efSTim Chen * Implement fast CRC32C with PCLMULQDQ instructions. (x86_64)
36a8ce1efSTim Chen *
4*918731faSTim Chen * The white papers on CRC32C calculations with PCLMULQDQ instruction can be
56a8ce1efSTim Chen * downloaded from:
6*918731faSTim Chen * http://www.intel.com/content/dam/www/public/us/en/documents/white-papers/crc-iscsi-polynomial-crc32-instruction-paper.pdf
7*918731faSTim Chen * http://www.intel.com/content/dam/www/public/us/en/documents/white-papers/fast-crc-computation-paper.pdf
86a8ce1efSTim Chen *
96a8ce1efSTim Chen * Copyright (C) 2012 Intel Corporation.
106a8ce1efSTim Chen *
116a8ce1efSTim Chen * Authors:
126a8ce1efSTim Chen *	Wajdi Feghali <wajdi.k.feghali@intel.com>
136a8ce1efSTim Chen *	James Guilford <james.guilford@intel.com>
146a8ce1efSTim Chen *	David Cote <david.m.cote@intel.com>
156a8ce1efSTim Chen *	Tim Chen <tim.c.chen@linux.intel.com>
166a8ce1efSTim Chen *
176a8ce1efSTim Chen * This software is available to you under a choice of one of two
186a8ce1efSTim Chen * licenses.  You may choose to be licensed under the terms of the GNU
196a8ce1efSTim Chen * General Public License (GPL) Version 2, available from the file
206a8ce1efSTim Chen * COPYING in the main directory of this source tree, or the
216a8ce1efSTim Chen * OpenIB.org BSD license below:
226a8ce1efSTim Chen *
236a8ce1efSTim Chen *     Redistribution and use in source and binary forms, with or
246a8ce1efSTim Chen *     without modification, are permitted provided that the following
256a8ce1efSTim Chen *     conditions are met:
266a8ce1efSTim Chen *
276a8ce1efSTim Chen *      - Redistributions of source code must retain the above
286a8ce1efSTim Chen *        copyright notice, this list of conditions and the following
296a8ce1efSTim Chen *        disclaimer.
306a8ce1efSTim Chen *
316a8ce1efSTim Chen *      - Redistributions in binary form must reproduce the above
326a8ce1efSTim Chen *        copyright notice, this list of conditions and the following
336a8ce1efSTim Chen *        disclaimer in the documentation and/or other materials
346a8ce1efSTim Chen *        provided with the distribution.
356a8ce1efSTim Chen *
366a8ce1efSTim Chen * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
376a8ce1efSTim Chen * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
386a8ce1efSTim Chen * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
396a8ce1efSTim Chen * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
406a8ce1efSTim Chen * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
416a8ce1efSTim Chen * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
426a8ce1efSTim Chen * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
436a8ce1efSTim Chen * SOFTWARE.
446a8ce1efSTim Chen */
456a8ce1efSTim Chen
46698a5abbSJussi Kivilinna#include <linux/linkage.h>
47698a5abbSJussi Kivilinna
486a8ce1efSTim Chen## ISCSI CRC 32 Implementation with crc32 and pclmulqdq Instruction
496a8ce1efSTim Chen
506a8ce1efSTim Chen.macro LABEL prefix n
516a8ce1efSTim Chen\prefix\n\():
526a8ce1efSTim Chen.endm
536a8ce1efSTim Chen
546a8ce1efSTim Chen.macro JMPTBL_ENTRY i
556a8ce1efSTim Chen.word crc_\i - crc_array
566a8ce1efSTim Chen.endm
576a8ce1efSTim Chen
586a8ce1efSTim Chen.macro JNC_LESS_THAN j
596a8ce1efSTim Chen	jnc less_than_\j
606a8ce1efSTim Chen.endm
616a8ce1efSTim Chen
626a8ce1efSTim Chen# Define threshold where buffers are considered "small" and routed to more
636a8ce1efSTim Chen# efficient "by-1" code. This "by-1" code only handles up to 255 bytes, so
646a8ce1efSTim Chen# SMALL_SIZE can be no larger than 255.
656a8ce1efSTim Chen
666a8ce1efSTim Chen#define SMALL_SIZE 200
676a8ce1efSTim Chen
686a8ce1efSTim Chen.if (SMALL_SIZE > 255)
696a8ce1efSTim Chen.error "SMALL_ SIZE must be < 256"
706a8ce1efSTim Chen.endif
716a8ce1efSTim Chen
726a8ce1efSTim Chen# unsigned int crc_pcl(u8 *buffer, int len, unsigned int crc_init);
736a8ce1efSTim Chen
74698a5abbSJussi KivilinnaENTRY(crc_pcl)
756a8ce1efSTim Chen#define    bufp		%rdi
766a8ce1efSTim Chen#define    bufp_dw	%edi
776a8ce1efSTim Chen#define    bufp_w	%di
786a8ce1efSTim Chen#define    bufp_b	%dil
796a8ce1efSTim Chen#define    bufptmp	%rcx
806a8ce1efSTim Chen#define    block_0	%rcx
816a8ce1efSTim Chen#define    block_1	%rdx
826a8ce1efSTim Chen#define    block_2	%r11
836a8ce1efSTim Chen#define    len		%rsi
846a8ce1efSTim Chen#define    len_dw	%esi
856a8ce1efSTim Chen#define    len_w	%si
866a8ce1efSTim Chen#define    len_b	%sil
876a8ce1efSTim Chen#define    crc_init_arg %rdx
886a8ce1efSTim Chen#define    tmp		%rbx
896a8ce1efSTim Chen#define    crc_init	%r8
906a8ce1efSTim Chen#define    crc_init_dw	%r8d
916a8ce1efSTim Chen#define    crc1		%r9
926a8ce1efSTim Chen#define    crc2		%r10
936a8ce1efSTim Chen
946a8ce1efSTim Chen	pushq   %rbx
956a8ce1efSTim Chen	pushq   %rdi
966a8ce1efSTim Chen	pushq   %rsi
976a8ce1efSTim Chen
986a8ce1efSTim Chen	## Move crc_init for Linux to a different
996a8ce1efSTim Chen	mov     crc_init_arg, crc_init
1006a8ce1efSTim Chen
1016a8ce1efSTim Chen	################################################################
1026a8ce1efSTim Chen	## 1) ALIGN:
1036a8ce1efSTim Chen	################################################################
1046a8ce1efSTim Chen
1056a8ce1efSTim Chen	mov     bufp, bufptmp		# rdi = *buf
1066a8ce1efSTim Chen	neg     bufp
1076a8ce1efSTim Chen	and     $7, bufp		# calculate the unalignment amount of
1086a8ce1efSTim Chen					# the address
1096a8ce1efSTim Chen	je      proc_block		# Skip if aligned
1106a8ce1efSTim Chen
1116a8ce1efSTim Chen	## If len is less than 8 and we're unaligned, we need to jump
1126a8ce1efSTim Chen	## to special code to avoid reading beyond the end of the buffer
1136a8ce1efSTim Chen	cmp     $8, len
1146a8ce1efSTim Chen	jae     do_align
1156a8ce1efSTim Chen	# less_than_8 expects length in upper 3 bits of len_dw
1166a8ce1efSTim Chen	# less_than_8_post_shl1 expects length = carryflag * 8 + len_dw[31:30]
1176a8ce1efSTim Chen	shl     $32-3+1, len_dw
1186a8ce1efSTim Chen	jmp     less_than_8_post_shl1
1196a8ce1efSTim Chen
1206a8ce1efSTim Chendo_align:
1216a8ce1efSTim Chen	#### Calculate CRC of unaligned bytes of the buffer (if any)
1226a8ce1efSTim Chen	movq    (bufptmp), tmp		# load a quadward from the buffer
1236a8ce1efSTim Chen	add     bufp, bufptmp		# align buffer pointer for quadword
1246a8ce1efSTim Chen					# processing
1256a8ce1efSTim Chen	sub     bufp, len		# update buffer length
1266a8ce1efSTim Chenalign_loop:
1276a8ce1efSTim Chen	crc32b  %bl, crc_init_dw 	# compute crc32 of 1-byte
1286a8ce1efSTim Chen	shr     $8, tmp			# get next byte
1296a8ce1efSTim Chen	dec     bufp
1306a8ce1efSTim Chen	jne     align_loop
1316a8ce1efSTim Chen
1326a8ce1efSTim Chenproc_block:
1336a8ce1efSTim Chen
1346a8ce1efSTim Chen	################################################################
1356a8ce1efSTim Chen	## 2) PROCESS  BLOCKS:
1366a8ce1efSTim Chen	################################################################
1376a8ce1efSTim Chen
1386a8ce1efSTim Chen	## compute num of bytes to be processed
1396a8ce1efSTim Chen	movq    len, tmp		# save num bytes in tmp
1406a8ce1efSTim Chen
1416a8ce1efSTim Chen	cmpq    $128*24, len
1426a8ce1efSTim Chen	jae     full_block
1436a8ce1efSTim Chen
1446a8ce1efSTim Chencontinue_block:
1456a8ce1efSTim Chen	cmpq    $SMALL_SIZE, len
1466a8ce1efSTim Chen	jb      small
1476a8ce1efSTim Chen
1486a8ce1efSTim Chen	## len < 128*24
1496a8ce1efSTim Chen	movq    $2731, %rax		# 2731 = ceil(2^16 / 24)
1506a8ce1efSTim Chen	mul     len_dw
1516a8ce1efSTim Chen	shrq    $16, %rax
1526a8ce1efSTim Chen
1536a8ce1efSTim Chen	## eax contains floor(bytes / 24) = num 24-byte chunks to do
1546a8ce1efSTim Chen
1556a8ce1efSTim Chen	## process rax 24-byte chunks (128 >= rax >= 0)
1566a8ce1efSTim Chen
1576a8ce1efSTim Chen	## compute end address of each block
1586a8ce1efSTim Chen	## block 0 (base addr + RAX * 8)
1596a8ce1efSTim Chen	## block 1 (base addr + RAX * 16)
1606a8ce1efSTim Chen	## block 2 (base addr + RAX * 24)
1616a8ce1efSTim Chen	lea     (bufptmp, %rax, 8), block_0
1626a8ce1efSTim Chen	lea     (block_0, %rax, 8), block_1
1636a8ce1efSTim Chen	lea     (block_1, %rax, 8), block_2
1646a8ce1efSTim Chen
1656a8ce1efSTim Chen	xor     crc1, crc1
1666a8ce1efSTim Chen	xor     crc2, crc2
1676a8ce1efSTim Chen
1686a8ce1efSTim Chen	## branch into array
1696a8ce1efSTim Chen	lea	jump_table(%rip), bufp
1706a8ce1efSTim Chen	movzxw  (bufp, %rax, 2), len
1716a8ce1efSTim Chen	offset=crc_array-jump_table
1726a8ce1efSTim Chen	lea     offset(bufp, len, 1), bufp
1736a8ce1efSTim Chen	jmp     *bufp
1746a8ce1efSTim Chen
1756a8ce1efSTim Chen	################################################################
1766a8ce1efSTim Chen	## 2a) PROCESS FULL BLOCKS:
1776a8ce1efSTim Chen	################################################################
1786a8ce1efSTim Chenfull_block:
1796a8ce1efSTim Chen	movq    $128,%rax
1806a8ce1efSTim Chen	lea     128*8*2(block_0), block_1
1816a8ce1efSTim Chen	lea     128*8*3(block_0), block_2
1826a8ce1efSTim Chen	add     $128*8*1, block_0
1836a8ce1efSTim Chen
1846a8ce1efSTim Chen	xor     crc1,crc1
1856a8ce1efSTim Chen	xor     crc2,crc2
1866a8ce1efSTim Chen
1876a8ce1efSTim Chen	# Fall thruogh into top of crc array (crc_128)
1886a8ce1efSTim Chen
1896a8ce1efSTim Chen	################################################################
1906a8ce1efSTim Chen	## 3) CRC Array:
1916a8ce1efSTim Chen	################################################################
1926a8ce1efSTim Chen
1936a8ce1efSTim Chencrc_array:
1946a8ce1efSTim Chen	i=128
1956a8ce1efSTim Chen.rept 128-1
1966a8ce1efSTim Chen.altmacro
1976a8ce1efSTim ChenLABEL crc_ %i
1986a8ce1efSTim Chen.noaltmacro
1996a8ce1efSTim Chen	crc32q   -i*8(block_0), crc_init
2006a8ce1efSTim Chen	crc32q   -i*8(block_1), crc1
2016a8ce1efSTim Chen	crc32q   -i*8(block_2), crc2
2026a8ce1efSTim Chen	i=(i-1)
2036a8ce1efSTim Chen.endr
2046a8ce1efSTim Chen
2056a8ce1efSTim Chen.altmacro
2066a8ce1efSTim ChenLABEL crc_ %i
2076a8ce1efSTim Chen.noaltmacro
2086a8ce1efSTim Chen	crc32q   -i*8(block_0), crc_init
2096a8ce1efSTim Chen	crc32q   -i*8(block_1), crc1
2106a8ce1efSTim Chen# SKIP  crc32  -i*8(block_2), crc2 ; Don't do this one yet
2116a8ce1efSTim Chen
2126a8ce1efSTim Chen	mov     block_2, block_0
2136a8ce1efSTim Chen
2146a8ce1efSTim Chen	################################################################
2156a8ce1efSTim Chen	## 4) Combine three results:
2166a8ce1efSTim Chen	################################################################
2176a8ce1efSTim Chen
2186a8ce1efSTim Chen	lea	(K_table-16)(%rip), bufp	# first entry is for idx 1
2196a8ce1efSTim Chen	shlq    $3, %rax			# rax *= 8
2206a8ce1efSTim Chen	subq    %rax, tmp			# tmp -= rax*8
2216a8ce1efSTim Chen	shlq    $1, %rax
2226a8ce1efSTim Chen	subq    %rax, tmp			# tmp -= rax*16
2236a8ce1efSTim Chen						# (total tmp -= rax*24)
2246a8ce1efSTim Chen	addq    %rax, bufp
2256a8ce1efSTim Chen
2266a8ce1efSTim Chen	movdqa  (bufp), %xmm0			# 2 consts: K1:K2
2276a8ce1efSTim Chen
2286a8ce1efSTim Chen	movq    crc_init, %xmm1			# CRC for block 1
2296a8ce1efSTim Chen	pclmulqdq $0x00,%xmm0,%xmm1		# Multiply by K2
2306a8ce1efSTim Chen
2316a8ce1efSTim Chen	movq    crc1, %xmm2			# CRC for block 2
2326a8ce1efSTim Chen	pclmulqdq $0x10, %xmm0, %xmm2		# Multiply by K1
2336a8ce1efSTim Chen
2346a8ce1efSTim Chen	pxor    %xmm2,%xmm1
2356a8ce1efSTim Chen	movq    %xmm1, %rax
2366a8ce1efSTim Chen	xor     -i*8(block_2), %rax
2376a8ce1efSTim Chen	mov     crc2, crc_init
2386a8ce1efSTim Chen	crc32   %rax, crc_init
2396a8ce1efSTim Chen
2406a8ce1efSTim Chen################################################################
2416a8ce1efSTim Chen## 5) Check for end:
2426a8ce1efSTim Chen################################################################
2436a8ce1efSTim Chen
2446a8ce1efSTim ChenLABEL crc_ 0
2456a8ce1efSTim Chen	mov     tmp, len
2466a8ce1efSTim Chen	cmp     $128*24, tmp
2476a8ce1efSTim Chen	jae     full_block
2486a8ce1efSTim Chen	cmp     $24, tmp
2496a8ce1efSTim Chen	jae     continue_block
2506a8ce1efSTim Chen
2516a8ce1efSTim Chenless_than_24:
2526a8ce1efSTim Chen	shl     $32-4, len_dw			# less_than_16 expects length
2536a8ce1efSTim Chen						# in upper 4 bits of len_dw
2546a8ce1efSTim Chen	jnc     less_than_16
2556a8ce1efSTim Chen	crc32q  (bufptmp), crc_init
2566a8ce1efSTim Chen	crc32q  8(bufptmp), crc_init
2576a8ce1efSTim Chen	jz      do_return
2586a8ce1efSTim Chen	add     $16, bufptmp
2596a8ce1efSTim Chen	# len is less than 8 if we got here
2606a8ce1efSTim Chen	# less_than_8 expects length in upper 3 bits of len_dw
2616a8ce1efSTim Chen	# less_than_8_post_shl1 expects length = carryflag * 8 + len_dw[31:30]
2626a8ce1efSTim Chen	shl     $2, len_dw
2636a8ce1efSTim Chen	jmp     less_than_8_post_shl1
2646a8ce1efSTim Chen
2656a8ce1efSTim Chen	#######################################################################
2666a8ce1efSTim Chen	## 6) LESS THAN 256-bytes REMAIN AT THIS POINT (8-bits of len are full)
2676a8ce1efSTim Chen	#######################################################################
2686a8ce1efSTim Chensmall:
2696a8ce1efSTim Chen	shl $32-8, len_dw		# Prepare len_dw for less_than_256
2706a8ce1efSTim Chen	j=256
2716a8ce1efSTim Chen.rept 5					# j = {256, 128, 64, 32, 16}
2726a8ce1efSTim Chen.altmacro
2736a8ce1efSTim ChenLABEL less_than_ %j			# less_than_j: Length should be in
2746a8ce1efSTim Chen					# upper lg(j) bits of len_dw
2756a8ce1efSTim Chen	j=(j/2)
2766a8ce1efSTim Chen	shl     $1, len_dw		# Get next MSB
2776a8ce1efSTim Chen	JNC_LESS_THAN %j
2786a8ce1efSTim Chen.noaltmacro
2796a8ce1efSTim Chen	i=0
2806a8ce1efSTim Chen.rept (j/8)
2816a8ce1efSTim Chen	crc32q  i(bufptmp), crc_init	# Compute crc32 of 8-byte data
2826a8ce1efSTim Chen	i=i+8
2836a8ce1efSTim Chen.endr
2846a8ce1efSTim Chen	jz      do_return		# Return if remaining length is zero
2856a8ce1efSTim Chen	add     $j, bufptmp		# Advance buf
2866a8ce1efSTim Chen.endr
2876a8ce1efSTim Chen
2886a8ce1efSTim Chenless_than_8:				# Length should be stored in
2896a8ce1efSTim Chen					# upper 3 bits of len_dw
2906a8ce1efSTim Chen	shl     $1, len_dw
2916a8ce1efSTim Chenless_than_8_post_shl1:
2926a8ce1efSTim Chen	jnc     less_than_4
2936a8ce1efSTim Chen	crc32l  (bufptmp), crc_init_dw	# CRC of 4 bytes
2946a8ce1efSTim Chen	jz      do_return		# return if remaining data is zero
2956a8ce1efSTim Chen	add     $4, bufptmp
2966a8ce1efSTim Chenless_than_4:				# Length should be stored in
2976a8ce1efSTim Chen					# upper 2 bits of len_dw
2986a8ce1efSTim Chen	shl     $1, len_dw
2996a8ce1efSTim Chen	jnc     less_than_2
3006a8ce1efSTim Chen	crc32w  (bufptmp), crc_init_dw	# CRC of 2 bytes
3016a8ce1efSTim Chen	jz      do_return		# return if remaining data is zero
3026a8ce1efSTim Chen	add     $2, bufptmp
3036a8ce1efSTim Chenless_than_2:				# Length should be stored in the MSB
3046a8ce1efSTim Chen					# of len_dw
3056a8ce1efSTim Chen	shl     $1, len_dw
3066a8ce1efSTim Chen	jnc     less_than_1
3076a8ce1efSTim Chen	crc32b  (bufptmp), crc_init_dw	# CRC of 1 byte
3086a8ce1efSTim Chenless_than_1:				# Length should be zero
3096a8ce1efSTim Chendo_return:
3106a8ce1efSTim Chen	movq    crc_init, %rax
3116a8ce1efSTim Chen	popq    %rsi
3126a8ce1efSTim Chen	popq    %rdi
3136a8ce1efSTim Chen	popq    %rbx
3146a8ce1efSTim Chen        ret
3156a8ce1efSTim Chen
3166a8ce1efSTim Chen        ################################################################
3176a8ce1efSTim Chen        ## jump table        Table is 129 entries x 2 bytes each
3186a8ce1efSTim Chen        ################################################################
3196a8ce1efSTim Chen.align 4
3206a8ce1efSTim Chenjump_table:
3216a8ce1efSTim Chen	i=0
3226a8ce1efSTim Chen.rept 129
3236a8ce1efSTim Chen.altmacro
3246a8ce1efSTim ChenJMPTBL_ENTRY %i
3256a8ce1efSTim Chen.noaltmacro
3266a8ce1efSTim Chen	i=i+1
3276a8ce1efSTim Chen.endr
328698a5abbSJussi Kivilinna
329698a5abbSJussi KivilinnaENDPROC(crc_pcl)
330698a5abbSJussi Kivilinna
3316a8ce1efSTim Chen	################################################################
3326a8ce1efSTim Chen	## PCLMULQDQ tables
3336a8ce1efSTim Chen	## Table is 128 entries x 2 quad words each
3346a8ce1efSTim Chen	################################################################
3356a8ce1efSTim Chen.data
3366a8ce1efSTim Chen.align 64
3376a8ce1efSTim ChenK_table:
3386a8ce1efSTim Chen        .quad 0x14cd00bd6,0x105ec76f0
3396a8ce1efSTim Chen        .quad 0x0ba4fc28e,0x14cd00bd6
3406a8ce1efSTim Chen        .quad 0x1d82c63da,0x0f20c0dfe
3416a8ce1efSTim Chen        .quad 0x09e4addf8,0x0ba4fc28e
3426a8ce1efSTim Chen        .quad 0x039d3b296,0x1384aa63a
3436a8ce1efSTim Chen        .quad 0x102f9b8a2,0x1d82c63da
3446a8ce1efSTim Chen        .quad 0x14237f5e6,0x01c291d04
3456a8ce1efSTim Chen        .quad 0x00d3b6092,0x09e4addf8
3466a8ce1efSTim Chen        .quad 0x0c96cfdc0,0x0740eef02
3476a8ce1efSTim Chen        .quad 0x18266e456,0x039d3b296
3486a8ce1efSTim Chen        .quad 0x0daece73e,0x0083a6eec
3496a8ce1efSTim Chen        .quad 0x0ab7aff2a,0x102f9b8a2
3506a8ce1efSTim Chen        .quad 0x1248ea574,0x1c1733996
3516a8ce1efSTim Chen        .quad 0x083348832,0x14237f5e6
3526a8ce1efSTim Chen        .quad 0x12c743124,0x02ad91c30
3536a8ce1efSTim Chen        .quad 0x0b9e02b86,0x00d3b6092
3546a8ce1efSTim Chen        .quad 0x018b33a4e,0x06992cea2
3556a8ce1efSTim Chen        .quad 0x1b331e26a,0x0c96cfdc0
3566a8ce1efSTim Chen        .quad 0x17d35ba46,0x07e908048
3576a8ce1efSTim Chen        .quad 0x1bf2e8b8a,0x18266e456
3586a8ce1efSTim Chen        .quad 0x1a3e0968a,0x11ed1f9d8
3596a8ce1efSTim Chen        .quad 0x0ce7f39f4,0x0daece73e
3606a8ce1efSTim Chen        .quad 0x061d82e56,0x0f1d0f55e
3616a8ce1efSTim Chen        .quad 0x0d270f1a2,0x0ab7aff2a
3626a8ce1efSTim Chen        .quad 0x1c3f5f66c,0x0a87ab8a8
3636a8ce1efSTim Chen        .quad 0x12ed0daac,0x1248ea574
3646a8ce1efSTim Chen        .quad 0x065863b64,0x08462d800
3656a8ce1efSTim Chen        .quad 0x11eef4f8e,0x083348832
3666a8ce1efSTim Chen        .quad 0x1ee54f54c,0x071d111a8
3676a8ce1efSTim Chen        .quad 0x0b3e32c28,0x12c743124
3686a8ce1efSTim Chen        .quad 0x0064f7f26,0x0ffd852c6
3696a8ce1efSTim Chen        .quad 0x0dd7e3b0c,0x0b9e02b86
3706a8ce1efSTim Chen        .quad 0x0f285651c,0x0dcb17aa4
3716a8ce1efSTim Chen        .quad 0x010746f3c,0x018b33a4e
3726a8ce1efSTim Chen        .quad 0x1c24afea4,0x0f37c5aee
3736a8ce1efSTim Chen        .quad 0x0271d9844,0x1b331e26a
3746a8ce1efSTim Chen        .quad 0x08e766a0c,0x06051d5a2
3756a8ce1efSTim Chen        .quad 0x093a5f730,0x17d35ba46
3766a8ce1efSTim Chen        .quad 0x06cb08e5c,0x11d5ca20e
3776a8ce1efSTim Chen        .quad 0x06b749fb2,0x1bf2e8b8a
3786a8ce1efSTim Chen        .quad 0x1167f94f2,0x021f3d99c
3796a8ce1efSTim Chen        .quad 0x0cec3662e,0x1a3e0968a
3806a8ce1efSTim Chen        .quad 0x19329634a,0x08f158014
3816a8ce1efSTim Chen        .quad 0x0e6fc4e6a,0x0ce7f39f4
3826a8ce1efSTim Chen        .quad 0x08227bb8a,0x1a5e82106
3836a8ce1efSTim Chen        .quad 0x0b0cd4768,0x061d82e56
3846a8ce1efSTim Chen        .quad 0x13c2b89c4,0x188815ab2
3856a8ce1efSTim Chen        .quad 0x0d7a4825c,0x0d270f1a2
3866a8ce1efSTim Chen        .quad 0x10f5ff2ba,0x105405f3e
3876a8ce1efSTim Chen        .quad 0x00167d312,0x1c3f5f66c
3886a8ce1efSTim Chen        .quad 0x0f6076544,0x0e9adf796
3896a8ce1efSTim Chen        .quad 0x026f6a60a,0x12ed0daac
3906a8ce1efSTim Chen        .quad 0x1a2adb74e,0x096638b34
3916a8ce1efSTim Chen        .quad 0x19d34af3a,0x065863b64
3926a8ce1efSTim Chen        .quad 0x049c3cc9c,0x1e50585a0
3936a8ce1efSTim Chen        .quad 0x068bce87a,0x11eef4f8e
3946a8ce1efSTim Chen        .quad 0x1524fa6c6,0x19f1c69dc
3956a8ce1efSTim Chen        .quad 0x16cba8aca,0x1ee54f54c
3966a8ce1efSTim Chen        .quad 0x042d98888,0x12913343e
3976a8ce1efSTim Chen        .quad 0x1329d9f7e,0x0b3e32c28
3986a8ce1efSTim Chen        .quad 0x1b1c69528,0x088f25a3a
3996a8ce1efSTim Chen        .quad 0x02178513a,0x0064f7f26
4006a8ce1efSTim Chen        .quad 0x0e0ac139e,0x04e36f0b0
4016a8ce1efSTim Chen        .quad 0x0170076fa,0x0dd7e3b0c
4026a8ce1efSTim Chen        .quad 0x141a1a2e2,0x0bd6f81f8
4036a8ce1efSTim Chen        .quad 0x16ad828b4,0x0f285651c
4046a8ce1efSTim Chen        .quad 0x041d17b64,0x19425cbba
4056a8ce1efSTim Chen        .quad 0x1fae1cc66,0x010746f3c
4066a8ce1efSTim Chen        .quad 0x1a75b4b00,0x18db37e8a
4076a8ce1efSTim Chen        .quad 0x0f872e54c,0x1c24afea4
4086a8ce1efSTim Chen        .quad 0x01e41e9fc,0x04c144932
4096a8ce1efSTim Chen        .quad 0x086d8e4d2,0x0271d9844
4106a8ce1efSTim Chen        .quad 0x160f7af7a,0x052148f02
4116a8ce1efSTim Chen        .quad 0x05bb8f1bc,0x08e766a0c
4126a8ce1efSTim Chen        .quad 0x0a90fd27a,0x0a3c6f37a
4136a8ce1efSTim Chen        .quad 0x0b3af077a,0x093a5f730
4146a8ce1efSTim Chen        .quad 0x04984d782,0x1d22c238e
4156a8ce1efSTim Chen        .quad 0x0ca6ef3ac,0x06cb08e5c
4166a8ce1efSTim Chen        .quad 0x0234e0b26,0x063ded06a
4176a8ce1efSTim Chen        .quad 0x1d88abd4a,0x06b749fb2
4186a8ce1efSTim Chen        .quad 0x04597456a,0x04d56973c
4196a8ce1efSTim Chen        .quad 0x0e9e28eb4,0x1167f94f2
4206a8ce1efSTim Chen        .quad 0x07b3ff57a,0x19385bf2e
4216a8ce1efSTim Chen        .quad 0x0c9c8b782,0x0cec3662e
4226a8ce1efSTim Chen        .quad 0x13a9cba9e,0x0e417f38a
4236a8ce1efSTim Chen        .quad 0x093e106a4,0x19329634a
4246a8ce1efSTim Chen        .quad 0x167001a9c,0x14e727980
4256a8ce1efSTim Chen        .quad 0x1ddffc5d4,0x0e6fc4e6a
4266a8ce1efSTim Chen        .quad 0x00df04680,0x0d104b8fc
4276a8ce1efSTim Chen        .quad 0x02342001e,0x08227bb8a
4286a8ce1efSTim Chen        .quad 0x00a2a8d7e,0x05b397730
4296a8ce1efSTim Chen        .quad 0x168763fa6,0x0b0cd4768
4306a8ce1efSTim Chen        .quad 0x1ed5a407a,0x0e78eb416
4316a8ce1efSTim Chen        .quad 0x0d2c3ed1a,0x13c2b89c4
4326a8ce1efSTim Chen        .quad 0x0995a5724,0x1641378f0
4336a8ce1efSTim Chen        .quad 0x19b1afbc4,0x0d7a4825c
4346a8ce1efSTim Chen        .quad 0x109ffedc0,0x08d96551c
4356a8ce1efSTim Chen        .quad 0x0f2271e60,0x10f5ff2ba
4366a8ce1efSTim Chen        .quad 0x00b0bf8ca,0x00bf80dd2
4376a8ce1efSTim Chen        .quad 0x123888b7a,0x00167d312
4386a8ce1efSTim Chen        .quad 0x1e888f7dc,0x18dcddd1c
4396a8ce1efSTim Chen        .quad 0x002ee03b2,0x0f6076544
4406a8ce1efSTim Chen        .quad 0x183e8d8fe,0x06a45d2b2
4416a8ce1efSTim Chen        .quad 0x133d7a042,0x026f6a60a
4426a8ce1efSTim Chen        .quad 0x116b0f50c,0x1dd3e10e8
4436a8ce1efSTim Chen        .quad 0x05fabe670,0x1a2adb74e
4446a8ce1efSTim Chen        .quad 0x130004488,0x0de87806c
4456a8ce1efSTim Chen        .quad 0x000bcf5f6,0x19d34af3a
4466a8ce1efSTim Chen        .quad 0x18f0c7078,0x014338754
4476a8ce1efSTim Chen        .quad 0x017f27698,0x049c3cc9c
4486a8ce1efSTim Chen        .quad 0x058ca5f00,0x15e3e77ee
4496a8ce1efSTim Chen        .quad 0x1af900c24,0x068bce87a
4506a8ce1efSTim Chen        .quad 0x0b5cfca28,0x0dd07448e
4516a8ce1efSTim Chen        .quad 0x0ded288f8,0x1524fa6c6
4526a8ce1efSTim Chen        .quad 0x059f229bc,0x1d8048348
4536a8ce1efSTim Chen        .quad 0x06d390dec,0x16cba8aca
4546a8ce1efSTim Chen        .quad 0x037170390,0x0a3e3e02c
4556a8ce1efSTim Chen        .quad 0x06353c1cc,0x042d98888
4566a8ce1efSTim Chen        .quad 0x0c4584f5c,0x0d73c7bea
4576a8ce1efSTim Chen        .quad 0x1f16a3418,0x1329d9f7e
4586a8ce1efSTim Chen        .quad 0x0531377e2,0x185137662
4596a8ce1efSTim Chen        .quad 0x1d8d9ca7c,0x1b1c69528
4606a8ce1efSTim Chen        .quad 0x0b25b29f2,0x18a08b5bc
4616a8ce1efSTim Chen        .quad 0x19fb2a8b0,0x02178513a
4626a8ce1efSTim Chen        .quad 0x1a08fe6ac,0x1da758ae0
4636a8ce1efSTim Chen        .quad 0x045cddf4e,0x0e0ac139e
4646a8ce1efSTim Chen        .quad 0x1a91647f2,0x169cf9eb0
4656a8ce1efSTim Chen        .quad 0x1a0f717c4,0x0170076fa
466