| David Howells | b920de1 | 2008-02-08 04:19:31 -0800 | [diff] [blame] | 1 | /* Optimised simple memory checksum |
| 2 | * |
| 3 | * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved. |
| 4 | * Written by David Howells (dhowells@redhat.com) |
| 5 | * |
| 6 | * This program is free software; you can redistribute it and/or |
| 7 | * modify it under the terms of the GNU General Public Licence |
| 8 | * as published by the Free Software Foundation; either version |
| 9 | * 2 of the Licence, or (at your option) any later version. |
| 10 | */ |
| 11 | #include <asm/cache.h> |
| 12 | |
| 13 | .section .text |
| 14 | .balign L1_CACHE_BYTES |
| 15 | |
| 16 | ############################################################################### |
| 17 | # |
| 18 | # unsigned int do_csum(const unsigned char *buff, size_t len) |
| 19 | # |
| 20 | ############################################################################### |
| 21 | .globl do_csum |
| 22 | .type do_csum,@function |
| 23 | do_csum: |
| 24 | movm [d2,d3],(sp) |
| 25 | mov d0,(12,sp) |
| 26 | mov d1,(16,sp) |
| 27 | mov d1,d2 # count |
| 28 | mov d0,a0 # buff |
| 29 | clr d1 # accumulator |
| 30 | |
| 31 | cmp +0,d2 |
| 32 | beq do_csum_done # return if zero-length buffer |
| 33 | |
| 34 | # 4-byte align the buffer pointer |
| 35 | btst +3,a0 |
| 36 | beq do_csum_now_4b_aligned |
| 37 | |
| 38 | btst +1,a0 |
| 39 | beq do_csum_addr_not_odd |
| 40 | movbu (a0),d0 |
| 41 | inc a0 |
| 42 | asl +8,d0 |
| 43 | add d0,d1 |
| 44 | addc +0,d1 |
| 45 | add -1,d2 |
| 46 | do_csum_addr_not_odd: |
| 47 | |
| 48 | cmp +2,d2 |
| 49 | bcs do_csum_fewer_than_4 |
| 50 | btst +2,a0 |
| 51 | beq do_csum_now_4b_aligned |
| 52 | movhu (a0+),d0 |
| 53 | add d0,d1 |
| 54 | addc +0,d1 |
| 55 | add -2,d2 |
| 56 | cmp +4,d2 |
| 57 | bcs do_csum_fewer_than_4 |
| 58 | |
| 59 | do_csum_now_4b_aligned: |
| 60 | # we want to checksum as much as we can in chunks of 32 bytes |
| 61 | cmp +31,d2 |
| 62 | bls do_csum_remainder # 4-byte aligned remainder |
| 63 | |
| 64 | add -32,d2 |
| 65 | mov +32,d3 |
| 66 | |
| 67 | do_csum_loop: |
| 68 | mov (a0+),d0 |
| 69 | add d0,d1 |
| 70 | mov (a0+),e0 |
| 71 | addc e0,d1 |
| 72 | mov (a0+),e1 |
| 73 | addc e1,d1 |
| 74 | mov (a0+),e3 |
| 75 | addc e3,d1 |
| 76 | mov (a0+),d0 |
| 77 | addc d0,d1 |
| 78 | mov (a0+),e0 |
| 79 | addc e0,d1 |
| 80 | mov (a0+),e1 |
| 81 | addc e1,d1 |
| 82 | mov (a0+),e3 |
| 83 | addc e3,d1 |
| 84 | addc +0,d1 |
| 85 | |
| 86 | sub d3,d2 |
| 87 | bcc do_csum_loop |
| 88 | |
| 89 | add d3,d2 |
| 90 | beq do_csum_done |
| 91 | |
| 92 | do_csum_remainder: |
| 93 | # cut 16-31 bytes down to 0-15 |
| 94 | cmp +16,d2 |
| 95 | bcs do_csum_fewer_than_16 |
| 96 | mov (a0+),d0 |
| 97 | add d0,d1 |
| 98 | mov (a0+),e0 |
| 99 | addc e0,d1 |
| 100 | mov (a0+),e1 |
| 101 | addc e1,d1 |
| 102 | mov (a0+),e3 |
| 103 | addc e3,d1 |
| 104 | addc +0,d1 |
| 105 | add -16,d2 |
| 106 | beq do_csum_done |
| 107 | |
| 108 | do_csum_fewer_than_16: |
| 109 | # copy the remaining whole words |
| 110 | cmp +4,d2 |
| 111 | bcs do_csum_fewer_than_4 |
| 112 | cmp +8,d2 |
| 113 | bcs do_csum_one_word |
| 114 | cmp +12,d2 |
| 115 | bcs do_csum_two_words |
| 116 | mov (a0+),d0 |
| 117 | add d0,d1 |
| 118 | addc +0,d1 |
| 119 | do_csum_two_words: |
| 120 | mov (a0+),d0 |
| 121 | add d0,d1 |
| 122 | addc +0,d1 |
| 123 | do_csum_one_word: |
| 124 | mov (a0+),d0 |
| 125 | add d0,d1 |
| 126 | addc +0,d1 |
| 127 | |
| 128 | do_csum_fewer_than_4: |
| 129 | and +3,d2 |
| 130 | beq do_csum_done |
| 131 | xor_cmp d0,d0,+2,d2 |
| 132 | bcs do_csum_fewer_than_2 |
| 133 | movhu (a0+),d0 |
| 134 | do_csum_fewer_than_2: |
| 135 | and +1,d2 |
| 136 | beq do_csum_add_last_bit |
| 137 | movbu (a0),d3 |
| 138 | add d3,d0 |
| 139 | do_csum_add_last_bit: |
| 140 | add d0,d1 |
| 141 | addc +0,d1 |
| 142 | |
| 143 | do_csum_done: |
| 144 | # compress the checksum down to 16 bits |
| 145 | mov +0xffff0000,d2 |
| 146 | and d1,d2 |
| 147 | asl +16,d1 |
| 148 | add d2,d1,d0 |
| 149 | addc +0xffff,d0 |
| 150 | lsr +16,d0 |
| 151 | |
| 152 | # flip the halves of the word result if the buffer was oddly aligned |
| 153 | mov (12,sp),d1 |
| 154 | and +1,d1 |
| 155 | beq do_csum_not_oddly_aligned |
| 156 | swaph d0,d0 # exchange bits 15:8 with 7:0 |
| 157 | |
| 158 | do_csum_not_oddly_aligned: |
| 159 | ret [d2,d3],8 |
| 160 | |
| 161 | do_csum_end: |
| 162 | .size do_csum, do_csum_end-do_csum |