SiSK / EthernetInterface

fdsf

Dependents:   sisk_proj_stat MQTT Hello_FXOS8700Q WireFSHandControl ... more

lwip-sys/arch/checksum.c@0:d7bd7384a37c, 2016-12-07 (annotated)

Committer:
grzemich
Date:
Wed Dec 07 23:47:50 2016 +0000
Revision:
0:d7bd7384a37c
dgd

Who changed what in which revision?

UserRevisionLine numberNew contents of line
grzemich 0:d7bd7384a37c 1 /* Copyright (C) 2013 - Adam Green (https://github.com/adamgreen)
grzemich 0:d7bd7384a37c 2
grzemich 0:d7bd7384a37c 3 Licensed under the Apache License, Version 2.0 (the "License");
grzemich 0:d7bd7384a37c 4 you may not use this file except in compliance with the License.
grzemich 0:d7bd7384a37c 5 You may obtain a copy of the License at
grzemich 0:d7bd7384a37c 6
grzemich 0:d7bd7384a37c 7 http://www.apache.org/licenses/LICENSE-2.0
grzemich 0:d7bd7384a37c 8
grzemich 0:d7bd7384a37c 9 Unless required by applicable law or agreed to in writing, software
grzemich 0:d7bd7384a37c 10 distributed under the License is distributed on an "AS IS" BASIS,
grzemich 0:d7bd7384a37c 11 WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
grzemich 0:d7bd7384a37c 12 See the License for the specific language governing permissions and
grzemich 0:d7bd7384a37c 13 limitations under the License.
grzemich 0:d7bd7384a37c 14 */
grzemich 0:d7bd7384a37c 15 #if defined(TOOLCHAIN_GCC) && defined(__thumb2__)
grzemich 0:d7bd7384a37c 16
grzemich 0:d7bd7384a37c 17
grzemich 0:d7bd7384a37c 18 /* This is a hand written Thumb-2 assembly language version of the
grzemich 0:d7bd7384a37c 19 algorithm 3 version of lwip_standard_chksum in lwIP's inet_chksum.c. It
grzemich 0:d7bd7384a37c 20 performs the checksumming 32-bits at a time and even unrolls the loop to
grzemich 0:d7bd7384a37c 21 perform two of these 32-bit adds per loop iteration.
grzemich 0:d7bd7384a37c 22
grzemich 0:d7bd7384a37c 23 Returns:
grzemich 0:d7bd7384a37c 24 16-bit 1's complement summation (not inversed).
grzemich 0:d7bd7384a37c 25
grzemich 0:d7bd7384a37c 26 NOTE: This function does return a uint16_t from the assembly language code
grzemich 0:d7bd7384a37c 27 but is marked as void so that GCC doesn't issue warning because it
grzemich 0:d7bd7384a37c 28 doesn't know about this low level return.
grzemich 0:d7bd7384a37c 29 */
grzemich 0:d7bd7384a37c 30 __attribute__((naked)) void /*uint16_t*/ thumb2_checksum(const void* pData, int length)
grzemich 0:d7bd7384a37c 31 {
grzemich 0:d7bd7384a37c 32 __asm (
grzemich 0:d7bd7384a37c 33 ".syntax unified\n"
grzemich 0:d7bd7384a37c 34 ".thumb\n"
grzemich 0:d7bd7384a37c 35
grzemich 0:d7bd7384a37c 36 // Push non-volatile registers we use on stack. Push link register too to
grzemich 0:d7bd7384a37c 37 // keep stack 8-byte aligned and allow single pop to restore and return.
grzemich 0:d7bd7384a37c 38 " push {r4, lr}\n"
grzemich 0:d7bd7384a37c 39 // Initialize sum, r2, to 0.
grzemich 0:d7bd7384a37c 40 " movs r2, #0\n"
grzemich 0:d7bd7384a37c 41 // Remember whether pData was at odd address in r3. This is used later to
grzemich 0:d7bd7384a37c 42 // know if it needs to swap the result since the summation will be done at
grzemich 0:d7bd7384a37c 43 // an offset of 1, rather than 0.
grzemich 0:d7bd7384a37c 44 " ands r3, r0, #1\n"
grzemich 0:d7bd7384a37c 45 // Need to 2-byte align? If not skip ahead.
grzemich 0:d7bd7384a37c 46 " beq 1$\n"
grzemich 0:d7bd7384a37c 47 // We can return if there are no bytes to sum.
grzemich 0:d7bd7384a37c 48 " cbz r1, 9$\n"
grzemich 0:d7bd7384a37c 49
grzemich 0:d7bd7384a37c 50 // 2-byte align.
grzemich 0:d7bd7384a37c 51 // Place the first data byte in odd summation location since it needs to be
grzemich 0:d7bd7384a37c 52 // swapped later. It's ok to overwrite r2 here as it only had a value of 0
grzemich 0:d7bd7384a37c 53 // up until now. Advance r0 pointer and decrement r1 length as we go.
grzemich 0:d7bd7384a37c 54 " ldrb r2, [r0], #1\n"
grzemich 0:d7bd7384a37c 55 " lsls r2, r2, #8\n"
grzemich 0:d7bd7384a37c 56 " subs r1, r1, #1\n"
grzemich 0:d7bd7384a37c 57
grzemich 0:d7bd7384a37c 58 // Need to 4-byte align? If not skip ahead.
grzemich 0:d7bd7384a37c 59 "1$:\n"
grzemich 0:d7bd7384a37c 60 " ands r4, r0, #3\n"
grzemich 0:d7bd7384a37c 61 " beq 2$\n"
grzemich 0:d7bd7384a37c 62 // Have more than 1 byte left to align? If not skip ahead to take care of
grzemich 0:d7bd7384a37c 63 // trailing byte.
grzemich 0:d7bd7384a37c 64 " cmp r1, #2\n"
grzemich 0:d7bd7384a37c 65 " blt 7$\n"
grzemich 0:d7bd7384a37c 66
grzemich 0:d7bd7384a37c 67 // 4-byte align.
grzemich 0:d7bd7384a37c 68 " ldrh r4, [r0], #2\n"
grzemich 0:d7bd7384a37c 69 " adds r2, r2, r4\n"
grzemich 0:d7bd7384a37c 70 " subs r1, r1, #2\n"
grzemich 0:d7bd7384a37c 71
grzemich 0:d7bd7384a37c 72 // Main summing loop which sums up data 2 words at a time.
grzemich 0:d7bd7384a37c 73 // Make sure that we have more than 7 bytes left to sum.
grzemich 0:d7bd7384a37c 74 "2$:\n"
grzemich 0:d7bd7384a37c 75 " cmp r1, #8\n"
grzemich 0:d7bd7384a37c 76 " blt 3$\n"
grzemich 0:d7bd7384a37c 77 // Sum next two words. Applying previous upper 16-bit carry to
grzemich 0:d7bd7384a37c 78 // lower 16-bits.
grzemich 0:d7bd7384a37c 79 " ldr r4, [r0], #4\n"
grzemich 0:d7bd7384a37c 80 " adds r2, r4\n"
grzemich 0:d7bd7384a37c 81 " adc r2, r2, #0\n"
grzemich 0:d7bd7384a37c 82 " ldr r4, [r0], #4\n"
grzemich 0:d7bd7384a37c 83 " adds r2, r4\n"
grzemich 0:d7bd7384a37c 84 " adc r2, r2, #0\n"
grzemich 0:d7bd7384a37c 85 " subs r1, r1, #8\n"
grzemich 0:d7bd7384a37c 86 " b 2$\n"
grzemich 0:d7bd7384a37c 87
grzemich 0:d7bd7384a37c 88 // Sum up any remaining half-words.
grzemich 0:d7bd7384a37c 89 "3$:\n"
grzemich 0:d7bd7384a37c 90 // Make sure that we have more than 1 byte left to sum.
grzemich 0:d7bd7384a37c 91 " cmp r1, #2\n"
grzemich 0:d7bd7384a37c 92 " blt 7$\n"
grzemich 0:d7bd7384a37c 93 // Sum up next half word, continue to apply carry.
grzemich 0:d7bd7384a37c 94 " ldrh r4, [r0], #2\n"
grzemich 0:d7bd7384a37c 95 " adds r2, r4\n"
grzemich 0:d7bd7384a37c 96 " adc r2, r2, #0\n"
grzemich 0:d7bd7384a37c 97 " subs r1, r1, #2\n"
grzemich 0:d7bd7384a37c 98 " b 3$\n"
grzemich 0:d7bd7384a37c 99
grzemich 0:d7bd7384a37c 100 // Handle trailing byte, if it exists
grzemich 0:d7bd7384a37c 101 "7$:\n"
grzemich 0:d7bd7384a37c 102 " cbz r1, 8$\n"
grzemich 0:d7bd7384a37c 103 " ldrb r4, [r0]\n"
grzemich 0:d7bd7384a37c 104 " adds r2, r4\n"
grzemich 0:d7bd7384a37c 105 " adc r2, r2, #0\n"
grzemich 0:d7bd7384a37c 106
grzemich 0:d7bd7384a37c 107 // Fold 32-bit checksum into 16-bit checksum.
grzemich 0:d7bd7384a37c 108 "8$:\n"
grzemich 0:d7bd7384a37c 109 " ubfx r4, r2, #16, #16\n"
grzemich 0:d7bd7384a37c 110 " ubfx r2, r2, #0, #16\n"
grzemich 0:d7bd7384a37c 111 " adds r2, r4\n"
grzemich 0:d7bd7384a37c 112 " ubfx r4, r2, #16, #16\n"
grzemich 0:d7bd7384a37c 113 " ubfx r2, r2, #0, #16\n"
grzemich 0:d7bd7384a37c 114 " adds r2, r4\n"
grzemich 0:d7bd7384a37c 115
grzemich 0:d7bd7384a37c 116 // Swap bytes if started at odd address
grzemich 0:d7bd7384a37c 117 " cbz r3, 9$\n"
grzemich 0:d7bd7384a37c 118 " rev16 r2, r2\n"
grzemich 0:d7bd7384a37c 119
grzemich 0:d7bd7384a37c 120 // Return final sum.
grzemich 0:d7bd7384a37c 121 "9$: mov r0, r2\n"
grzemich 0:d7bd7384a37c 122 " pop {r4, pc}\n"
grzemich 0:d7bd7384a37c 123 );
grzemich 0:d7bd7384a37c 124 }
grzemich 0:d7bd7384a37c 125
grzemich 0:d7bd7384a37c 126 #endif