| Mikael Starvik | 51533b6 | 2005-07-27 11:44:44 -0700 | [diff] [blame^] | 1 | /* |
| 2 | * A fast checksum+copy routine using movem |
| 3 | * Copyright (c) 1998, 2001, 2003 Axis Communications AB |
| 4 | * |
| 5 | * Authors: Bjorn Wesen |
| 6 | * |
| 7 | * csum_partial_copy_nocheck(const char *src, char *dst, |
| 8 | * int len, unsigned int sum) |
| 9 | */ |
| 10 | |
| 11 | .globl csum_partial_copy_nocheck |
| 12 | csum_partial_copy_nocheck: |
| 13 | |
| 14 | ;; r10 - src |
| 15 | ;; r11 - dst |
| 16 | ;; r12 - length |
| 17 | ;; r13 - checksum |
| 18 | |
| 19 | ;; check for breakeven length between movem and normal word looping versions |
| 20 | ;; we also do _NOT_ want to compute a checksum over more than the |
| 21 | ;; actual length when length < 40 |
| 22 | |
| 23 | cmpu.w 80,$r12 |
| 24 | blo _word_loop |
| 25 | nop |
| 26 | |
| 27 | ;; need to save the registers we use below in the movem loop |
| 28 | ;; this overhead is why we have a check above for breakeven length |
| 29 | ;; only r0 - r8 have to be saved, the other ones are clobber-able |
| 30 | ;; according to the ABI |
| 31 | |
| 32 | subq 9*4,$sp |
| 33 | subq 10*4,$r12 ; update length for the first loop |
| 34 | movem $r8,[$sp] |
| 35 | |
| 36 | ;; do a movem copy and checksum |
| 37 | |
| 38 | 1: ;; A failing userspace access (the read) will have this as PC. |
| 39 | _mloop: movem [$r10+],$r9 ; read 10 longwords |
| 40 | movem $r9,[$r11+] ; write 10 longwords |
| 41 | |
| 42 | ;; perform dword checksumming on the 10 longwords |
| 43 | |
| 44 | add.d $r0,$r13 |
| 45 | addc $r1,$r13 |
| 46 | addc $r2,$r13 |
| 47 | addc $r3,$r13 |
| 48 | addc $r4,$r13 |
| 49 | addc $r5,$r13 |
| 50 | addc $r6,$r13 |
| 51 | addc $r7,$r13 |
| 52 | addc $r8,$r13 |
| 53 | addc $r9,$r13 |
| 54 | |
| 55 | ;; fold the carry into the checksum, to avoid having to loop the carry |
| 56 | ;; back into the top |
| 57 | |
| 58 | addc 0,$r13 |
| 59 | addc 0,$r13 ; do it again, since we might have generated a carry |
| 60 | |
| 61 | subq 10*4,$r12 |
| 62 | bge _mloop |
| 63 | nop |
| 64 | |
| 65 | addq 10*4,$r12 ; compensate for last loop underflowing length |
| 66 | |
| 67 | movem [$sp+],$r8 ; restore regs |
| 68 | |
| 69 | _word_loop: |
| 70 | ;; only fold if there is anything to fold. |
| 71 | |
| 72 | cmpq 0,$r13 |
| 73 | beq _no_fold |
| 74 | |
| 75 | ;; fold 32-bit checksum into a 16-bit checksum, to avoid carries below |
| 76 | ;; r9 can be used as temporary. |
| 77 | |
| 78 | move.d $r13,$r9 |
| 79 | lsrq 16,$r9 ; r0 = checksum >> 16 |
| 80 | and.d 0xffff,$r13 ; checksum = checksum & 0xffff |
| 81 | add.d $r9,$r13 ; checksum += r0 |
| 82 | move.d $r13,$r9 ; do the same again, maybe we got a carry last add |
| 83 | lsrq 16,$r9 |
| 84 | and.d 0xffff,$r13 |
| 85 | add.d $r9,$r13 |
| 86 | |
| 87 | _no_fold: |
| 88 | cmpq 2,$r12 |
| 89 | blt _no_words |
| 90 | nop |
| 91 | |
| 92 | ;; copy and checksum the rest of the words |
| 93 | |
| 94 | subq 2,$r12 |
| 95 | |
| 96 | 2: ;; A failing userspace access for the read below will have this as PC. |
| 97 | _wloop: move.w [$r10+],$r9 |
| 98 | addu.w $r9,$r13 |
| 99 | subq 2,$r12 |
| 100 | bge _wloop |
| 101 | move.w $r9,[$r11+] |
| 102 | |
| 103 | addq 2,$r12 |
| 104 | |
| 105 | _no_words: |
| 106 | ;; see if we have one odd byte more |
| 107 | cmpq 1,$r12 |
| 108 | beq _do_byte |
| 109 | nop |
| 110 | ret |
| 111 | move.d $r13,$r10 |
| 112 | |
| 113 | _do_byte: |
| 114 | ;; copy and checksum the last byte |
| 115 | 3: ;; A failing userspace access for the read below will have this as PC. |
| 116 | move.b [$r10],$r9 |
| 117 | addu.b $r9,$r13 |
| 118 | move.b $r9,[$r11] |
| 119 | ret |
| 120 | move.d $r13,$r10 |