1/* 2 * A fast checksum+copy routine using movem 3 * Copyright (c) 1998-2007 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 12csum_partial_copy_nocheck: 13 14 ;; r10 - src 15 ;; r11 - dst 16 ;; r12 - length 17 ;; r13 - checksum 18 19 ;; Optimized for large packets 20 subq 10*4, $r12 21 blt _word_loop 22 move.d $r12, $acr 23 24 subq 9*4,$sp 25 clearf c 26 movem $r8,[$sp] 27 28 ;; do a movem copy and checksum 291: ;; A failing userspace access (the read) will have this as PC. 30_mloop: movem [$r10+],$r9 ; read 10 longwords 31 addoq -10*4, $acr, $acr ; loop counter in latency cycle 32 movem $r9,[$r11+] ; write 10 longwords 33 34 ;; perform dword checksumming on the 10 longwords 35 addc $r0,$r13 36 addc $r1,$r13 37 addc $r2,$r13 38 addc $r3,$r13 39 addc $r4,$r13 40 addc $r5,$r13 41 addc $r6,$r13 42 addc $r7,$r13 43 addc $r8,$r13 44 addc $r9,$r13 45 46 ;; test $acr, without trashing carry. 47 move.d $acr, $acr 48 bpl _mloop 49 ;; r12 <= acr is needed after mloop and in the exception handlers. 50 move.d $acr, $r12 51 52 ;; fold the last carry into r13 53 addc 0, $r13 54 movem [$sp+],$r8 ; restore regs 55 56_word_loop: 57 addq 10*4,$r12 ; compensate for last loop underflowing length 58 59 ;; fold 32-bit checksum into a 16-bit checksum, to avoid carries below 60 ;; r9 can be used as temporary. 61 move.d $r13,$r9 62 lsrq 16,$r9 ; r0 = checksum >> 16 63 and.d 0xffff,$r13 ; checksum = checksum & 0xffff 64 65 subq 2, $r12 66 blt _no_words 67 add.d $r9,$r13 ; checksum += r0 68 69 ;; copy and checksum the rest of the words 702: ;; A failing userspace access for the read below will have this as PC. 71_wloop: move.w [$r10+],$r9 72 addu.w $r9,$r13 73 subq 2,$r12 74 bge _wloop 75 move.w $r9,[$r11+] 76 77_no_words: 78 addq 2,$r12 79 bne _do_byte 80 nop 81 ret 82 move.d $r13,$r10 83 84_do_byte: 85 ;; copy and checksum the last byte 863: ;; A failing userspace access for the read below will have this as PC. 87 move.b [$r10],$r9 88 addu.b $r9,$r13 89 move.b $r9,[$r11] 90 ret 91 move.d $r13,$r10 92