github.com/sandwichdev/go-internals@v0.0.0-20210605002614-12311ac6b2c5/bytealg/compare_arm.s (about) 1 // Copyright 2018 The Go Authors. All rights reserved. 2 // Use of this source code is governed by a BSD-style 3 // license that can be found in the LICENSE file. 4 5 #include "go_asm.h" 6 #include "textflag.h" 7 8 TEXT ·Compare(SB),NOSPLIT|NOFRAME,$0-28 9 MOVW a_base+0(FP), R2 10 MOVW a_len+4(FP), R0 11 MOVW b_base+12(FP), R3 12 MOVW b_len+16(FP), R1 13 ADD $28, R13, R7 14 B cmpbody<>(SB) 15 16 TEXT runtime·cmpstring(SB),NOSPLIT|NOFRAME,$0-20 17 MOVW a_base+0(FP), R2 18 MOVW a_len+4(FP), R0 19 MOVW b_base+8(FP), R3 20 MOVW b_len+12(FP), R1 21 ADD $20, R13, R7 22 B cmpbody<>(SB) 23 24 // On entry: 25 // R0 is the length of a 26 // R1 is the length of b 27 // R2 points to the start of a 28 // R3 points to the start of b 29 // R7 points to return value (-1/0/1 will be written here) 30 // 31 // On exit: 32 // R4, R5, R6 and R8 are clobbered 33 TEXT cmpbody<>(SB),NOSPLIT|NOFRAME,$0-0 34 CMP R2, R3 35 BEQ samebytes 36 CMP R0, R1 37 MOVW R0, R6 38 MOVW.LT R1, R6 // R6 is min(R0, R1) 39 40 CMP $0, R6 41 BEQ samebytes 42 CMP $4, R6 43 ADD R2, R6 // R2 is current byte in a, R6 is the end of the range to compare 44 BLT byte_loop // length < 4 45 AND $3, R2, R8 46 CMP $0, R8 47 BNE byte_loop // unaligned a, use byte-wise compare (TODO: try to align a) 48 aligned_a: 49 AND $3, R3, R8 50 CMP $0, R8 51 BNE byte_loop // unaligned b, use byte-wise compare 52 AND $0xfffffffc, R6, R8 53 // length >= 4 54 chunk4_loop: 55 MOVW.P 4(R2), R4 56 MOVW.P 4(R3), R5 57 CMP R4, R5 58 BNE cmp 59 CMP R2, R8 60 BNE chunk4_loop 61 CMP R2, R6 62 BEQ samebytes // all compared bytes were the same; compare lengths 63 byte_loop: 64 MOVBU.P 1(R2), R4 65 MOVBU.P 1(R3), R5 66 CMP R4, R5 67 BNE ret 68 CMP R2, R6 69 BNE byte_loop 70 samebytes: 71 CMP R0, R1 72 MOVW.LT $1, R0 73 MOVW.GT $-1, R0 74 MOVW.EQ $0, R0 75 MOVW R0, (R7) 76 RET 77 ret: 78 // bytes differed 79 MOVW.LT $1, R0 80 MOVW.GT $-1, R0 81 MOVW R0, (R7) 82 RET 83 cmp: 84 SUB $4, R2, R2 85 SUB $4, R3, R3 86 B byte_loop