github.com/icodeface/tls@v0.0.0-20230910023335-34df9250cd12/internal/bytealg/count_arm64.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 ·Count(SB),NOSPLIT,$0-40 9 MOVD b_base+0(FP), R0 10 MOVD b_len+8(FP), R2 11 MOVBU c+24(FP), R1 12 MOVD $ret+32(FP), R8 13 B countbytebody<>(SB) 14 15 TEXT ·CountString(SB),NOSPLIT,$0-32 16 MOVD s_base+0(FP), R0 17 MOVD s_len+8(FP), R2 18 MOVBU c+16(FP), R1 19 MOVD $ret+24(FP), R8 20 B countbytebody<>(SB) 21 22 // input: 23 // R0: data 24 // R2: data len 25 // R1: byte to find 26 // R8: address to put result 27 TEXT countbytebody<>(SB),NOSPLIT,$0 28 // R11 = count of byte to search 29 MOVD $0, R11 30 // short path to handle 0-byte case 31 CBZ R2, done 32 CMP $0x20, R2 33 // jump directly to tail if length < 32 34 BLO tail 35 ANDS $0x1f, R0, R9 36 BEQ chunk 37 // Work with not 32-byte aligned head 38 BIC $0x1f, R0, R3 39 ADD $0x20, R3 40 head_loop: 41 MOVBU.P 1(R0), R5 42 CMP R5, R1 43 CINC EQ, R11, R11 44 SUB $1, R2, R2 45 CMP R0, R3 46 BNE head_loop 47 // Work with 32-byte aligned chunks 48 chunk: 49 BIC $0x1f, R2, R9 50 // The first chunk can also be the last 51 CBZ R9, tail 52 // R3 = end of 32-byte chunks 53 ADD R0, R9, R3 54 MOVD $1, R5 55 VMOV R5, V5.B16 56 // R2 = length of tail 57 SUB R9, R2, R2 58 // Duplicate R1 (byte to search) to 16 1-byte elements of V0 59 VMOV R1, V0.B16 60 // Clear the low 64-bit element of V7 and V8 61 VEOR V7.B8, V7.B8, V7.B8 62 VEOR V8.B8, V8.B8, V8.B8 63 // Count the target byte in 32-byte chunk 64 chunk_loop: 65 VLD1.P (R0), [V1.B16, V2.B16] 66 CMP R0, R3 67 VCMEQ V0.B16, V1.B16, V3.B16 68 VCMEQ V0.B16, V2.B16, V4.B16 69 // Clear the higher 7 bits 70 VAND V5.B16, V3.B16, V3.B16 71 VAND V5.B16, V4.B16, V4.B16 72 // Count lanes match the requested byte 73 VADDP V4.B16, V3.B16, V6.B16 // 32B->16B 74 VUADDLV V6.B16, V7 75 // Accumulate the count in low 64-bit element of V8 when inside the loop 76 VADD V7, V8 77 BNE chunk_loop 78 VMOV V8.D[0], R6 79 ADD R6, R11, R11 80 CBZ R2, done 81 tail: 82 // Work with tail shorter than 32 bytes 83 MOVBU.P 1(R0), R5 84 SUB $1, R2, R2 85 CMP R5, R1 86 CINC EQ, R11, R11 87 CBNZ R2, tail 88 done: 89 MOVD R11, (R8) 90 RET