Text file src/internal/bytealg/compare_loong64.s

     1  // Copyright 2022 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<ABIInternal>(SB),NOSPLIT,$0-56
     9  	// R4 = a_base
    10  	// R5 = a_len
    11  	// R6 = a_cap (unused)
    12  	// R7 = b_base (want in R6)
    13  	// R8 = b_len (want in R7)
    14  	// R9 = b_cap (unused)
    15  	MOVV	R7, R6
    16  	MOVV	R8, R7
    17  	JMP	cmpbody<>(SB)
    18  
    19  TEXT runtime·cmpstring<ABIInternal>(SB),NOSPLIT,$0-40
    20  	// R4 = a_base
    21  	// R5 = a_len
    22  	// R6 = b_base
    23  	// R7 = b_len
    24  	JMP	cmpbody<>(SB)
    25  
    26  // On entry:
    27  // R5 length of a
    28  // R7 length of b
    29  // R4 points to the start of a
    30  // R6 points to the start of b
    31  // R13 points to the return value (-1/0/1)
    32  TEXT cmpbody<>(SB),NOSPLIT|NOFRAME,$0
    33  	BEQ	R4, R6, samebytes // same start of a and b
    34  
    35  	SGTU	R5, R7, R9
    36  	BNE	R0, R9, r2_lt_r1
    37  	MOVV	R5, R14
    38  	JMP	entry
    39  r2_lt_r1:
    40  	MOVV	R7, R14	// R14 is min(R4, R5)
    41  entry:
    42  	ADDV	R4, R14, R12	// R6 start of a, R14 end of a
    43  	BEQ	R4, R12, samebytes // length is 0
    44  
    45  	SRLV	$4, R14		// R14 is number of chunks
    46  	BEQ	R0, R14, byte_loop
    47  
    48  	// make sure both a and b are aligned.
    49  	OR	R4, R6, R15
    50  	AND	$7, R15
    51  	BNE	R0, R15, byte_loop
    52  
    53  	PCALIGN	$16
    54  chunk16_loop:
    55  	BEQ	R0, R14, byte_loop
    56  	MOVV	(R4), R8
    57  	MOVV	(R6), R9
    58  	BNE	R8, R9, byte_loop
    59  	MOVV	8(R4), R16
    60  	MOVV	8(R6), R17
    61  	ADDV	$16, R4
    62  	ADDV	$16, R6
    63  	SUBVU	$1, R14
    64  	BEQ	R16, R17, chunk16_loop
    65  	SUBV	$8, R4
    66  	SUBV	$8, R6
    67  
    68  byte_loop:
    69  	BEQ	R4, R12, samebytes
    70  	MOVBU	(R4), R8
    71  	ADDVU	$1, R4
    72  	MOVBU	(R6), R9
    73  	ADDVU	$1, R6
    74  	BEQ	R8, R9, byte_loop
    75  
    76  byte_cmp:
    77  	SGTU	R8, R9, R4 // R12 = 1 if (R8 > R9)
    78  	BNE	R0, R4, ret
    79  	MOVV	$-1, R4
    80  	JMP	ret
    81  
    82  samebytes:
    83  	SGTU	R5, R7, R8
    84  	SGTU	R7, R5, R9
    85  	SUBV	R9, R8, R4
    86  
    87  ret:
    88  	RET
    89  

View as plain text