memclr_arm64.s

     1  // Copyright 2014 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 "textflag.h"
     6  
     7  // See memclrNoHeapPointers Go doc for important implementation constraints.
     8  
     9  // func memclrNoHeapPointers(ptr unsafe.Pointer, n uintptr)
    10  // Also called from assembly in sys_windows_arm64.s without g (but using Go stack convention).
    11  TEXT runtime·memclrNoHeapPointers<ABIInternal>(SB),NOSPLIT,$0-16
    12  	CMP	$16, R1
    13  	// If n is equal to 16 bytes, use zero_exact_16 to zero
    14  	BEQ	zero_exact_16
    15  
    16  	// If n is greater than 16 bytes, use zero_by_16 to zero
    17  	BHI	zero_by_16
    18  
    19  	// n is less than 16 bytes
    20  	ADD	R1, R0, R7
    21  	TBZ	$3, R1, less_than_8
    22  	MOVD	ZR, (R0)
    23  	MOVD	ZR, -8(R7)
    24  	RET
    25  
    26  less_than_8:
    27  	TBZ	$2, R1, less_than_4
    28  	MOVW	ZR, (R0)
    29  	MOVW	ZR, -4(R7)
    30  	RET
    31  
    32  less_than_4:
    33  	CBZ	R1, ending
    34  	MOVB	ZR, (R0)
    35  	TBZ	$1, R1, ending
    36  	MOVH	ZR, -2(R7)
    37  
    38  ending:
    39  	RET
    40  
    41  zero_exact_16:
    42  	// n is exactly 16 bytes
    43  	STP	(ZR, ZR), (R0)
    44  	RET
    45  
    46  zero_by_16:
    47  	// n greater than 16 bytes, check if the start address is aligned
    48  	NEG	R0, R4
    49  	ANDS	$15, R4, R4
    50  	// Try zeroing using zva if the start address is aligned with 16
    51  	BEQ	try_zva
    52  
    53  	// Non-aligned store
    54  	STP	(ZR, ZR), (R0)
    55  	// Make the destination aligned
    56  	SUB	R4, R1, R1
    57  	ADD	R4, R0, R0
    58  	B	try_zva
    59  
    60  tail_maybe_long:
    61  	CMP	$64, R1
    62  	BHS	no_zva
    63  
    64  tail63:
    65  	ANDS	$48, R1, R3
    66  	BEQ	last16
    67  	CMPW	$32, R3
    68  	BEQ	last48
    69  	BLT	last32
    70  	STP.P	(ZR, ZR), 16(R0)
    71  last48:
    72  	STP.P	(ZR, ZR), 16(R0)
    73  last32:
    74  	STP.P	(ZR, ZR), 16(R0)
    75  	// The last store length is at most 16, so it is safe to use
    76  	// stp to write last 16 bytes
    77  last16:
    78  	ANDS	$15, R1, R1
    79  	CBZ	R1, last_end
    80  	ADD	R1, R0, R0
    81  	STP	(ZR, ZR), -16(R0)
    82  last_end:
    83  	RET
    84  
    85  no_zva:
    86  	SUB	$16, R0, R0
    87  	SUB	$64, R1, R1
    88  
    89  loop_64:
    90  	STP	(ZR, ZR), 16(R0)
    91  	STP	(ZR, ZR), 32(R0)
    92  	STP	(ZR, ZR), 48(R0)
    93  	STP.W	(ZR, ZR), 64(R0)
    94  	SUBS	$64, R1, R1
    95  	BGE	loop_64
    96  	ANDS	$63, R1, ZR
    97  	ADD	$16, R0, R0
    98  	BNE	tail63
    99  	RET
   100  
   101  try_zva:
   102  	// Try using the ZVA feature to zero entire cache lines
   103  	// It is not meaningful to use ZVA if the block size is less than 64,
   104  	// so make sure that n is greater than or equal to 64
   105  	CMP	$63, R1
   106  	BLE	tail63
   107  
   108  	CMP	$128, R1
   109  	// Ensure n is at least 128 bytes, so that there is enough to copy after
   110  	// alignment.
   111  	BLT	no_zva
   112  	// Check if ZVA is allowed from user code, and if so get the block size
   113  	MOVW	block_size<>(SB), R5
   114  	TBNZ	$31, R5, no_zva
   115  	CBNZ	R5, zero_by_line
   116  	// DCZID_EL0 bit assignments
   117  	// [63:5] Reserved
   118  	// [4]    DZP, if bit set DC ZVA instruction is prohibited, else permitted
   119  	// [3:0]  log2 of the block size in words, eg. if it returns 0x4 then block size is 16 words
   120  	MRS	DCZID_EL0, R3
   121  	TBZ	$4, R3, init
   122  	// ZVA not available
   123  	MOVW	$~0, R5
   124  	MOVW	R5, block_size<>(SB)
   125  	B	no_zva
   126  
   127  init:
   128  	MOVW	$4, R9
   129  	ANDW	$15, R3, R5
   130  	LSLW	R5, R9, R5
   131  	MOVW	R5, block_size<>(SB)
   132  
   133  	ANDS	$63, R5, R9
   134  	// Block size is less than 64.
   135  	BNE	no_zva
   136  
   137  zero_by_line:
   138  	CMP	R5, R1
   139  	// Not enough memory to reach alignment
   140  	BLO	no_zva
   141  	SUB	$1, R5, R6
   142  	NEG	R0, R4
   143  	ANDS	R6, R4, R4
   144  	// Already aligned
   145  	BEQ	aligned
   146  
   147  	// check there is enough to copy after alignment
   148  	SUB	R4, R1, R3
   149  
   150  	// Check that the remaining length to ZVA after alignment
   151  	// is greater than 64.
   152  	CMP	$64, R3
   153  	CCMP	GE, R3, R5, $10  // condition code GE, NZCV=0b1010
   154  	BLT	no_zva
   155  
   156  	// We now have at least 64 bytes to zero, update n
   157  	MOVD	R3, R1
   158  
   159  loop_zva_prolog:
   160  	STP	(ZR, ZR), (R0)
   161  	STP	(ZR, ZR), 16(R0)
   162  	STP	(ZR, ZR), 32(R0)
   163  	SUBS	$64, R4, R4
   164  	STP	(ZR, ZR), 48(R0)
   165  	ADD	$64, R0, R0
   166  	BGE	loop_zva_prolog
   167  
   168  	ADD	R4, R0, R0
   169  
   170  aligned:
   171  	SUB	R5, R1, R1
   172  
   173  loop_zva:
   174  	WORD	$0xd50b7420 // DC ZVA, R0
   175  	ADD	R5, R0, R0
   176  	SUBS	R5, R1, R1
   177  	BHS	loop_zva
   178  	ANDS	R6, R1, R1
   179  	BNE	tail_maybe_long
   180  	RET
   181  
   182  GLOBL block_size<>(SB), NOPTR, $8
   183
View as plain text