github.com/slayercat/go@v0.0.0-20170428012452-c51559813f61/src/sync/atomic/asm_386.s (about)

     1  // Copyright 2011 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  // +build !race
     6  
     7  #include "textflag.h"
     8  
     9  TEXT ·SwapInt32(SB),NOSPLIT,$0-12
    10  	JMP	·SwapUint32(SB)
    11  
    12  TEXT ·SwapUint32(SB),NOSPLIT,$0-12
    13  	MOVL	addr+0(FP), BP
    14  	MOVL	new+4(FP), AX
    15  	XCHGL	AX, 0(BP)
    16  	MOVL	AX, old+8(FP)
    17  	RET
    18  
    19  TEXT ·SwapInt64(SB),NOSPLIT,$0-20
    20  	JMP	·SwapUint64(SB)
    21  
    22  TEXT ·SwapUint64(SB),NOSPLIT,$0-20
    23  	// no XCHGQ so use CMPXCHG8B loop
    24  	MOVL	addr+0(FP), BP
    25  	TESTL	$7, BP
    26  	JZ	2(PC)
    27  	MOVL	0, AX // crash with nil ptr deref
    28  	// CX:BX = new
    29  	MOVL	new_lo+4(FP), BX
    30  	MOVL	new_hi+8(FP), CX
    31  	// DX:AX = *addr
    32  	MOVL	0(BP), AX
    33  	MOVL	4(BP), DX
    34  swaploop:
    35  	// if *addr == DX:AX
    36  	//	*addr = CX:BX
    37  	// else
    38  	//	DX:AX = *addr
    39  	// all in one instruction
    40  	LOCK
    41  	CMPXCHG8B	0(BP)
    42  	JNZ	swaploop
    43  
    44  	// success
    45  	// return DX:AX
    46  	MOVL	AX, old_lo+12(FP)
    47  	MOVL	DX, old_hi+16(FP)
    48  	RET
    49  
    50  TEXT ·SwapUintptr(SB),NOSPLIT,$0-12
    51  	JMP	·SwapUint32(SB)
    52  
    53  TEXT ·CompareAndSwapInt32(SB),NOSPLIT,$0-13
    54  	JMP	·CompareAndSwapUint32(SB)
    55  
    56  TEXT ·CompareAndSwapUint32(SB),NOSPLIT,$0-13
    57  	MOVL	addr+0(FP), BP
    58  	MOVL	old+4(FP), AX
    59  	MOVL	new+8(FP), CX
    60  	// CMPXCHGL was introduced on the 486.
    61  	LOCK
    62  	CMPXCHGL	CX, 0(BP)
    63  	SETEQ	swapped+12(FP)
    64  	RET
    65  
    66  TEXT ·CompareAndSwapUintptr(SB),NOSPLIT,$0-13
    67  	JMP	·CompareAndSwapUint32(SB)
    68  
    69  TEXT ·CompareAndSwapInt64(SB),NOSPLIT,$0-21
    70  	JMP	·CompareAndSwapUint64(SB)
    71  
    72  TEXT ·CompareAndSwapUint64(SB),NOSPLIT,$0-21
    73  	MOVL	addr+0(FP), BP
    74  	TESTL	$7, BP
    75  	JZ	2(PC)
    76  	MOVL	0, AX // crash with nil ptr deref
    77  	MOVL	old_lo+4(FP), AX
    78  	MOVL	old_hi+8(FP), DX
    79  	MOVL	new_lo+12(FP), BX
    80  	MOVL	new_hi+16(FP), CX
    81  	// CMPXCHG8B was introduced on the Pentium.
    82  	LOCK
    83  	CMPXCHG8B	0(BP)
    84  	SETEQ	swapped+20(FP)
    85  	RET
    86  
    87  TEXT ·AddInt32(SB),NOSPLIT,$0-12
    88  	JMP	·AddUint32(SB)
    89  
    90  TEXT ·AddUint32(SB),NOSPLIT,$0-12
    91  	MOVL	addr+0(FP), BP
    92  	MOVL	delta+4(FP), AX
    93  	MOVL	AX, CX
    94  	// XADD was introduced on the 486.
    95  	LOCK
    96  	XADDL	AX, 0(BP)
    97  	ADDL	AX, CX
    98  	MOVL	CX, new+8(FP)
    99  	RET
   100  
   101  TEXT ·AddUintptr(SB),NOSPLIT,$0-12
   102  	JMP	·AddUint32(SB)
   103  
   104  TEXT ·AddInt64(SB),NOSPLIT,$0-20
   105  	JMP	·AddUint64(SB)
   106  
   107  TEXT ·AddUint64(SB),NOSPLIT,$0-20
   108  	// no XADDQ so use CMPXCHG8B loop
   109  	MOVL	addr+0(FP), BP
   110  	TESTL	$7, BP
   111  	JZ	2(PC)
   112  	MOVL	0, AX // crash with nil ptr deref
   113  	// DI:SI = delta
   114  	MOVL	delta_lo+4(FP), SI
   115  	MOVL	delta_hi+8(FP), DI
   116  	// DX:AX = *addr
   117  	MOVL	0(BP), AX
   118  	MOVL	4(BP), DX
   119  addloop:
   120  	// CX:BX = DX:AX (*addr) + DI:SI (delta)
   121  	MOVL	AX, BX
   122  	MOVL	DX, CX
   123  	ADDL	SI, BX
   124  	ADCL	DI, CX
   125  
   126  	// if *addr == DX:AX {
   127  	//	*addr = CX:BX
   128  	// } else {
   129  	//	DX:AX = *addr
   130  	// }
   131  	// all in one instruction
   132  	LOCK
   133  	CMPXCHG8B	0(BP)
   134  
   135  	JNZ	addloop
   136  
   137  	// success
   138  	// return CX:BX
   139  	MOVL	BX, new_lo+12(FP)
   140  	MOVL	CX, new_hi+16(FP)
   141  	RET
   142  
   143  TEXT ·LoadInt32(SB),NOSPLIT,$0-8
   144  	JMP	·LoadUint32(SB)
   145  
   146  TEXT ·LoadUint32(SB),NOSPLIT,$0-8
   147  	MOVL	addr+0(FP), AX
   148  	MOVL	0(AX), AX
   149  	MOVL	AX, val+4(FP)
   150  	RET
   151  
   152  TEXT ·LoadInt64(SB),NOSPLIT,$0-12
   153  	JMP	·LoadUint64(SB)
   154  
   155  TEXT ·LoadUint64(SB),NOSPLIT,$0-12
   156  	MOVL	addr+0(FP), AX
   157  	TESTL	$7, AX
   158  	JZ	2(PC)
   159  	MOVL	0, AX // crash with nil ptr deref
   160  	// MOVQ and EMMS were introduced on the Pentium MMX.
   161  	// MOVQ (%EAX), %MM0
   162  	BYTE $0x0f; BYTE $0x6f; BYTE $0x00
   163  	// MOVQ %MM0, 0x8(%ESP)
   164  	BYTE $0x0f; BYTE $0x7f; BYTE $0x44; BYTE $0x24; BYTE $0x08
   165  	EMMS
   166  	RET
   167  
   168  TEXT ·LoadUintptr(SB),NOSPLIT,$0-8
   169  	JMP	·LoadUint32(SB)
   170  
   171  TEXT ·LoadPointer(SB),NOSPLIT,$0-8
   172  	JMP	·LoadUint32(SB)
   173  
   174  TEXT ·StoreInt32(SB),NOSPLIT,$0-8
   175  	JMP	·StoreUint32(SB)
   176  
   177  TEXT ·StoreUint32(SB),NOSPLIT,$0-8
   178  	MOVL	addr+0(FP), BP
   179  	MOVL	val+4(FP), AX
   180  	XCHGL	AX, 0(BP)
   181  	RET
   182  
   183  TEXT ·StoreInt64(SB),NOSPLIT,$0-12
   184  	JMP	·StoreUint64(SB)
   185  
   186  TEXT ·StoreUint64(SB),NOSPLIT,$0-12
   187  	MOVL	addr+0(FP), AX
   188  	TESTL	$7, AX
   189  	JZ	2(PC)
   190  	MOVL	0, AX // crash with nil ptr deref
   191  	// MOVQ and EMMS were introduced on the Pentium MMX.
   192  	// MOVQ 0x8(%ESP), %MM0
   193  	BYTE $0x0f; BYTE $0x6f; BYTE $0x44; BYTE $0x24; BYTE $0x08
   194  	// MOVQ %MM0, (%EAX)
   195  	BYTE $0x0f; BYTE $0x7f; BYTE $0x00 
   196  	EMMS
   197  	// This is essentially a no-op, but it provides required memory fencing.
   198  	// It can be replaced with MFENCE, but MFENCE was introduced only on the Pentium4 (SSE2).
   199  	XORL	AX, AX
   200  	LOCK
   201  	XADDL	AX, (SP)
   202  	RET
   203  
   204  TEXT ·StoreUintptr(SB),NOSPLIT,$0-8
   205  	JMP	·StoreUint32(SB)