ref: 4f1090a10d585fabda5c0bed6881c8d680b6955b
dir: /atomic-386.s/
#define CMPXCHG /* (CX) */\
BYTE $0x0F; BYTE $0xB1; BYTE $0x11
#define CMPXCHG64 /* (DI) */\
BYTE $0x0F; BYTE $0xC7; BYTE $0x0F
#define XADDL /* CX, (AX) */ \
BYTE $0x67; BYTE $0x0F; BYTE $0xC1; BYTE $0x08
#define MFENCE
BYTE $0x0F; BYTE $0xAE; BYTE $0xF0
/* get variants */
TEXT ageti+0(SB),1,$0
TEXT agetl+0(SB),1,$0
TEXT agetp+0(SB),1,$0
MOVL p+0(FP), AX
MOVL 0(AX), AX
RET
TEXT agetv+0(SB),1,$0
MOVL p+0(FP),AX
MOVL p+4(FP),BX
FMOVD (BX),F0
FMOVD F0,(AX)
RET
/* set variants */
TEXT aseti+0(SB),1,$0
TEXT asetl+0(SB),1,$0
TEXT asetp+0(SB),1,$0
MOVL v+0(FP), AX
MOVL 0(AX), AX
MOVL v+4(FP), BX
LOCK; XCHGL (AX), BX
RET
TEXT asetv+0(SB),1,$0
MOVL p+4(FP), DI
MOVL nv+8(FP), BX
MOVL nv+12(FP), CX
MOVL 0(DI), AX
MOVL 4(DI), DX
loop:
LOCK; CMPXCHG64
JNE loop
MOVL p+0(FP),DI
MOVL AX, 0(DI)
MOVL DX, 4(DI)
RET
/* inc variants */
TEXT ainci+0(SB),1,$0
TEXT aincl+0(SB),1,$0
TEXT aincp+0(SB),1,$0
MOVL p+0(FP), AX
MOVL 0(AX), AX
MOVL v+4(FP), CX
LOCK; XADDL // AX, (DX)
RET
TEXT aincv+0(SB),1,$0
MOVL p+4(FP), DI
retry:
MOVL 0x0(DI), AX // ov = *p
MOVL 0x4(DI), DX
MOVL AX, BX // nv = ov
MOVL DX, CX
ADDL dv+0x8(FP), BX // nv += dv
ADCL dv+0x12(FP), CX
LOCK; CMPXCHG64
JNE retry
MOVL r+0(FP), DI
MOVL BX, (DI)
MOVL CX, (DI)
RET
/* cas variants */
TEXT acasi+0(SB),1,$0
TEXT acasl+0(SB),1,$0
TEXT acasp+0(SB),1,$0
MOVL p+0(FP), CX
MOVL ov+4(FP), AX
MOVL nv+8(FP), DX
LOCK; CMPXCHG // (CX)
JNE fail32
MOVL $1,AX
RET
fail32:
MOVL $0,AX
RET
TEXT acasv+0(SB),1,$0
MOVL p+0(FP), DI
MOVL ov+0x4(FP), AX
MOVL ov+0x8(FP), DX
MOVL nv+0xc(FP), BX
MOVL nv+0x10(FP), CX
LOCK; CMPXCHG64 // (DI)
JNE fail64
MOVL $1,AX
fail64:
MOVL $0,AX
RET
/* barriers (do we want to distinguish types?) */
TEXT coherence+0(SB),1,$0
MFENCE
RET