// MOVQ and EMMS were introduced on the Pentium MMX.
	MOVQ	val+4(FP), M0
	MOVQ	M0, (AX)
	EMMS
	// This is essentially a no-op, but it provides required memory fencing.
	// It can be replaced with MFENCE, but MFENCE was introduced only on the Pentium4 (SSE2).
	XORL	AX, AX
	LOCK
	XADDL	AX, (SP)
	RET

// void	·Or8(byte volatile*, byte);
TEXT ·Or8(SB), NOSPLIT, $0-5
	MOVL	ptr+0(FP), AX
	MOVB	val+4(FP), BX

done:
	MOVL	AX, ret_lo+0(FP)
	MOVL	DX, ret_hi+4(FP)
	RET
fences:
	// MFENCE is instruction stream serializing and flushes the
	// store buffers on AMD. The serialization semantics of LFENCE on AMD
	// are dependent on MSR C001_1029 and CPU generation.
	// LFENCE on Intel does wait for all previous instructions to have executed.
	// Intel recommends MFENCE;LFENCE in its manuals before RDTSC to have all

//
// These four modes correspond to other well-known memory models on other CPUs.
// On ARM, aq corresponds to a dmb ishst, aq+rl corresponds to a dmb ish. On
// Intel, aq corresponds to an lfence, rl to an sfence, and aq+rl to an mfence
// (or a lock prefix).
//
// Go's memory model requires that
//   - if a read happens after a write, the read must observe the write, and
//     that

	SHLQ	$32, DX
	ADDQ	DX, AX
	MOVQ	AX, ret+0(FP)
	RET
fences:
	// MFENCE is instruction stream serializing and flushes the
	// store buffers on AMD. The serialization semantics of LFENCE on AMD
	// are dependent on MSR C001_1029 and CPU generation.
	// LFENCE on Intel does wait for all previous instructions to have executed.
	// Intel recommends MFENCE;LFENCE in its manuals before RDTSC to have all