VMOVDQU8 15(DX)(BX*1), K3, X31                     // 62617f0b6fbc1a0f000000
	VMOVDQU8 -7(CX)(DX*2), K3, X31                     // 62617f0b6fbc51f9ffffff
	VMOVDQU8 Y3, K4, Y6                                // 62f17f2c7fde
	VMOVDQU8 Y3, K4, 7(SI)(DI*1)                       // 62f17f2c7f9c3e07000000
	VMOVDQU8 Y3, K4, 15(DX)(BX*8)                      // 62f17f2c7f9cda0f000000

func (c *sigctxt) setsp(x uintptr) { c.u.sp = uint64(x) }
func (c *sigctxt) setlr(x uintptr) {}

func (c *sigctxt) savelr(x uintptr) {}

func dumpregs(u *ureg) {
	print("ax    ", hex(u.ax), "\n")
	print("bx    ", hex(u.bx), "\n")
	print("cx    ", hex(u.cx), "\n")
	print("dx    ", hex(u.dx), "\n")
	print("di    ", hex(u.di), "\n")
	print("si    ", hex(u.si), "\n")
	print("bp    ", hex(u.bp), "\n")
	print("sp    ", hex(u.sp), "\n")

// R11 = caller's return address
TEXT	racefuncenter<>(SB), NOSPLIT|NOFRAME, $0-0
	MOVQ	DX, BX		// save function entry context (for closures)
	MOVQ	g_racectx(R14), RARG0	// goroutine context
	MOVQ	R11, RARG1
	// void __tsan_func_enter(ThreadState *thr, void *pc);
	MOVQ	$__tsan_func_enter(SB), AX
	// racecall<> preserves BX
	CALL	racecall<>(SB)
	MOVQ	BX, DX	// restore function entry context
	RET

// func runtime·racefuncexit()

// (TESTQ x y) if the ANDQ has other uses. If that rewrite happens, then one
// of the args of the ANDQ needs to be saved so it can be used as the arg to TESTQ.
func andWithUse(x, y int) int {
	z := x & y
	// amd64:`TESTQ\s(AX, AX|BX, BX|CX, CX|DX, DX|SI, SI|DI, DI|R8, R8|R9, R9|R10, R10|R11, R11|R12, R12|R13, R13|R15, R15)`
	if z == 0 {
		return 77
	}
	// use z by returning it
	return z
}

// Use a local trampoline, to avoid taking the address of a dynamically exported
// function.
TEXT ·set_crosscall2(SB),NOSPLIT,$0-0
	MOVQ	_crosscall2_ptr(SB), AX
	MOVQ	$crosscall2_trampoline<>(SB), BX
	MOVQ	BX, (AX)
	RET

TEXT crosscall2_trampoline<>(SB),NOSPLIT,$0-0
	JMP	crosscall2(SB)

// Called by C code generated by cmd/cgo.
// func crosscall2(fn, a unsafe.Pointer, n int32, ctxt uintptr)

// library is loaded. For static libraries it is called when the
// final executable starts, during the C runtime initialization
// phase.
TEXT _rt0_386_windows_lib(SB),NOSPLIT,$0x1C
	MOVL	BP, 0x08(SP)
	MOVL	BX, 0x0C(SP)
	MOVL	AX, 0x10(SP)
	MOVL  CX, 0x14(SP)
	MOVL  DX, 0x18(SP)

	// Create a new thread to do the runtime initialization and return.
	MOVL	_cgo_sys_thread_create(SB), AX

#define INVOKE_SYSCALL	INT	$0x80

// func Syscall6(num, a1, a2, a3, a4, a5, a6 uintptr) (r1, r2, errno uintptr)
//
// Syscall # in AX, args in BX CX DX SI DI BP, return in AX
TEXT ·Syscall6(SB),NOSPLIT,$0-40
	MOVL	num+0(FP), AX	// syscall entry
	MOVL	a1+4(FP), BX
	MOVL	a2+8(FP), CX
	MOVL	a3+12(FP), DX
	MOVL	a4+16(FP), SI
	MOVL	a5+20(FP), DI
	MOVL	a6+24(FP), BP
	INVOKE_SYSCALL
	CMPL	AX, $0xfffff001
	JLS	ok

	// Move input to stack in order to free registers
	MOVQ res+0(FP), AX
	MOVQ in1+8(FP), BX
	MOVQ in2+16(FP), CX
	MOVQ sign+24(FP), DX
	MOVQ sel+32(FP), t1
	MOVQ zero+40(FP), t2

	MOVOU (16*0)(BX), X0
	MOVOU (16*1)(BX), X1
	MOVOU (16*2)(BX), X2
	MOVOU (16*3)(BX), X3
	MOVOU (16*4)(BX), X4
	MOVOU (16*5)(BX), X5

	MOVOU X0, x1in(16*0)
	MOVOU X1, x1in(16*1)
	MOVOU X2, y1in(16*0)

	VRCP28PS Z21, K3, Z28                              // 62227d4bcae5
	VRCP28PS 15(DX)(BX*1), K3, Z28                     // 62627d4bcaa41a0f000000
	VRCP28PS -7(CX)(DX*2), K3, Z28                     // 62627d4bcaa451f9ffffff
	VRCP28PS Z13, K3, Z6                               // 62d27d4bcaf5
	VRCP28PS Z21, K3, Z6                               // 62b27d4bcaf5
	VRCP28PS 15(DX)(BX*1), K3, Z6                      // 62f27d4bcab41a0f000000

	MOVQ	old+16(FP), DX		// arg 3 oact
	MOVQ	_cgo_sigaction(SB), AX
	MOVQ	SP, BX			// callee-saved
	ANDQ	$~15, SP		// alignment as per amd64 psABI
	CALL	AX
	MOVQ	BX, SP
	MOVL	AX, ret+24(FP)
	RET

TEXT runtime·sigfwd(SB),NOSPLIT,$0-32
	MOVQ	fn+0(FP),    AX
	MOVL	sig+8(FP),   DI
	MOVQ	info+16(FP), SI
	MOVQ	ctx+24(FP),  DX
	MOVQ	SP, BX		// callee-saved
	ANDQ	$~15, SP	// alignment for x86_64 ABI
	CALL	AX