EBX:  "BX",
	ESP:  "SP",
	EBP:  "BP",
	ESI:  "SI",
	EDI:  "DI",
	R8L:  "R8",
	R9L:  "R9",
	R10L: "R10",
	R11L: "R11",
	R12L: "R12",
	R13L: "R13",
	R14L: "R14",
	R15L: "R15",
	RAX:  "AX",
	RCX:  "CX",
	RDX:  "DX",
	RBX:  "BX",
	RSP:  "SP",
	RBP:  "BP",
	RSI:  "SI",
	RDI:  "DI",
	R8:   "R8",
	R9:   "R9",
	R10:  "R10",
	R11:  "R11",
	R12:  "R12",
	R13:  "R13",

	MOVL CX, 0(SP)
	INCL CX
	MOVL CX, 4(SP)
	INCL CX
	MOVL CX, 8(SP)
	INCL CX
	MOVL CX, 12(SP)
	MOVOU 0(SP), X12

	// Load seed words into next two rows and into DI, SI, R8..R13
	SEED(0, DI, X4)
	SEED(1, SI, X5)
	SEED(2, R8, X6)
	SEED(3, R9, X7)
	SEED(4, R10, X8)
	SEED(5, R11, X9)
	SEED(6, R12, X10)
	SEED(7, R13, X11)

	// Zeros for remaining two matrix entries.

.globl __umodti3
__umodti3:
	# specialized to u128 % u64, so verify that
	test %rcx,%rcx
	jne 1f

	# save divisor
	movq %rdx, %r8

	# reduce top 64 bits mod divisor
	movq %rsi, %rax
	xorl %edx, %edx
	divq %r8

	# reduce full 128-bit mod divisor
	# quotient fits in 64 bits because top 64 bits have been reduced < divisor.

TEXT ·Syscall9(SB),NOSPLIT,$28-52
	NO_LOCAL_POINTERS
	JAL	runtime·entersyscall(SB)
	MOVW	a1+4(FP), R4
	MOVW	a2+8(FP), R5
	MOVW	a3+12(FP), R6
	MOVW	a4+16(FP), R7
	MOVW	a5+20(FP), R8
	MOVW	a6+24(FP), R9
	MOVW	a7+28(FP), R10
	MOVW	a8+32(FP), R11
	MOVW	R8, 16(R29)
	MOVW	R9, 20(R29)
	MOVW	R10, 24(R29)
	MOVW	R11, 28(R29)
	MOVW	trap+0(FP), R2	// syscall entry
	MOVW	R0, R3	// reset R3 to zero as 1-ret SYSCALL keeps it
	SYSCALL

//go:build libfuzzer

#include "go_asm.h"
#include "go_tls.h"
#include "textflag.h"

// Based on race_amd64.s; see commentary there.

#ifdef GOOS_windows
#define RARG0 CX
#define RARG1 DX
#define RARG2 R8
#define RARG3 R9
#else
#define RARG0 DI
#define RARG1 SI
#define RARG2 DX
#define RARG3 CX
#endif

// void runtime·libfuzzerCall4(fn, hookId int, s1, s2 unsafe.Pointer, result uintptr)

	// The C arguments arrive in R0, R1, R2, and R3. We want to
	// pass R0, R1, and R3 to Go, so we push those on the stack.
	// Also, save C callee-save registers R4-R12.
	MOVM.WP	[R0, R1, R3, R4, R5, R6, R7, R8, R9, g, R11, R12], (R13)
	// Finally, save the link register R14. This also puts the
	// arguments we pushed for cgocallback where they need to be,
	// starting at 4(R13).
	MOVW.W	R14, -4(R13)

	CALL	libc_thrkill(SB)
	RET

TEXT runtime·thrsleep_trampoline(SB),NOSPLIT,$0
	MOVL	8(DI), SI		// arg 2 - clock_id
	MOVQ	16(DI), DX		// arg 3 - abstime
	MOVQ	24(DI), CX		// arg 4 - lock
	MOVQ	32(DI), R8		// arg 5 - abort
	MOVQ	0(DI), DI		// arg 1 - id
	CALL	libc_thrsleep(SB)
	RET

TEXT runtime·thrwakeup_trampoline(SB),NOSPLIT,$0
	MOVL	8(DI), SI		// arg 2 - count
	MOVQ	0(DI), DI		// arg 1 - id

func (c *sigctxt) rsi() uint64 { return c.regs().rsi }
func (c *sigctxt) rbp() uint64 { return c.regs().rbp }
func (c *sigctxt) rsp() uint64 { return c.regs().rsp }
func (c *sigctxt) r8() uint64  { return c.regs().r8 }
func (c *sigctxt) r9() uint64  { return c.regs().r9 }
func (c *sigctxt) r10() uint64 { return c.regs().r10 }
func (c *sigctxt) r11() uint64 { return c.regs().r11 }

#include "go_asm.h"
#include "textflag.h"

TEXT ·asyncPreempt(SB),NOSPLIT|NOFRAME,$0-0
	MOVV R1, -480(R3)
	SUBV $480, R3
	MOVV R4, 8(R3)
	MOVV R5, 16(R3)
	MOVV R6, 24(R3)
	MOVV R7, 32(R3)
	MOVV R8, 40(R3)
	MOVV R9, 48(R3)
	MOVV R10, 56(R3)
	MOVV R11, 64(R3)
	MOVV R12, 72(R3)
	MOVV R13, 80(R3)
	MOVV R14, 88(R3)
	MOVV R15, 96(R3)
	MOVV R16, 104(R3)
	MOVV R17, 112(R3)
	MOVV R18, 120(R3)

	VLD1.P	(R8)(R9.SXTX<<2), [V2.B16]                       // ERROR "invalid extended register"
	VLD1.P	(R8)(R9<<2), [V2.B16]                            // ERROR "invalid extended register"
	VST1.P	[V1.B16], (R8)(R9.UXTW)                          // ERROR "invalid extended register"
	VST1.P	[V1.B16], (R8)(R9<<1)                            // ERROR "invalid extended register"