// LTYPE3 rimrem	{ outcode($1, &$2); }
	SUBQ $4, DI
	SUBQ R11, DI
	SUBQ 4(R11), DI
	SUBQ foo+4(SB), DI
	SUBQ $4, 8(R12)
	SUBQ R11, 8(R12)
	SUBQ R11, foo+4(SB)

// LTYPE4 remrim	{ outcode($1, &$2); }
	CMPB	CX, $4

// LTYPER nonrel	{ outcode($1, &$2); }
label:
	JB	-4(PC) // JCS -4(PC)
	JB	label // JCS 17

// LTYPEC spec3	{ outcode($1, &$2); }
	JCS	2(PC)
	JMP	-4(PC)
	JCS	2(PC)
	JMP	label // JMP 17

	MOVQ    $~(1<<63), AX // sign bit mask
	MOVQ    BX, DX
	ANDQ    AX, DX
	MOVQ    $PosInf, AX
	CMPQ    AX, DX
	JLE     notFinite
	// check if argument will overflow
	MOVQ    BX, X0
	MOVSD   $Overflow, X1
	COMISD  X1, X0
	JA      overflow
	MOVSD   $LOG2E, X1
	MULSD   X0, X1
	CVTSD2SL X1, BX // BX = exponent
	CVTSL2SD BX, X1
	CMPB ·useFMA(SB), $1
	JE   avxfma
	MOVSD   $LN2U, X2
	MULSD   X1, X2
	SUBSD   X2, X0

		{name: "CMPL", argLength: 2, reg: gp2flags, asm: "CMPL", typ: "Flags"},                    // arg0 compare to arg1
		{name: "CMPW", argLength: 2, reg: gp2flags, asm: "CMPW", typ: "Flags"},                    // arg0 compare to arg1
		{name: "CMPB", argLength: 2, reg: gp2flags, asm: "CMPB", typ: "Flags"},                    // arg0 compare to arg1

		for x := 0; x <= size-n; x = x*7 + 1 { // offset in src
			for y := 0; y <= size-n; y = y*9 + 1 { // offset in dst
				copy(dst[y:y+n], src[x:x+n])
				copyref(ref[y:y+n], src[x:x+n])
				p := cmpb(dst, ref)
				if p >= 0 {
					t.Fatalf("Copy failed, copying from src[%d:%d] to dst[%d:%d].\nOffset %d is different, %v != %v", x, x+n, y, y+n, p, dst[p], ref[p])
				}
			}
		}
	}
}

//go:build !purego

// func addMulVVW1024(z *uint, x *uint, y uint) (c uint)
// Requires: ADX, BMI2
TEXT ·addMulVVW1024(SB), $0-32
	CMPB ·supportADX+0(SB), $0x01
	JEQ  adx
	MOVQ z+0(FP), CX
	MOVQ x+8(FP), BX
	MOVQ y+16(FP), SI
	XORQ DI, DI

	// Iteration 0
	MOVQ (BX), AX
	MULQ SI
	ADDQ (CX), AX
	ADCQ $0x00, DX
	ADDQ DI, AX

		// CMPx: compare arg0 to arg1.
		{name: "CMPQ", argLength: 2, reg: gp2flags, asm: "CMPQ", typ: "Flags"},
		{name: "CMPL", argLength: 2, reg: gp2flags, asm: "CMPL", typ: "Flags"},
		{name: "CMPW", argLength: 2, reg: gp2flags, asm: "CMPW", typ: "Flags"},
		{name: "CMPB", argLength: 2, reg: gp2flags, asm: "CMPB", typ: "Flags"},

		// CMPxconst: compare arg0 to auxint.

	"jmp",
	"callq",
	"cmpb",
}

var i386GnuNeed = []string{
	"jmp",
	"call",
	"cmp",
}

var armNeed = []string{
	"B main.main(SB)",
	"BL main.Println(SB)",
	"RET",
}

var arm64Need = []string{
	"JMP main.main(SB)",
	"CALL main.Println(SB)",
	"RET",
}

var armGnuNeed = []string{ // for both ARM and AMR64
	"ldr",
	"bl",
	"cmp",
}

var ppcNeed = []string{

	MOVL	$1, 0(SP)
	CALL	runtime·exit(SB)
	CALL	runtime·abort(SB)

has_cpuid:
	MOVL	$0, AX
	CPUID
	MOVL	AX, SI
	CMPL	AX, $0
	JE	nocpuinfo

	CMPL	BX, $0x756E6547  // "Genu"
	JNE	notintel
	CMPL	DX, $0x49656E69  // "ineI"
	JNE	notintel
	CMPL	CX, $0x6C65746E  // "ntel"
	JNE	notintel
	MOVB	$1, runtime·isIntel(SB)
notintel:

	// Load EAX=1 cpuid flags
	MOVL	$1, AX
	CPUID

	v_2 := v.Args[2]
	v_1 := v.Args[1]
	v_0 := v.Args[0]
	b := v.Block
	typ := &b.Func.Config.Types
	// match: (CMPBload {sym} [off] ptr x mem)
	// result: (CMPB (MOVBload {sym} [off] ptr mem) x)
	for {
		off := auxIntToInt32(v.AuxInt)
		sym := auxToSym(v.Aux)
		ptr := v_0
		x := v_1
		mem := v_2
		v.reset(OpAMD64CMPB)

	"CMOVWLS",
	"CMOVWLT",
	"CMOVWMI",
	"CMOVWNE",
	"CMOVWOC",
	"CMOVWOS",
	"CMOVWPC",
	"CMOVWPL",
	"CMOVWPS",
	"CMPB",
	"CMPL",
	"CMPPD",
	"CMPPS",
	"CMPQ",
	"CMPSB",
	"CMPSD",
	"CMPSL",
	"CMPSQ",
	"CMPSS",
	"CMPSW",
	"CMPW",
	"CMPXCHG16B",
	"CMPXCHG8B",
	"CMPXCHGB",
	"CMPXCHGL",
	"CMPXCHGQ",
	"CMPXCHGW",
	"COMISD",
	"COMISS",
	"CPUID",