MOVQ DI, (R8)
	RET

// handle for lengths < 16
small:
	TESTQ	BX, BX
	JEQ	failure

	// Check if we'll load across a page boundary.
	LEAQ	16(SI), AX
	TESTW	$0xff0, AX
	JEQ	endofpage

	MOVOU	(SI), X1 // Load data
	PCMPEQB	X0, X1	// Compare target byte with each byte in data.
	PMOVMSKB X1, DX	// Move result bits to integer register.
	BSFL	DX, DX	// Find first set bit.

	v[i] = (v[i] & 0xFF00) >> 6
	i++
	// ppc64x: "MOVW\tR0"
	v[i] = (v[i] & 0xFF) >> 8
	i++
	// ppc64x: "MOVW\tR0"
	v[i] = (v[i] & 0xF000000) >> 28
	i++
	// ppc64x: "RLWNM\t[$]26, R[0-9]+, [$]24, [$]31, R[0-9]+"
	v[i] = (v[i] >> 6) & 0xFF
	i++
	// ppc64x: "RLWNM\t[$]26, R[0-9]+, [$]12, [$]19, R[0-9]+"
	v[i] = (v[i] >> 6) & 0xFF000
	i++
	// ppc64x: "MOVW\tR0"

			{".debug_ranges", SHT_PROGBITS, 0x0, 0x0, 0xe80, 0x90, 0x0, 0x0, 0x10, 0x0, 0x90},
			{".shstrtab", SHT_STRTAB, 0x0, 0x0, 0xf10, 0x149, 0x0, 0x0, 0x1, 0x0, 0x149},
			{".symtab", SHT_SYMTAB, 0x0, 0x0, 0x19a0, 0x6f0, 0x24, 0x39, 0x8, 0x18, 0x6f0},
			{".strtab", SHT_STRTAB, 0x0, 0x0, 0x2090, 0x1fc, 0x0, 0x0, 0x1, 0x0, 0x1fc},
		},
		[]ProgHeader{

	ADDQ DX, R12
end:
	MOVQ R12, (R8)
	RET

// handle for lengths < 16
small:
	TESTQ	BX, BX
	JEQ	endzero

	// Check if we'll load across a page boundary.
	LEAQ	16(SI), AX
	TESTW	$0xff0, AX
	JEQ	endofpage

	// We must ignore high bytes as they aren't part of our slice.
	// Create mask.
	MOVB BX, CX
	MOVQ $1, R10
	SALQ CL, R10
	SUBQ $1, R10

	// Load data
	MOVOU	(SI), X1

		t.Errorf("got %d %d 0x%x, expected -32 -128 0x7fffffe0,", v, w, z)
	}
	if v, w, z := variableShiftOverflow16x16(-64, makeU16(0xffff), 2); v != -32 || w != -128 || z != 0x7fe0 {
		t.Errorf("got %d %d 0x%x, expected -32 -128 0x7fe0", v, w, z)
	}
	if v, w, z := variableShiftOverflow8x16(-64, makeU16(0xffff), 2); v != -32 || w != -128 || z != 0x60 {

	a[i] = bits.RotateLeft32(a[i], 16) & 0xFF0000
	i++
	// ppc64x: "RLWNM\t[$]16, R[0-9]+, [$]8, [$]15, R[0-9]+"
	a[i] = bits.RotateLeft32(a[i]&0xFF, 16)
	i++
	// ppc64x: "RLWNM\t[$]4, R[0-9]+, [$]20, [$]27, R[0-9]+"
	a[i] = bits.RotateLeft32(a[i], 4) & 0xFF0
	i++
	// ppc64x: "RLWNM\t[$]16, R[0-9]+, [$]24, [$]31, R[0-9]+"
	a[i] = bits.RotateLeft32(a[i]&0xFF0000, 16)
	i++

	PCMPEQB X1, X2
	PMOVMSKB X2, SI
	CMPQ  SI, $0xffff
	JE   success
	ADDQ $1,DI
	CMPQ DI,DX
	JB loop16
	JMP fail
_17_or_more:
	CMPQ AX, $31
	JA   _32_or_more
	LEAQ 1(DI)(DX*1), DX
	SUBQ AX, DX
	MOVOU -16(R8)(AX*1), X0
	MOVOU (R8), X1
loop17to31:
	MOVOU (DI), X2
	PCMPEQB X1,X2
	PMOVMSKB X2, SI
	CMPQ  SI, $0xffff
	JE   partial_success17to31
	ADDQ $1,DI
	CMPQ DI,DX

	use(a1k[i&^1])
	use(a1k[i8&^0])
	use(a1k[i8&^-128]) // ERROR "index bounds check elided"
	use(a1k[ui8&^1])   // ERROR "index bounds check elided"
	use(a1k[ui16&^0xf000])
	use(a1k[ui16&^0xff00]) // ERROR "index bounds check elided"

	// Right shift cuts the effective number of bits in the index,
	// but only for unsigned (signed stays negative).
	use(s[i32>>22])
	use(a1[i32>>22])

	SWPW.S	R1, (R2), R3       // ERROR "invalid .S suffix"
	MOVF.S	(R0), F1           // ERROR "invalid .S suffix"
	MOVF.S	F9, (R4)           // ERROR "invalid .S suffix"
	MOVF.S	0xfff0(R0), F1     // ERROR "invalid .S suffix"
	MOVF.S	F9, 0xfff0(R4)     // ERROR "invalid .S suffix"
	ADDF.S	F1, F2, F3         // ERROR "invalid .S suffix"
	SUBD.U	F1, F2             // ERROR "invalid .U suffix"

	MOVD $2, R15      // Set up index

	// Set up masks for use with VSEL
	MOVD   $0xff, R21        // Set up mask 0xff000000ff000000...
	SLD    $24, R21
	MTVSRD R21, V10
	VSPLTW $1, V10, V29
	VSLDOI $2, V29, V29, V30 // Mask 0x0000ff000000ff00...
	MOVD   $0xffff, R21
	SLD    $16, R21
	MTVSRD R21, V10
	VSPLTW $1, V10, V31      // Mask 0xffff0000ffff0000...