// R4 = length
	// R6 = byte to count
	MTVRD	R6, V1		// move compare byte
	MOVD	R6, R5
	VSPLTB	$7, V1, V1	// replicate byte across V1
	BR	countbytebody<>(SB)

TEXT ·CountString<ABIInternal>(SB), NOSPLIT|NOFRAME, $0-32
	// R3 = byte array pointer
	// R4 = length
	// R5 = byte to count
	MTVRD	R5, V1		// move compare byte
	VSPLTB	$7, V1, V1	// replicate byte across V1
	BR	countbytebody<>(SB)

// R3: addr of string

	"VCMPNEB",
	"VCMPNEBCC",
	"VCMPNEH",
	"VCMPNEHCC",
	"VCMPNEW",
	"VCMPNEWCC",
	"VPERM",
	"VPERMXOR",
	"VPERMR",
	"VBPERMQ",
	"VBPERMD",
	"VSEL",
	"VSPLTB",
	"VSPLTH",
	"VSPLTW",
	"VSPLTISB",
	"VSPLTISH",
	"VSPLTISW",
	"VCIPH",
	"VCIPHER",
	"VCIPHERLAST",
	"VNCIPH",
	"VNCIPHER",
	"VNCIPHERLAST",
	"VSBOX",
	"VSHASIGMA",
	"VSHASIGMAW",
	"VSHASIGMAD",

	CMP     R8, R4
	BGE     loopexit        // Already at end?

	// vectorize if len(z) is >=3, else jump to scalar loop
	CMP     R4, $3
	BLT     scalar
	MTVSRD  R9, VS38        // s
	VSPLTB  $7, V6, V4
	MTVSRD  R5, VS39        // ŝ
	VSPLTB  $7, V7, V2
	ADD     $-2, R4, R16
	PCALIGN $16
loopback:
	ADD     $-1, R8, R10
	SLD     $3, R10
	LXVD2X  (R6)(R10), VS32 // load x[i-1], x[i]
	SLD     $3, R8, R12

	LXVDSX   (R1)(R18), SEL1 // VLREPG idx+32(FP), SEL1
	VSPLTB   $7, SEL1, IDX    // splat byte
	VSPLTISB $1, ONE          // VREPIB $1, ONE
	VSPLTISB $1, SEL2         // VREPIB $1, SEL2
	MOVD     $17, COUNT
	MOVD     COUNT, CTR       // set up ctr

	VSPLTISB $0, X1H // VZERO  X1H
	VSPLTISB $0, X1L // VZERO  X1L
	VSPLTISB $0, Y1H // VZERO  Y1H
	VSPLTISB $0, Y1L // VZERO  Y1L
	VSPLTISB $0, Z1H // VZERO  Z1H
	VSPLTISB $0, Z1L // VZERO  Z1L

	MOVD   $0x30, R10
	LXVD2X (HTBL)(R0), VH // Load H

	VSPLTISB $-16, XC2           // 0xf0
	VSPLTISB $1, T0              // one
	VADDUBM  XC2, XC2, XC2       // 0xe0
	VXOR     ZERO, ZERO, ZERO
	VOR      XC2, T0, XC2        // 0xe1
	VSLDOI   $15, XC2, ZERO, XC2 // 0xe1...
	VSLDOI   $1, ZERO, T0, T1    // ...1
	VADDUBM  XC2, XC2, XC2       // 0xc2...
	VSPLTISB $7, T2
	VOR      XC2, T1, XC2        // 0xc2....01

	VBPERMQ V1, V2, V3              // 1061154c
	VBPERMD V1, V2, V3              // 106115cc
	VSEL V1, V2, V3, V4             // 108110ea
	VSPLTB $1, V1, V2               // 10410a0c
	VSPLTH $1, V1, V2               // 10410a4c
	VSPLTW $1, V1, V2               // 10410a8c
	VSPLTISB $1, V1                 // 1021030c
	VSPLTISW $1, V1                 // 1021038c
	VSPLTISH $1, V1                 // 1021034c

		case AVBPERMQ: /* vbpermq, vbpermd */
			opset(AVBPERMD, r0)

		case AVSEL: /* vsel */
			opset(AVSEL, r0)

		case AVSPLTB: /* vspltb, vsplth, vspltw */
			opset(AVSPLTH, r0)
			opset(AVSPLTW, r0)

		case AVSPLTISB: /* vspltisb, vspltish, vspltisw */
			opset(AVSPLTISH, r0)
			opset(AVSPLTISW, r0)

		case AVCIPH: /* vcipher, vcipherlast */
			opset(AVCIPHER, r0)

	VXOR	IN0, TMP, IN0       // vxor 1,1,6
	VSLDOI	$12, ZERO, TMP, TMP // vsldoi 6,0,6,12
	VXOR	IN0, TMP, IN0       // vxor 1,1,6

	VSLDOI	$8, ZERO, IN1, STAGE  // vsldoi 7,0,2,8
	VSPLTW	$3, IN0, TMP          // vspltw 6,1,3
	VXOR	TMP, IN1, TMP         // vxor 6,6,2
	VSLDOI	$12, ZERO, IN1, IN1   // vsldoi 2,0,2,12
	VADDUWM	RCON, RCON, RCON      // vadduwm 4,4,4
	VXOR	IN1, TMP, IN1         // vxor 2,2,6

	// splat values from V17, V18 into V4-V11
	VSPLTW $0, V17, V4
	VSPLTW $1, V17, V5
	VSPLTW $2, V17, V6
	VSPLTW $3, V17, V7
	VSPLTW $0, V18, V8
	VSPLTW $1, V18, V9
	VSPLTW $2, V18, V10
	VSPLTW $3, V18, V11

	// VOR
	VOR V26, V26, V12

	// splat values from V19 -> V13, V14, V15
	VSPLTW $1, V19, V13
	VSPLTW $2, V19, V14
	VSPLTW $3, V19, V15

	// splat   const values

	"VSUMQ",
	"VSUMQF",
	"VSUMQG",
	"VSUM",
	"VSUMB",
	"VSUMH",
	"VTM",
	"VUPH",
	"VUPHB",
	"VUPHH",
	"VUPHF",
	"VUPLH",
	"VUPLHB",
	"VUPLHH",
	"VUPLHF",
	"VUPLL",
	"VUPLLB",
	"VUPLLH",
	"VUPLLF",
	"VUPL",
	"VUPLB",
	"VUPLHW",
	"VUPLF",
	"VMSLG",
	"VMSLEG",
	"VMSLOG",
	"VMSLEOG",
	"NOPH",
	"BYTE",
	"WORD",
	"DWORD",
	"BRRK",
	"LAST",