SHA512ROUND(V3, V0, V4, V2, V1, V26, V30, V18, V19, V17, V14, V15)
	SHA512ROUND(V2, V3, V1, V4, V0, V27, V31, V19, V12, V18, V15, V16)
	SHA512ROUND(V4, V2, V0, V1, V3, V28, V24, V12, V13, V19, V16, V17)
	SHA512ROUND(V1, V4, V3, V0, V2, V29, V25, V13, V14, V12, V17, V18)

	SHA512ROUND(V0, V1, V2, V3, V4, V30, V26, V14, V15, V13, V18, V19)
	SHA512ROUND(V3, V0, V4, V2, V1, V31, V27, V15, V16, V14, V19, V12)

	LVX	(R3+off112),V17
	VPMSUMW	V15,V17,V15

	VXOR	V19,V15,V19
v14:	VXOR	V20,V14,V20
v13:	VXOR	V19,V13,V19
v12:	VXOR	V20,V12,V20
v11:	VXOR	V19,V11,V19
v10:	VXOR	V20,V10,V20
v9:	VXOR	V19,V9,V19
v8:	VXOR	V20,V8,V20
v7:	VXOR	V19,V7,V19
v6:	VXOR	V20,V6,V20
v5:	VXOR	V19,V5,V19
v4:	VXOR	V20,V4,V20
v3:	VXOR	V19,V3,V19
v2:	VXOR	V20,V2,V20
v1:	VXOR	V19,V1,V19
v0:	VXOR	V20,V0,V20

	VXOR	V19,V20,V0

    T6 v6, T7 v7, T8 v8, T9 v9, T10 v10, T11 v11, T12 v12, T13 v13, T14 v14,
    T15 v15, T16 v16, T17 v17, T18 v18, T19 v19) {
  return internal::ValueArray19<T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11,
      T12, T13, T14, T15, T16, T17, T18, T19>(v1, v2, v3, v4, v5, v6, v7, v8,
      v9, v10, v11, v12, v13, v14, v15, v16, v17, v18, v19);
}

template <typename T1, typename T2, typename T3, typename T4, typename T5,

	LXVW4X (KEY)(R8), VS50

	// Load CNT, NONCE into V19
	LXVW4X (CNT)(R0), VS51

	// Clear V27
	VXOR V27, V27, V27

	// V28
	LXVW4X (CONSTBASE)(R11), VS60

	// Load mask constants for VPERMXOR
	LXVW4X (MASK)(R0), V20
	LXVW4X (MASK)(R20), V21

	// splat slot from V19 -> V26
	VSPLTW $0, V19, V26

	VSLDOI $4, V19, V27, V19
	VSLDOI $12, V27, V19, V19

	VADDUWM V26, V28, V26

	MOVD $10, R14

	VLD1.P	16(R1), [V7.B16]
	VLD1	(R2), [V19.S4]                               // load constant k0-k79
	VMOV	V0.B16, V2.B16
	VMOV	V20.S[0], V1
	VMOV	V2.B16, V3.B16
	VDUP	V19.S[0], V17.S4
	VREV32	V4.B16, V4.B16                               // prepare for using message in Byte format
	VREV32	V5.B16, V5.B16
	VREV32	V6.B16, V6.B16
	VREV32	V7.B16, V7.B16


	VDUP	V19.S[1], V18.S4
	VADD	V17.S4, V4.S4, V16.S4

	SHA512ROUND1(V7, V0, V1, V2, V3, V4, V5, V6, V17, V18, V19, V11, V16, R_x090)
	SHA512ROUND1(V6, V7, V0, V1, V2, V3, V4, V5, V18, V19, V20, V12, V17, R_x0a0)
	SHA512ROUND1(V5, V6, V7, V0, V1, V2, V3, V4, V19, V20, V21, V13, V18, R_x0b0)
	SHA512ROUND1(V4, V5, V6, V7, V0, V1, V2, V3, V20, V21, V22, V14, V19, R_x0c0)
	SHA512ROUND1(V3, V4, V5, V6, V7, V0, V1, V2, V21, V22, V23, V15, V20, R_x0d0)

	SHA256ROUND1(V7, V0, V1, V2, V3, V4, V5, V6, V17, V18, V19, V11, V16, R_x090)
	SHA256ROUND1(V6, V7, V0, V1, V2, V3, V4, V5, V18, V19, V20, V12, V17, R_x0a0)
	SHA256ROUND1(V5, V6, V7, V0, V1, V2, V3, V4, V19, V20, V21, V13, V18, R_x0b0)
	SHA256ROUND1(V4, V5, V6, V7, V0, V1, V2, V3, V20, V21, V22, V14, V19, R_x0c0)
	SHA256ROUND1(V3, V4, V5, V6, V7, V0, V1, V2, V21, V22, V23, V15, V20, R_x0d0)

      T10 v10, T11 v11, T12 v12, T13 v13, T14 v14, T15 v15, T16 v16, T17 v17,
      T18 v18, T19 v19) : v1_(v1), v2_(v2), v3_(v3), v4_(v4), v5_(v5), v6_(v6),
      v7_(v7), v8_(v8), v9_(v9), v10_(v10), v11_(v11), v12_(v12), v13_(v13),
      v14_(v14), v15_(v15), v16_(v16), v17_(v17), v18_(v18), v19_(v19) {}

  template <typename T>
  operator ParamGenerator<T>() const {

	MOVD	p_len+32(FP), R3                           // message length
	VLD1	(R0), [V0.S4, V1.S4]                       // load h(a,b,c,d,e,f,g,h)
	VLD1.P	64(R2), [V16.S4, V17.S4, V18.S4, V19.S4]
	VLD1.P	64(R2), [V20.S4, V21.S4, V22.S4, V23.S4]
	VLD1.P	64(R2), [V24.S4, V25.S4, V26.S4, V27.S4]
	VLD1	(R2), [V28.S4, V29.S4, V30.S4, V31.S4]     //load 64*4bytes K constant(K0-K63)

blockloop:

	VCIPHERLAST     V18, V23, V18; \
	VCIPHERLAST     V19, V23, V19; \
	VCIPHERLAST     V20, V23, V20; \
	VCIPHERLAST     V21, V23, V21; \
	VCIPHERLAST     V22, V23, V22; \
	XXLXOR          V1, V15, V1; \
	XXLXOR          V2, V16, V2; \
	XXLXOR          V3, V17, V3; \
	XXLXOR          V4, V18, V4; \
	XXLXOR          V5, V19, V5; \
	XXLXOR          V6, V20, V6; \
	XXLXOR          V7, V21, V7; \