puma-plus-core 0.1.0

data/go/vendor/golang.org/x/crypto/internal/poly1305/sum_generic.go

@@ -0,0 +1,312 @@
+// Copyright 2018 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// This file provides the generic implementation of Sum and MAC. Other files
+// might provide optimized assembly implementations of some of this code.
+
+package poly1305
+
+import (
+	"encoding/binary"
+	"math/bits"
+)
+
+// Poly1305 [RFC 7539] is a relatively simple algorithm: the authentication tag
+// for a 64 bytes message is approximately
+//
+//     s + m[0:16] * r⁴ + m[16:32] * r³ + m[32:48] * r² + m[48:64] * r  mod  2¹³⁰ - 5
+//
+// for some secret r and s. It can be computed sequentially like
+//
+//     for len(msg) > 0:
+//         h += read(msg, 16)
+//         h *= r
+//         h %= 2¹³⁰ - 5
+//     return h + s
+//
+// All the complexity is about doing performant constant-time math on numbers
+// larger than any available numeric type.
+
+func sumGeneric(out *[TagSize]byte, msg []byte, key *[32]byte) {
+	h := newMACGeneric(key)
+	h.Write(msg)
+	h.Sum(out)
+}
+
+func newMACGeneric(key *[32]byte) macGeneric {
+	m := macGeneric{}
+	initialize(key, &m.macState)
+	return m
+}
+
+// macState holds numbers in saturated 64-bit little-endian limbs. That is,
+// the value of [x0, x1, x2] is x[0] + x[1] * 2⁶⁴ + x[2] * 2¹²⁸.
+type macState struct {
+	// h is the main accumulator. It is to be interpreted modulo 2¹³⁰ - 5, but
+	// can grow larger during and after rounds. It must, however, remain below
+	// 2 * (2¹³⁰ - 5).
+	h [3]uint64
+	// r and s are the private key components.
+	r [2]uint64
+	s [2]uint64
+}
+
+type macGeneric struct {
+	macState
+
+	buffer [TagSize]byte
+	offset int
+}
+
+// Write splits the incoming message into TagSize chunks, and passes them to
+// update. It buffers incomplete chunks.
+func (h *macGeneric) Write(p []byte) (int, error) {
+	nn := len(p)
+	if h.offset > 0 {
+		n := copy(h.buffer[h.offset:], p)
+		if h.offset+n < TagSize {
+			h.offset += n
+			return nn, nil
+		}
+		p = p[n:]
+		h.offset = 0
+		updateGeneric(&h.macState, h.buffer[:])
+	}
+	if n := len(p) - (len(p) % TagSize); n > 0 {
+		updateGeneric(&h.macState, p[:n])
+		p = p[n:]
+	}
+	if len(p) > 0 {
+		h.offset += copy(h.buffer[h.offset:], p)
+	}
+	return nn, nil
+}
+
+// Sum flushes the last incomplete chunk from the buffer, if any, and generates
+// the MAC output. It does not modify its state, in order to allow for multiple
+// calls to Sum, even if no Write is allowed after Sum.
+func (h *macGeneric) Sum(out *[TagSize]byte) {
+	state := h.macState
+	if h.offset > 0 {
+		updateGeneric(&state, h.buffer[:h.offset])
+	}
+	finalize(out, &state.h, &state.s)
+}
+
+// [rMask0, rMask1] is the specified Poly1305 clamping mask in little-endian. It
+// clears some bits of the secret coefficient to make it possible to implement
+// multiplication more efficiently.
+const (
+	rMask0 = 0x0FFFFFFC0FFFFFFF
+	rMask1 = 0x0FFFFFFC0FFFFFFC
+)
+
+// initialize loads the 256-bit key into the two 128-bit secret values r and s.
+func initialize(key *[32]byte, m *macState) {
+	m.r[0] = binary.LittleEndian.Uint64(key[0:8]) & rMask0
+	m.r[1] = binary.LittleEndian.Uint64(key[8:16]) & rMask1
+	m.s[0] = binary.LittleEndian.Uint64(key[16:24])
+	m.s[1] = binary.LittleEndian.Uint64(key[24:32])
+}
+
+// uint128 holds a 128-bit number as two 64-bit limbs, for use with the
+// bits.Mul64 and bits.Add64 intrinsics.
+type uint128 struct {
+	lo, hi uint64
+}
+
+func mul64(a, b uint64) uint128 {
+	hi, lo := bits.Mul64(a, b)
+	return uint128{lo, hi}
+}
+
+func add128(a, b uint128) uint128 {
+	lo, c := bits.Add64(a.lo, b.lo, 0)
+	hi, c := bits.Add64(a.hi, b.hi, c)
+	if c != 0 {
+		panic("poly1305: unexpected overflow")
+	}
+	return uint128{lo, hi}
+}
+
+func shiftRightBy2(a uint128) uint128 {
+	a.lo = a.lo>>2 | (a.hi&3)<<62
+	a.hi = a.hi >> 2
+	return a
+}
+
+// updateGeneric absorbs msg into the state.h accumulator. For each chunk m of
+// 128 bits of message, it computes
+//
+//	h₊ = (h + m) * r  mod  2¹³⁰ - 5
+//
+// If the msg length is not a multiple of TagSize, it assumes the last
+// incomplete chunk is the final one.
+func updateGeneric(state *macState, msg []byte) {
+	h0, h1, h2 := state.h[0], state.h[1], state.h[2]
+	r0, r1 := state.r[0], state.r[1]
+
+	for len(msg) > 0 {
+		var c uint64
+
+		// For the first step, h + m, we use a chain of bits.Add64 intrinsics.
+		// The resulting value of h might exceed 2¹³⁰ - 5, but will be partially
+		// reduced at the end of the multiplication below.
+		//
+		// The spec requires us to set a bit just above the message size, not to
+		// hide leading zeroes. For full chunks, that's 1 << 128, so we can just
+		// add 1 to the most significant (2¹²⁸) limb, h2.
+		if len(msg) >= TagSize {
+			h0, c = bits.Add64(h0, binary.LittleEndian.Uint64(msg[0:8]), 0)
+			h1, c = bits.Add64(h1, binary.LittleEndian.Uint64(msg[8:16]), c)
+			h2 += c + 1
+
+			msg = msg[TagSize:]
+		} else {
+			var buf [TagSize]byte
+			copy(buf[:], msg)
+			buf[len(msg)] = 1
+
+			h0, c = bits.Add64(h0, binary.LittleEndian.Uint64(buf[0:8]), 0)
+			h1, c = bits.Add64(h1, binary.LittleEndian.Uint64(buf[8:16]), c)
+			h2 += c
+
+			msg = nil
+		}
+
+		// Multiplication of big number limbs is similar to elementary school
+		// columnar multiplication. Instead of digits, there are 64-bit limbs.
+		//
+		// We are multiplying a 3 limbs number, h, by a 2 limbs number, r.
+		//
+		//                        h2    h1    h0  x
+		//                              r1    r0  =
+		//                       ----------------
+		//                      h2r0  h1r0  h0r0     <-- individual 128-bit products
+		//            +   h2r1  h1r1  h0r1
+		//               ------------------------
+		//                 m3    m2    m1    m0      <-- result in 128-bit overlapping limbs
+		//               ------------------------
+		//         m3.hi m2.hi m1.hi m0.hi           <-- carry propagation
+		//     +         m3.lo m2.lo m1.lo m0.lo
+		//        -------------------------------
+		//           t4    t3    t2    t1    t0      <-- final result in 64-bit limbs
+		//
+		// The main difference from pen-and-paper multiplication is that we do
+		// carry propagation in a separate step, as if we wrote two digit sums
+		// at first (the 128-bit limbs), and then carried the tens all at once.
+
+		h0r0 := mul64(h0, r0)
+		h1r0 := mul64(h1, r0)
+		h2r0 := mul64(h2, r0)
+		h0r1 := mul64(h0, r1)
+		h1r1 := mul64(h1, r1)
+		h2r1 := mul64(h2, r1)
+
+		// Since h2 is known to be at most 7 (5 + 1 + 1), and r0 and r1 have their
+		// top 4 bits cleared by rMask{0,1}, we know that their product is not going
+		// to overflow 64 bits, so we can ignore the high part of the products.
+		//
+		// This also means that the product doesn't have a fifth limb (t4).
+		if h2r0.hi != 0 {
+			panic("poly1305: unexpected overflow")
+		}
+		if h2r1.hi != 0 {
+			panic("poly1305: unexpected overflow")
+		}
+
+		m0 := h0r0
+		m1 := add128(h1r0, h0r1) // These two additions don't overflow thanks again
+		m2 := add128(h2r0, h1r1) // to the 4 masked bits at the top of r0 and r1.
+		m3 := h2r1
+
+		t0 := m0.lo
+		t1, c := bits.Add64(m1.lo, m0.hi, 0)
+		t2, c := bits.Add64(m2.lo, m1.hi, c)
+		t3, _ := bits.Add64(m3.lo, m2.hi, c)
+
+		// Now we have the result as 4 64-bit limbs, and we need to reduce it
+		// modulo 2¹³⁰ - 5. The special shape of this Crandall prime lets us do
+		// a cheap partial reduction according to the reduction identity
+		//
+		//     c * 2¹³⁰ + n  =  c * 5 + n  mod  2¹³⁰ - 5
+		//
+		// because 2¹³⁰ = 5 mod 2¹³⁰ - 5. Partial reduction since the result is
+		// likely to be larger than 2¹³⁰ - 5, but still small enough to fit the
+		// assumptions we make about h in the rest of the code.
+		//
+		// See also https://speakerdeck.com/gtank/engineering-prime-numbers?slide=23
+
+		// We split the final result at the 2¹³⁰ mark into h and cc, the carry.
+		// Note that the carry bits are effectively shifted left by 2, in other
+		// words, cc = c * 4 for the c in the reduction identity.
+		h0, h1, h2 = t0, t1, t2&maskLow2Bits
+		cc := uint128{t2 & maskNotLow2Bits, t3}
+
+		// To add c * 5 to h, we first add cc = c * 4, and then add (cc >> 2) = c.
+
+		h0, c = bits.Add64(h0, cc.lo, 0)
+		h1, c = bits.Add64(h1, cc.hi, c)
+		h2 += c
+
+		cc = shiftRightBy2(cc)
+
+		h0, c = bits.Add64(h0, cc.lo, 0)
+		h1, c = bits.Add64(h1, cc.hi, c)
+		h2 += c
+
+		// h2 is at most 3 + 1 + 1 = 5, making the whole of h at most
+		//
+		//     5 * 2¹²⁸ + (2¹²⁸ - 1) = 6 * 2¹²⁸ - 1
+	}
+
+	state.h[0], state.h[1], state.h[2] = h0, h1, h2
+}
+
+const (
+	maskLow2Bits    uint64 = 0x0000000000000003
+	maskNotLow2Bits uint64 = ^maskLow2Bits
+)
+
+// select64 returns x if v == 1 and y if v == 0, in constant time.
+func select64(v, x, y uint64) uint64 { return ^(v-1)&x | (v-1)&y }
+
+// [p0, p1, p2] is 2¹³⁰ - 5 in little endian order.
+const (
+	p0 = 0xFFFFFFFFFFFFFFFB
+	p1 = 0xFFFFFFFFFFFFFFFF
+	p2 = 0x0000000000000003
+)
+
+// finalize completes the modular reduction of h and computes
+//
+//	out = h + s  mod  2¹²⁸
+func finalize(out *[TagSize]byte, h *[3]uint64, s *[2]uint64) {
+	h0, h1, h2 := h[0], h[1], h[2]
+
+	// After the partial reduction in updateGeneric, h might be more than
+	// 2¹³⁰ - 5, but will be less than 2 * (2¹³⁰ - 5). To complete the reduction
+	// in constant time, we compute t = h - (2¹³⁰ - 5), and select h as the
+	// result if the subtraction underflows, and t otherwise.
+
+	hMinusP0, b := bits.Sub64(h0, p0, 0)
+	hMinusP1, b := bits.Sub64(h1, p1, b)
+	_, b = bits.Sub64(h2, p2, b)
+
+	// h = h if h < p else h - p
+	h0 = select64(b, h0, hMinusP0)
+	h1 = select64(b, h1, hMinusP1)
+
+	// Finally, we compute the last Poly1305 step
+	//
+	//     tag = h + s  mod  2¹²⁸
+	//
+	// by just doing a wide addition with the 128 low bits of h and discarding
+	// the overflow.
+	h0, c := bits.Add64(h0, s[0], 0)
+	h1, _ = bits.Add64(h1, s[1], c)
+
+	binary.LittleEndian.PutUint64(out[0:8], h0)
+	binary.LittleEndian.PutUint64(out[8:16], h1)
+}

data/go/vendor/golang.org/x/crypto/internal/poly1305/sum_loong64.s

@@ -0,0 +1,123 @@
+// Copyright 2025 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+//go:build gc && !purego
+
+// func update(state *macState, msg []byte)
+TEXT ·update(SB), $0-32
+	MOVV	state+0(FP), R4
+	MOVV	msg_base+8(FP), R5
+	MOVV	msg_len+16(FP), R6
+
+	MOVV	$0x10, R7
+
+	MOVV	(R4), R8	// h0
+	MOVV	8(R4), R9	// h1
+	MOVV	16(R4), R10	// h2
+	MOVV	24(R4), R11	// r0
+	MOVV	32(R4), R12	// r1
+
+	BLT	R6, R7, bytes_between_0_and_15
+
+loop:
+	MOVV	(R5), R14	// msg[0:8]
+	MOVV	8(R5), R16	// msg[8:16]
+	ADDV	R14, R8, R8	// h0 (x1 + y1 = z1', if z1' < x1 then z1' overflow)
+	ADDV	R16, R9, R27
+	SGTU	R14, R8, R24	// h0.carry
+	SGTU	R9, R27, R28
+	ADDV	R27, R24, R9	// h1
+	SGTU	R27, R9, R24
+	OR	R24, R28, R24	// h1.carry
+	ADDV	$0x01, R24, R24
+	ADDV	R10, R24, R10	// h2
+
+	ADDV	$16, R5, R5	// msg = msg[16:]
+
+multiply:
+	MULV	R8, R11, R14	// h0r0.lo
+	MULHVU	R8, R11, R15	// h0r0.hi
+	MULV	R9, R11, R13	// h1r0.lo
+	MULHVU	R9, R11, R16	// h1r0.hi
+	ADDV	R13, R15, R15
+	SGTU	R13, R15, R24
+	ADDV	R24, R16, R16
+	MULV	R10, R11, R25
+	ADDV	R16, R25, R25
+	MULV	R8, R12, R13	// h0r1.lo
+	MULHVU	R8, R12, R16	// h0r1.hi
+	ADDV	R13, R15, R15
+	SGTU	R13, R15, R24
+	ADDV	R24, R16, R16
+	MOVV	R16, R8
+	MULV	R10, R12, R26	// h2r1
+	MULV	R9, R12, R13	// h1r1.lo
+	MULHVU	R9, R12, R16	// h1r1.hi
+	ADDV	R13, R25, R25
+	ADDV	R16, R26, R27
+	SGTU	R13, R25, R24
+	ADDV	R27, R24, R26
+	ADDV	R8, R25, R25
+	SGTU	R8, R25, R24
+	ADDV	R24, R26, R26
+	AND	$3, R25, R10
+	AND	$-4, R25, R17
+	ADDV	R17, R14, R8
+	ADDV	R26, R15, R27
+	SGTU	R17, R8, R24
+	SGTU	R26, R27, R28
+	ADDV	R27, R24, R9
+	SGTU	R27, R9, R24
+	OR	R24, R28, R24
+	ADDV	R24, R10, R10
+	SLLV	$62, R26, R27
+	SRLV	$2, R25, R28
+	SRLV	$2, R26, R26
+	OR	R27, R28, R25
+	ADDV	R25, R8, R8
+	ADDV	R26, R9, R27
+	SGTU	R25, R8, R24
+	SGTU	R26, R27, R28
+	ADDV	R27, R24, R9
+	SGTU	R27, R9, R24
+	OR	R24, R28, R24
+	ADDV	R24, R10, R10
+
+	SUBV	$16, R6, R6
+	BGE	R6, R7, loop
+
+bytes_between_0_and_15:
+	BEQ	R6, R0, done
+	MOVV	$1, R14
+	XOR	R15, R15
+	ADDV	R6, R5, R5
+
+flush_buffer:
+	MOVBU	-1(R5), R25
+	SRLV	$56, R14, R24
+	SLLV	$8, R15, R28
+	SLLV	$8, R14, R14
+	OR	R24, R28, R15
+	XOR	R25, R14, R14
+	SUBV	$1, R6, R6
+	SUBV	$1, R5, R5
+	BNE	R6, R0, flush_buffer
+
+	ADDV	R14, R8, R8
+	SGTU	R14, R8, R24
+	ADDV	R15, R9, R27
+	SGTU	R15, R27, R28
+	ADDV	R27, R24, R9
+	SGTU	R27, R9, R24
+	OR	R24, R28, R24
+	ADDV	R10, R24, R10
+
+	MOVV	$16, R6
+	JMP	multiply
+
+done:
+	MOVV	R8, (R4)
+	MOVV	R9, 8(R4)
+	MOVV	R10, 16(R4)
+	RET

data/go/vendor/golang.org/x/crypto/internal/poly1305/sum_ppc64x.s

@@ -0,0 +1,187 @@
+// Copyright 2019 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+//go:build gc && !purego && (ppc64 || ppc64le)
+
+#include "textflag.h"
+
+// This was ported from the amd64 implementation.
+
+#ifdef GOARCH_ppc64le
+#define LE_MOVD MOVD
+#define LE_MOVWZ MOVWZ
+#define LE_MOVHZ MOVHZ
+#else
+#define LE_MOVD MOVDBR
+#define LE_MOVWZ MOVWBR
+#define LE_MOVHZ MOVHBR
+#endif
+
+#define POLY1305_ADD(msg, h0, h1, h2, t0, t1, t2) \
+	LE_MOVD (msg)( R0), t0; \
+	LE_MOVD (msg)(R24), t1; \
+	MOVD $1, t2;     \
+	ADDC t0, h0, h0; \
+	ADDE t1, h1, h1; \
+	ADDE t2, h2;     \
+	ADD  $16, msg
+
+#define POLY1305_MUL(h0, h1, h2, r0, r1, t0, t1, t2, t3, t4, t5) \
+	MULLD  r0, h0, t0;  \
+	MULHDU r0, h0, t1;  \
+	MULLD  r0, h1, t4;  \
+	MULHDU r0, h1, t5;  \
+	ADDC   t4, t1, t1;  \
+	MULLD  r0, h2, t2;  \
+	MULHDU r1, h0, t4;  \
+	MULLD  r1, h0, h0;  \
+	ADDE   t5, t2, t2;  \
+	ADDC   h0, t1, t1;  \
+	MULLD  h2, r1, t3;  \
+	ADDZE  t4, h0;      \
+	MULHDU r1, h1, t5;  \
+	MULLD  r1, h1, t4;  \
+	ADDC   t4, t2, t2;  \
+	ADDE   t5, t3, t3;  \
+	ADDC   h0, t2, t2;  \
+	MOVD   $-4, t4;     \
+	ADDZE  t3;          \
+	RLDICL $0, t2, $62, h2; \
+	AND    t2, t4, h0;  \
+	ADDC   t0, h0, h0;  \
+	ADDE   t3, t1, h1;  \
+	SLD    $62, t3, t4; \
+	SRD    $2, t2;      \
+	ADDZE  h2;          \
+	OR     t4, t2, t2;  \
+	SRD    $2, t3;      \
+	ADDC   t2, h0, h0;  \
+	ADDE   t3, h1, h1;  \
+	ADDZE  h2
+
+// func update(state *[7]uint64, msg []byte)
+TEXT ·update(SB), $0-32
+	MOVD state+0(FP), R3
+	MOVD msg_base+8(FP), R4
+	MOVD msg_len+16(FP), R5
+
+	MOVD 0(R3), R8   // h0
+	MOVD 8(R3), R9   // h1
+	MOVD 16(R3), R10 // h2
+	MOVD 24(R3), R11 // r0
+	MOVD 32(R3), R12 // r1
+
+	MOVD $8, R24
+
+	CMP R5, $16
+	BLT bytes_between_0_and_15
+
+loop:
+	POLY1305_ADD(R4, R8, R9, R10, R20, R21, R22)
+
+	PCALIGN $16
+multiply:
+	POLY1305_MUL(R8, R9, R10, R11, R12, R16, R17, R18, R14, R20, R21)
+	ADD $-16, R5
+	CMP R5, $16
+	BGE loop
+
+bytes_between_0_and_15:
+	CMP  R5, $0
+	BEQ  done
+	MOVD $0, R16 // h0
+	MOVD $0, R17 // h1
+
+flush_buffer:
+	CMP R5, $8
+	BLE just1
+
+	MOVD $8, R21
+	SUB  R21, R5, R21
+
+	// Greater than 8 -- load the rightmost remaining bytes in msg
+	// and put into R17 (h1)
+	LE_MOVD (R4)(R21), R17
+	MOVD $16, R22
+
+	// Find the offset to those bytes
+	SUB R5, R22, R22
+	SLD $3, R22
+
+	// Shift to get only the bytes in msg
+	SRD R22, R17, R17
+
+	// Put 1 at high end
+	MOVD $1, R23
+	SLD  $3, R21
+	SLD  R21, R23, R23
+	OR   R23, R17, R17
+
+	// Remainder is 8
+	MOVD $8, R5
+
+just1:
+	CMP R5, $8
+	BLT less8
+
+	// Exactly 8
+	LE_MOVD (R4), R16
+
+	CMP R17, $0
+
+	// Check if we've already set R17; if not
+	// set 1 to indicate end of msg.
+	BNE  carry
+	MOVD $1, R17
+	BR   carry
+
+less8:
+	MOVD  $0, R16   // h0
+	MOVD  $0, R22   // shift count
+	CMP   R5, $4
+	BLT   less4
+	LE_MOVWZ (R4), R16
+	ADD   $4, R4
+	ADD   $-4, R5
+	MOVD  $32, R22
+
+less4:
+	CMP   R5, $2
+	BLT   less2
+	LE_MOVHZ (R4), R21
+	SLD   R22, R21, R21
+	OR    R16, R21, R16
+	ADD   $16, R22
+	ADD   $-2, R5
+	ADD   $2, R4
+
+less2:
+	CMP   R5, $0
+	BEQ   insert1
+	MOVBZ (R4), R21
+	SLD   R22, R21, R21
+	OR    R16, R21, R16
+	ADD   $8, R22
+
+insert1:
+	// Insert 1 at end of msg
+	MOVD $1, R21
+	SLD  R22, R21, R21
+	OR   R16, R21, R16
+
+carry:
+	// Add new values to h0, h1, h2
+	ADDC  R16, R8
+	ADDE  R17, R9
+	ADDZE R10, R10
+	MOVD  $16, R5
+	ADD   R5, R4
+	BR    multiply
+
+done:
+	// Save h0, h1, h2 in state
+	MOVD R8, 0(R3)
+	MOVD R9, 8(R3)
+	MOVD R10, 16(R3)
+	RET

data/go/vendor/golang.org/x/crypto/internal/poly1305/sum_s390x.go

@@ -0,0 +1,76 @@
+// Copyright 2018 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+//go:build gc && !purego
+
+package poly1305
+
+import (
+	"golang.org/x/sys/cpu"
+)
+
+// updateVX is an assembly implementation of Poly1305 that uses vector
+// instructions. It must only be called if the vector facility (vx) is
+// available.
+//
+//go:noescape
+func updateVX(state *macState, msg []byte)
+
+// mac is a replacement for macGeneric that uses a larger buffer and redirects
+// calls that would have gone to updateGeneric to updateVX if the vector
+// facility is installed.
+//
+// A larger buffer is required for good performance because the vector
+// implementation has a higher fixed cost per call than the generic
+// implementation.
+type mac struct {
+	macState
+
+	buffer [16 * TagSize]byte // size must be a multiple of block size (16)
+	offset int
+}
+
+func (h *mac) Write(p []byte) (int, error) {
+	nn := len(p)
+	if h.offset > 0 {
+		n := copy(h.buffer[h.offset:], p)
+		if h.offset+n < len(h.buffer) {
+			h.offset += n
+			return nn, nil
+		}
+		p = p[n:]
+		h.offset = 0
+		if cpu.S390X.HasVX {
+			updateVX(&h.macState, h.buffer[:])
+		} else {
+			updateGeneric(&h.macState, h.buffer[:])
+		}
+	}
+
+	tail := len(p) % len(h.buffer) // number of bytes to copy into buffer
+	body := len(p) - tail          // number of bytes to process now
+	if body > 0 {
+		if cpu.S390X.HasVX {
+			updateVX(&h.macState, p[:body])
+		} else {
+			updateGeneric(&h.macState, p[:body])
+		}
+	}
+	h.offset = copy(h.buffer[:], p[body:]) // copy tail bytes - can be 0
+	return nn, nil
+}
+
+func (h *mac) Sum(out *[TagSize]byte) {
+	state := h.macState
+	remainder := h.buffer[:h.offset]
+
+	// Use the generic implementation if we have 2 or fewer blocks left
+	// to sum. The vector implementation has a higher startup time.
+	if cpu.S390X.HasVX && len(remainder) > 2*TagSize {
+		updateVX(&state, remainder)
+	} else if len(remainder) > 0 {
+		updateGeneric(&state, remainder)
+	}
+	finalize(out, &state.h, &state.s)
+}