go.mod: update golang.org/x dependencies

Update golang.org/x dependencies to their latest tagged versions.

Change-Id: I1e9480e5897a08cac746309acdc8bbdf0bce8060
Reviewed-on: https://go-review.googlesource.com/c/crypto/+/663256
Reviewed-by: David Chase <drchase@google.com>
Reviewed-by: Dmitri Shuralyov <dmitshur@google.com>
Auto-Submit: Gopher Robot <gobot@golang.org>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>

README.md

@@ -2,20 +2,17 @@
 
 [![Go Reference](https://pkg.go.dev/badge/golang.org/x/crypto.svg)](https://pkg.go.dev/golang.org/x/crypto)
 
-This repository holds supplementary Go cryptography libraries.
-
-## Download/Install
-
-The easiest way to install is to run `go get -u golang.org/x/crypto/...`. You
-can also manually git clone the repository to `$GOPATH/src/golang.org/x/crypto`.
+This repository holds supplementary Go cryptography packages.
 
 ## Report Issues / Send Patches
 
 This repository uses Gerrit for code changes. To learn how to submit changes to
-this repository, see https://golang.org/doc/contribute.html.
+this repository, see https://go.dev/doc/contribute.
+
+The git repository is https://go.googlesource.com/crypto.
 
 The main issue tracker for the crypto repository is located at
-https://github.com/golang/go/issues. Prefix your issue with "x/crypto:" in the
+https://go.dev/issues. Prefix your issue with "x/crypto:" in the
 subject line, so it is easy to find.
 
 Note that contributions to the cryptography package receive additional scrutiny

acme/acme.go

@@ -514,7 +514,11 @@ func (c *Client) Accept(ctx context.Context, chal *Challenge) (*Challenge, error
 		return nil, err
 	}
 
-	res, err := c.post(ctx, nil, chal.URI, json.RawMessage("{}"), wantStatus(
+	payload := json.RawMessage("{}")
+	if len(chal.Payload) != 0 {
+		payload = chal.Payload
+	}
+	res, err := c.post(ctx, nil, chal.URI, payload, wantStatus(
 		http.StatusOK,       // according to the spec
 		http.StatusAccepted, // Let's Encrypt: see https://goo.gl/WsJ7VT (acme-divergences.md)
 	))

acme/acme_test.go

@@ -757,7 +757,7 @@ func TestTLSALPN01ChallengeCert(t *testing.T) {
 }
 
 func TestTLSChallengeCertOpt(t *testing.T) {
-	key, err := rsa.GenerateKey(rand.Reader, 512)
+	key, err := rsa.GenerateKey(rand.Reader, 1024)
 	if err != nil {
 		t.Fatal(err)
 	}

acme/autocert/autocert.go

@@ -292,6 +292,10 @@ func (m *Manager) GetCertificate(hello *tls.ClientHelloInfo) (*tls.Certificate,
 	}
 
 	// regular domain
+	if err := m.hostPolicy()(ctx, name); err != nil {
+		return nil, err
+	}
+
 	ck := certKey{
 		domain: strings.TrimSuffix(name, "."), // golang.org/issue/18114
 		isRSA:  !supportsECDSA(hello),
@@ -305,9 +309,6 @@ func (m *Manager) GetCertificate(hello *tls.ClientHelloInfo) (*tls.Certificate,
 	}
 
 	// first-time
-	if err := m.hostPolicy()(ctx, name); err != nil {
-		return nil, err
-	}
 	cert, err = m.createCert(ctx, ck)
 	if err != nil {
 		return nil, err

acme/autocert/autocert_test.go

@@ -619,7 +619,7 @@ func TestCache(t *testing.T) {
 		PrivateKey:  ecdsaKey,
 	}
 
-	rsaKey, err := rsa.GenerateKey(rand.Reader, 512)
+	rsaKey, err := rsa.GenerateKey(rand.Reader, 1024)
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -694,7 +694,7 @@ func TestValidCert(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
-	key3, err := rsa.GenerateKey(rand.Reader, 512)
+	key3, err := rsa.GenerateKey(rand.Reader, 1024)
 	if err != nil {
 		t.Fatal(err)
 	}

acme/autocert/internal/acmetest/ca.go

@@ -308,7 +308,7 @@ func (ca *CAServer) handle(w http.ResponseWriter, r *http.Request) {
 		}
 
 		if err := decodePayload(&req, r.Body); err != nil {
-			ca.httpErrorf(w, http.StatusBadRequest, err.Error())
+			ca.httpErrorf(w, http.StatusBadRequest, "%v", err)
 			return
 		}
 
@@ -328,7 +328,7 @@ func (ca *CAServer) handle(w http.ResponseWriter, r *http.Request) {
 			Identifiers []struct{ Value string }
 		}
 		if err := decodePayload(&req, r.Body); err != nil {
-			ca.httpErrorf(w, http.StatusBadRequest, err.Error())
+			ca.httpErrorf(w, http.StatusBadRequest, "%v", err)
 			return
 		}
 		ca.mu.Lock()
@@ -352,7 +352,7 @@ func (ca *CAServer) handle(w http.ResponseWriter, r *http.Request) {
 		defer ca.mu.Unlock()
 		o, err := ca.storedOrder(strings.TrimPrefix(r.URL.Path, "/orders/"))
 		if err != nil {
-			ca.httpErrorf(w, http.StatusBadRequest, err.Error())
+			ca.httpErrorf(w, http.StatusBadRequest, "%v", err)
 			return
 		}
 		if err := json.NewEncoder(w).Encode(o); err != nil {
@@ -412,7 +412,7 @@ func (ca *CAServer) handle(w http.ResponseWriter, r *http.Request) {
 		orderID := strings.TrimPrefix(r.URL.Path, "/new-cert/")
 		o, err := ca.storedOrder(orderID)
 		if err != nil {
-			ca.httpErrorf(w, http.StatusBadRequest, err.Error())
+			ca.httpErrorf(w, http.StatusBadRequest, "%v", err)
 			return
 		}
 		if o.Status != acme.StatusReady {
@@ -427,7 +427,7 @@ func (ca *CAServer) handle(w http.ResponseWriter, r *http.Request) {
 		b, _ := base64.RawURLEncoding.DecodeString(req.CSR)
 		csr, err := x509.ParseCertificateRequest(b)
 		if err != nil {
-			ca.httpErrorf(w, http.StatusBadRequest, err.Error())
+			ca.httpErrorf(w, http.StatusBadRequest, "%v", err)
 			return
 		}
 		// Issue the certificate.
@@ -449,7 +449,7 @@ func (ca *CAServer) handle(w http.ResponseWriter, r *http.Request) {
 		defer ca.mu.Unlock()
 		o, err := ca.storedOrder(strings.TrimPrefix(r.URL.Path, "/issued-cert/"))
 		if err != nil {
-			ca.httpErrorf(w, http.StatusBadRequest, err.Error())
+			ca.httpErrorf(w, http.StatusBadRequest, "%v", err)
 			return
 		}
 		if o.Status != acme.StatusValid {

acme/jws.go

@@ -92,7 +92,7 @@ func jwsEncodeJSON(claimset interface{}, key crypto.Signer, kid KeyID, nonce, ur
 	if err != nil {
 		return nil, err
 	}
-	phead := base64.RawURLEncoding.EncodeToString([]byte(phJSON))
+	phead := base64.RawURLEncoding.EncodeToString(phJSON)
 	var payload string
 	if val, ok := claimset.(string); ok {
 		payload = val

acme/types.go

@@ -7,6 +7,7 @@ package acme
 import (
 	"crypto"
 	"crypto/x509"
+	"encoding/json"
 	"errors"
 	"fmt"
 	"net/http"
@@ -288,7 +289,7 @@ type Directory struct {
 	// KeyChangeURL allows to perform account key rollover flow.
 	KeyChangeURL string
 
-	// Term is a URI identifying the current terms of service.
+	// Terms is a URI identifying the current terms of service.
 	Terms string
 
 	// Website is an HTTP or HTTPS URL locating a website
@@ -527,6 +528,16 @@ type Challenge struct {
 	// when this challenge was used.
 	// The type of a non-nil value is *Error.
 	Error error
+
+	// Payload is the JSON-formatted payload that the client sends
+	// to the server to indicate it is ready to respond to the challenge.
+	// When unset, it defaults to an empty JSON object: {}.
+	// For most challenges, the client must not set Payload,
+	// see https://tools.ietf.org/html/rfc8555#section-7.5.1.
+	// Payload is used only for newer challenges (such as "device-attest-01")
+	// where the client must send additional data for the server to validate
+	// the challenge.
+	Payload json.RawMessage
 }
 
 // wireChallenge is ACME JSON challenge representation.

argon2/_asm/blamka_amd64.go

@@ -0,0 +1,287 @@
+// Copyright 2024 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.
+
+package main
+
+import (
+	. "github.com/mmcloughlin/avo/build"
+	. "github.com/mmcloughlin/avo/operand"
+	. "github.com/mmcloughlin/avo/reg"
+	_ "golang.org/x/crypto/argon2"
+)
+
+//go:generate go run . -out ../blamka_amd64.s -pkg argon2
+
+func main() {
+	Package("golang.org/x/crypto/argon2")
+	ConstraintExpr("amd64,gc,!purego")
+
+	blamkaSSE4()
+	mixBlocksSSE2()
+	xorBlocksSSE2()
+	Generate()
+}
+
+func blamkaSSE4() {
+	Implement("blamkaSSE4")
+	Attributes(NOSPLIT)
+	AllocLocal(0)
+
+	Load(Param("b"), RAX)
+
+	c40 := c40_DATA()
+	c48 := c48_DATA()
+
+	MOVOU(c40, X10)
+	MOVOU(c48, X11)
+
+	BLAMKA_ROUND_0(AX, 0, X8, X9, X10, X11)
+	BLAMKA_ROUND_0(AX, 16, X8, X9, X10, X11)
+	BLAMKA_ROUND_0(AX, 32, X8, X9, X10, X11)
+	BLAMKA_ROUND_0(AX, 48, X8, X9, X10, X11)
+	BLAMKA_ROUND_0(AX, 64, X8, X9, X10, X11)
+	BLAMKA_ROUND_0(AX, 80, X8, X9, X10, X11)
+	BLAMKA_ROUND_0(AX, 96, X8, X9, X10, X11)
+	BLAMKA_ROUND_0(AX, 112, X8, X9, X10, X11)
+
+	BLAMKA_ROUND_1(AX, 0, X8, X9, X10, X11)
+	BLAMKA_ROUND_1(AX, 2, X8, X9, X10, X11)
+	BLAMKA_ROUND_1(AX, 4, X8, X9, X10, X11)
+	BLAMKA_ROUND_1(AX, 6, X8, X9, X10, X11)
+	BLAMKA_ROUND_1(AX, 8, X8, X9, X10, X11)
+	BLAMKA_ROUND_1(AX, 10, X8, X9, X10, X11)
+	BLAMKA_ROUND_1(AX, 12, X8, X9, X10, X11)
+	BLAMKA_ROUND_1(AX, 14, X8, X9, X10, X11)
+	RET()
+}
+
+func mixBlocksSSE2() {
+	Implement("mixBlocksSSE2")
+	Attributes(NOSPLIT)
+	AllocLocal(0)
+
+	Load(Param("out"), RDX)
+	Load(Param("a"), RAX)
+	Load(Param("b"), RBX)
+	Load(Param("c"), RCX)
+	MOVQ(U32(128), RDI)
+
+	Label("loop")
+	MOVOU(Mem{Base: AX}.Offset(0), X0)
+	MOVOU(Mem{Base: BX}.Offset(0), X1)
+	MOVOU(Mem{Base: CX}.Offset(0), X2)
+	PXOR(X1, X0)
+	PXOR(X2, X0)
+	MOVOU(X0, Mem{Base: DX}.Offset(0))
+	ADDQ(Imm(16), RAX)
+	ADDQ(Imm(16), RBX)
+	ADDQ(Imm(16), RCX)
+	ADDQ(Imm(16), RDX)
+	SUBQ(Imm(2), RDI)
+	JA(LabelRef("loop"))
+	RET()
+}
+
+func xorBlocksSSE2() {
+	Implement("xorBlocksSSE2")
+	Attributes(NOSPLIT)
+	AllocLocal(0)
+
+	Load(Param("out"), RDX)
+	Load(Param("a"), RAX)
+	Load(Param("b"), RBX)
+	Load(Param("c"), RCX)
+	MOVQ(U32(128), RDI)
+
+	Label("loop")
+	MOVOU(Mem{Base: AX}.Offset(0), X0)
+	MOVOU(Mem{Base: BX}.Offset(0), X1)
+	MOVOU(Mem{Base: CX}.Offset(0), X2)
+	MOVOU(Mem{Base: DX}.Offset(0), X3)
+	PXOR(X1, X0)
+	PXOR(X2, X0)
+	PXOR(X3, X0)
+	MOVOU(X0, Mem{Base: DX}.Offset(0))
+	ADDQ(Imm(16), RAX)
+	ADDQ(Imm(16), RBX)
+	ADDQ(Imm(16), RCX)
+	ADDQ(Imm(16), RDX)
+	SUBQ(Imm(2), RDI)
+	JA(LabelRef("loop"))
+	RET()
+}
+
+func SHUFFLE(v2, v3, v4, v5, v6, v7, t1, t2 VecPhysical) {
+	MOVO(v4, t1)
+	MOVO(v5, v4)
+	MOVO(t1, v5)
+	MOVO(v6, t1)
+	PUNPCKLQDQ(v6, t2)
+	PUNPCKHQDQ(v7, v6)
+	PUNPCKHQDQ(t2, v6)
+	PUNPCKLQDQ(v7, t2)
+	MOVO(t1, v7)
+	MOVO(v2, t1)
+	PUNPCKHQDQ(t2, v7)
+	PUNPCKLQDQ(v3, t2)
+	PUNPCKHQDQ(t2, v2)
+	PUNPCKLQDQ(t1, t2)
+	PUNPCKHQDQ(t2, v3)
+}
+
+func SHUFFLE_INV(v2, v3, v4, v5, v6, v7, t1, t2 VecPhysical) {
+	MOVO(v4, t1)
+	MOVO(v5, v4)
+	MOVO(t1, v5)
+	MOVO(v2, t1)
+	PUNPCKLQDQ(v2, t2)
+	PUNPCKHQDQ(v3, v2)
+	PUNPCKHQDQ(t2, v2)
+	PUNPCKLQDQ(v3, t2)
+	MOVO(t1, v3)
+	MOVO(v6, t1)
+	PUNPCKHQDQ(t2, v3)
+	PUNPCKLQDQ(v7, t2)
+	PUNPCKHQDQ(t2, v6)
+	PUNPCKLQDQ(t1, t2)
+	PUNPCKHQDQ(t2, v7)
+}
+
+func HALF_ROUND(v0, v1, v2, v3, v4, v5, v6, v7, t0, c40, c48 VecPhysical) {
+	MOVO(v0, t0)
+	PMULULQ(v2, t0)
+	PADDQ(v2, v0)
+	PADDQ(t0, v0)
+	PADDQ(t0, v0)
+	PXOR(v0, v6)
+	PSHUFD(Imm(0xB1), v6, v6)
+	MOVO(v4, t0)
+	PMULULQ(v6, t0)
+	PADDQ(v6, v4)
+	PADDQ(t0, v4)
+	PADDQ(t0, v4)
+	PXOR(v4, v2)
+	PSHUFB(c40, v2)
+	MOVO(v0, t0)
+	PMULULQ(v2, t0)
+	PADDQ(v2, v0)
+	PADDQ(t0, v0)
+	PADDQ(t0, v0)
+	PXOR(v0, v6)
+	PSHUFB(c48, v6)
+	MOVO(v4, t0)
+	PMULULQ(v6, t0)
+	PADDQ(v6, v4)
+	PADDQ(t0, v4)
+	PADDQ(t0, v4)
+	PXOR(v4, v2)
+	MOVO(v2, t0)
+	PADDQ(v2, t0)
+	PSRLQ(Imm(63), v2)
+	PXOR(t0, v2)
+	MOVO(v1, t0)
+	PMULULQ(v3, t0)
+	PADDQ(v3, v1)
+	PADDQ(t0, v1)
+	PADDQ(t0, v1)
+	PXOR(v1, v7)
+	PSHUFD(Imm(0xB1), v7, v7)
+	MOVO(v5, t0)
+	PMULULQ(v7, t0)
+	PADDQ(v7, v5)
+	PADDQ(t0, v5)
+	PADDQ(t0, v5)
+	PXOR(v5, v3)
+	PSHUFB(c40, v3)
+	MOVO(v1, t0)
+	PMULULQ(v3, t0)
+	PADDQ(v3, v1)
+	PADDQ(t0, v1)
+	PADDQ(t0, v1)
+	PXOR(v1, v7)
+	PSHUFB(c48, v7)
+	MOVO(v5, t0)
+	PMULULQ(v7, t0)
+	PADDQ(v7, v5)
+	PADDQ(t0, v5)
+	PADDQ(t0, v5)
+	PXOR(v5, v3)
+	MOVO(v3, t0)
+	PADDQ(v3, t0)
+	PSRLQ(Imm(63), v3)
+	PXOR(t0, v3)
+}
+
+func LOAD_MSG_0(block GPPhysical, off int) {
+	var registers = []VecPhysical{X0, X1, X2, X3, X4, X5, X6, X7}
+	for i, r := range registers {
+		MOVOU(Mem{Base: block}.Offset(8*(off+(i*2))), r)
+	}
+}
+
+func STORE_MSG_0(block GPPhysical, off int) {
+	var registers = []VecPhysical{X0, X1, X2, X3, X4, X5, X6, X7}
+	for i, r := range registers {
+		MOVOU(r, Mem{Base: block}.Offset(8*(off+(i*2))))
+	}
+}
+
+func LOAD_MSG_1(block GPPhysical, off int) {
+	var registers = []VecPhysical{X0, X1, X2, X3, X4, X5, X6, X7}
+	for i, r := range registers {
+		MOVOU(Mem{Base: block}.Offset(8*off+i*16*8), r)
+	}
+}
+
+func STORE_MSG_1(block GPPhysical, off int) {
+	var registers = []VecPhysical{X0, X1, X2, X3, X4, X5, X6, X7}
+	for i, r := range registers {
+		MOVOU(r, Mem{Base: block}.Offset(8*off+i*16*8))
+	}
+}
+
+func BLAMKA_ROUND_0(block GPPhysical, off int, t0, t1, c40, c48 VecPhysical) {
+	LOAD_MSG_0(block, off)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, t0, c40, c48)
+	SHUFFLE(X2, X3, X4, X5, X6, X7, t0, t1)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, t0, c40, c48)
+	SHUFFLE_INV(X2, X3, X4, X5, X6, X7, t0, t1)
+	STORE_MSG_0(block, off)
+}
+
+func BLAMKA_ROUND_1(block GPPhysical, off int, t0, t1, c40, c48 VecPhysical) {
+	LOAD_MSG_1(block, off)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, t0, c40, c48)
+	SHUFFLE(X2, X3, X4, X5, X6, X7, t0, t1)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, t0, c40, c48)
+	SHUFFLE_INV(X2, X3, X4, X5, X6, X7, t0, t1)
+	STORE_MSG_1(block, off)
+}
+
+// ##------------------DATA SECTION-------------------##
+
+var c40_DATA_ptr, c48_DATA_ptr *Mem
+
+func c40_DATA() Mem {
+	if c40_DATA_ptr != nil {
+		return *c40_DATA_ptr
+	}
+
+	c40_DATA := GLOBL("·c40", NOPTR|RODATA)
+	c40_DATA_ptr = &c40_DATA
+	DATA(0x00, U64(0x0201000706050403))
+	DATA(0x08, U64(0x0a09080f0e0d0c0b))
+	return c40_DATA
+}
+func c48_DATA() Mem {
+	if c48_DATA_ptr != nil {
+		return *c48_DATA_ptr
+	}
+
+	c48_DATA := GLOBL("·c48", NOPTR|RODATA)
+	c48_DATA_ptr = &c48_DATA
+	DATA(0x00, U64(0x0100070605040302))
+	DATA(0x08, U64(0x09080f0e0d0c0b0a))
+	return c48_DATA
+}

argon2/_asm/go.mod

@@ -0,0 +1,15 @@
+module argon2/_asm
+
+go 1.23
+
+require (
+	github.com/mmcloughlin/avo v0.6.0
+	golang.org/x/crypto v0.33.0
+)
+
+require (
+	golang.org/x/mod v0.20.0 // indirect
+	golang.org/x/sync v0.8.0 // indirect
+	golang.org/x/sys v0.30.0 // indirect
+	golang.org/x/tools v0.24.0 // indirect
+)

argon2/_asm/go.sum

@@ -0,0 +1,12 @@
+github.com/mmcloughlin/avo v0.6.0 h1:QH6FU8SKoTLaVs80GA8TJuLNkUYl4VokHKlPhVDg4YY=
+github.com/mmcloughlin/avo v0.6.0/go.mod h1:8CoAGaCSYXtCPR+8y18Y9aB/kxb8JSS6FRI7mSkvD+8=
+golang.org/x/crypto v0.33.0 h1:IOBPskki6Lysi0lo9qQvbxiQ+FvsCC/YWOecCHAixus=
+golang.org/x/crypto v0.33.0/go.mod h1:bVdXmD7IV/4GdElGPozy6U7lWdRXA4qyRVGJV57uQ5M=
+golang.org/x/mod v0.20.0 h1:utOm6MM3R3dnawAiJgn0y+xvuYRsm1RKM/4giyfDgV0=
+golang.org/x/mod v0.20.0/go.mod h1:hTbmBsO62+eylJbnUtE2MGJUyE7QWk4xUqPFrRgJ+7c=
+golang.org/x/sync v0.8.0 h1:3NFvSEYkUoMifnESzZl15y791HH1qU2xm6eCJU5ZPXQ=
+golang.org/x/sync v0.8.0/go.mod h1:Czt+wKu1gCyEFDUtn0jG5QVvpJ6rzVqr5aXyt9drQfk=
+golang.org/x/sys v0.30.0 h1:QjkSwP/36a20jFYWkSue1YwXzLmsV5Gfq7Eiy72C1uc=
+golang.org/x/sys v0.30.0/go.mod h1:/VUhepiaJMQUp4+oa/7Zr1D23ma6VTLIYjOOTFZPUcA=
+golang.org/x/tools v0.24.0 h1:J1shsA93PJUEVaUSaay7UXAyE8aimq3GW0pjlolpa24=
+golang.org/x/tools v0.24.0/go.mod h1:YhNqVBIfWHdzvTLs0d8LCuMhkKUgSUKldakyV7W/WDQ=

blake2b/_asm/AVX2/go.mod

@@ -0,0 +1,16 @@
+module blake2b/_asm/AVX2
+
+go 1.23
+
+require (
+	github.com/mmcloughlin/avo v0.6.0
+	golang.org/x/crypto v0.33.0
+
+)
+
+require (
+	golang.org/x/mod v0.20.0 // indirect
+	golang.org/x/sync v0.8.0 // indirect
+	golang.org/x/sys v0.30.0 // indirect
+	golang.org/x/tools v0.24.0 // indirect
+)

blake2b/_asm/AVX2/go.sum

@@ -0,0 +1,12 @@
+github.com/mmcloughlin/avo v0.6.0 h1:QH6FU8SKoTLaVs80GA8TJuLNkUYl4VokHKlPhVDg4YY=
+github.com/mmcloughlin/avo v0.6.0/go.mod h1:8CoAGaCSYXtCPR+8y18Y9aB/kxb8JSS6FRI7mSkvD+8=
+golang.org/x/crypto v0.33.0 h1:IOBPskki6Lysi0lo9qQvbxiQ+FvsCC/YWOecCHAixus=
+golang.org/x/crypto v0.33.0/go.mod h1:bVdXmD7IV/4GdElGPozy6U7lWdRXA4qyRVGJV57uQ5M=
+golang.org/x/mod v0.20.0 h1:utOm6MM3R3dnawAiJgn0y+xvuYRsm1RKM/4giyfDgV0=
+golang.org/x/mod v0.20.0/go.mod h1:hTbmBsO62+eylJbnUtE2MGJUyE7QWk4xUqPFrRgJ+7c=
+golang.org/x/sync v0.8.0 h1:3NFvSEYkUoMifnESzZl15y791HH1qU2xm6eCJU5ZPXQ=
+golang.org/x/sync v0.8.0/go.mod h1:Czt+wKu1gCyEFDUtn0jG5QVvpJ6rzVqr5aXyt9drQfk=
+golang.org/x/sys v0.30.0 h1:QjkSwP/36a20jFYWkSue1YwXzLmsV5Gfq7Eiy72C1uc=
+golang.org/x/sys v0.30.0/go.mod h1:/VUhepiaJMQUp4+oa/7Zr1D23ma6VTLIYjOOTFZPUcA=
+golang.org/x/tools v0.24.0 h1:J1shsA93PJUEVaUSaay7UXAyE8aimq3GW0pjlolpa24=
+golang.org/x/tools v0.24.0/go.mod h1:YhNqVBIfWHdzvTLs0d8LCuMhkKUgSUKldakyV7W/WDQ=

blake2b/_asm/standard/blake2b_amd64_asm.go

@@ -0,0 +1,361 @@
+// Copyright 2024 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.
+
+package main
+
+import (
+	. "github.com/mmcloughlin/avo/build"
+	. "github.com/mmcloughlin/avo/operand"
+	. "github.com/mmcloughlin/avo/reg"
+	_ "golang.org/x/crypto/blake2b"
+)
+
+//go:generate go run . -out ../../blake2b_amd64.s -pkg blake2b
+
+const ThatPeskyUnicodeDot = "\u00b7"
+
+var iv0_DATA_ptr, iv1_DATA_ptr, iv2_DATA_ptr, iv3_DATA_ptr, c40_DATA_ptr, c48_DATA_ptr *Mem
+
+func main() {
+	Package("golang.org/x/crypto/blake2b")
+	ConstraintExpr("amd64,gc,!purego")
+	hashBlocksSSE4()
+	Generate()
+}
+
+func SHUFFLE(v2, v3, v4, v5, v6, v7, t1, t2 VecPhysical) {
+	MOVO(v4, t1)
+	MOVO(v5, v4)
+	MOVO(t1, v5)
+	MOVO(v6, t1)
+	PUNPCKLQDQ(v6, t2)
+	PUNPCKHQDQ(v7, v6)
+	PUNPCKHQDQ(t2, v6)
+	PUNPCKLQDQ(v7, t2)
+	MOVO(t1, v7)
+	MOVO(v2, t1)
+	PUNPCKHQDQ(t2, v7)
+	PUNPCKLQDQ(v3, t2)
+	PUNPCKHQDQ(t2, v2)
+	PUNPCKLQDQ(t1, t2)
+	PUNPCKHQDQ(t2, v3)
+}
+
+func SHUFFLE_INV(v2, v3, v4, v5, v6, v7, t1, t2 VecPhysical) {
+	MOVO(v4, t1)
+	MOVO(v5, v4)
+	MOVO(t1, v5)
+	MOVO(v2, t1)
+	PUNPCKLQDQ(v2, t2)
+	PUNPCKHQDQ(v3, v2)
+	PUNPCKHQDQ(t2, v2)
+	PUNPCKLQDQ(v3, t2)
+	MOVO(t1, v3)
+	MOVO(v6, t1)
+	PUNPCKHQDQ(t2, v3)
+	PUNPCKLQDQ(v7, t2)
+	PUNPCKHQDQ(t2, v6)
+	PUNPCKLQDQ(t1, t2)
+	PUNPCKHQDQ(t2, v7)
+}
+
+func HALF_ROUND(v0, v1, v2, v3, v4, v5, v6, v7 VecPhysical, m0, m1, m2, m3 Op, t0, c40, c48 VecPhysical) {
+	PADDQ(m0, v0)
+	PADDQ(m1, v1)
+	PADDQ(v2, v0)
+	PADDQ(v3, v1)
+	PXOR(v0, v6)
+	PXOR(v1, v7)
+	PSHUFD(Imm(0xB1), v6, v6)
+	PSHUFD(Imm(0xB1), v7, v7)
+	PADDQ(v6, v4)
+	PADDQ(v7, v5)
+	PXOR(v4, v2)
+	PXOR(v5, v3)
+	PSHUFB(c40, v2)
+	PSHUFB(c40, v3)
+	PADDQ(m2, v0)
+	PADDQ(m3, v1)
+	PADDQ(v2, v0)
+	PADDQ(v3, v1)
+	PXOR(v0, v6)
+	PXOR(v1, v7)
+	PSHUFB(c48, v6)
+	PSHUFB(c48, v7)
+	PADDQ(v6, v4)
+	PADDQ(v7, v5)
+	PXOR(v4, v2)
+	PXOR(v5, v3)
+	MOVOU(v2, t0)
+	PADDQ(v2, t0)
+	PSRLQ(Imm(63), v2)
+	PXOR(t0, v2)
+	MOVOU(v3, t0)
+	PADDQ(v3, t0)
+	PSRLQ(Imm(63), v3)
+	PXOR(t0, v3)
+}
+
+func LOAD_MSG(m0, m1, m2, m3 VecPhysical, src GPPhysical, i0, i1, i2, i3, i4, i5, i6, i7 int) {
+	MOVQ(Mem{Base: src}.Offset(i0*8), m0)
+	PINSRQ(Imm(1), Mem{Base: src}.Offset(i1*8), m0)
+	MOVQ(Mem{Base: src}.Offset(i2*8), m1)
+	PINSRQ(Imm(1), Mem{Base: src}.Offset(i3*8), m1)
+	MOVQ(Mem{Base: src}.Offset(i4*8), m2)
+	PINSRQ(Imm(1), Mem{Base: src}.Offset(i5*8), m2)
+	MOVQ(Mem{Base: src}.Offset(i6*8), m3)
+	PINSRQ(Imm(1), Mem{Base: src}.Offset(i7*8), m3)
+}
+
+func hashBlocksSSE4() {
+	Implement("hashBlocksSSE4")
+	Attributes(4)
+	AllocLocal(288) // frame size = 272 + 16 byte alignment
+
+	Load(Param("h"), RAX)
+	Load(Param("c"), RBX)
+	Load(Param("flag"), RCX)
+	Load(Param("blocks").Base(), RSI)
+	Load(Param("blocks").Len(), RDI)
+
+	MOVQ(RSP, R10)
+	ADDQ(Imm(15), R10)
+	ANDQ(I32(-16), R10)
+
+	iv3 := iv3_DATA()
+	MOVOU(iv3, X0)
+	MOVO(X0, Mem{Base: R10}.Offset(0))
+	XORQ(RCX, Mem{Base: R10}.Offset(0)) // 0(R10) = ·iv3 ^ (CX || 0)
+
+	c40 := c40_DATA()
+	c48 := c48_DATA()
+	MOVOU(c40, X13)
+	MOVOU(c48, X14)
+
+	MOVOU(Mem{Base: AX}.Offset(0), X12)
+	MOVOU(Mem{Base: AX}.Offset(16), X15)
+
+	MOVQ(Mem{Base: BX}.Offset(0), R8)
+	MOVQ(Mem{Base: BX}.Offset(8), R9)
+
+	Label("loop")
+	ADDQ(Imm(128), R8)
+	CMPQ(R8, Imm(128))
+	JGE(LabelRef("noinc"))
+	INCQ(R9)
+
+	Label("noinc")
+	MOVQ(R8, X8)
+	PINSRQ(Imm(1), R9, X8)
+
+	iv0 := iv0_DATA()
+	iv1 := iv1_DATA()
+	iv2 := iv2_DATA()
+
+	MOVO(X12, X0)
+	MOVO(X15, X1)
+	MOVOU(Mem{Base: AX}.Offset(32), X2)
+	MOVOU(Mem{Base: AX}.Offset(48), X3)
+	MOVOU(iv0, X4)
+	MOVOU(iv1, X5)
+	MOVOU(iv2, X6)
+
+	PXOR(X8, X6)
+	MOVO(Mem{Base: R10}.Offset(0), X7)
+
+	LOAD_MSG(X8, X9, X10, X11, SI, 0, 2, 4, 6, 1, 3, 5, 7)
+	MOVO(X8, Mem{Base: R10}.Offset(16))
+	MOVO(X9, Mem{Base: R10}.Offset(32))
+	MOVO(X10, Mem{Base: R10}.Offset(48))
+	MOVO(X11, Mem{Base: R10}.Offset(64))
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE(X2, X3, X4, X5, X6, X7, X8, X9)
+	LOAD_MSG(X8, X9, X10, X11, SI, 8, 10, 12, 14, 9, 11, 13, 15)
+	MOVO(X8, Mem{Base: R10}.Offset(80))
+	MOVO(X9, Mem{Base: R10}.Offset(96))
+	MOVO(X10, Mem{Base: R10}.Offset(112))
+	MOVO(X11, Mem{Base: R10}.Offset(128))
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE_INV(X2, X3, X4, X5, X6, X7, X8, X9)
+
+	LOAD_MSG(X8, X9, X10, X11, SI, 14, 4, 9, 13, 10, 8, 15, 6)
+	MOVO(X8, Mem{Base: R10}.Offset(144))
+	MOVO(X9, Mem{Base: R10}.Offset(160))
+	MOVO(X10, Mem{Base: R10}.Offset(176))
+	MOVO(X11, Mem{Base: R10}.Offset(192))
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE(X2, X3, X4, X5, X6, X7, X8, X9)
+	LOAD_MSG(X8, X9, X10, X11, SI, 1, 0, 11, 5, 12, 2, 7, 3)
+	MOVO(X8, Mem{Base: R10}.Offset(208))
+	MOVO(X9, Mem{Base: R10}.Offset(224))
+	MOVO(X10, Mem{Base: R10}.Offset(240))
+	MOVO(X11, Mem{Base: R10}.Offset(256))
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE_INV(X2, X3, X4, X5, X6, X7, X8, X9)
+
+	LOAD_MSG(X8, X9, X10, X11, SI, 11, 12, 5, 15, 8, 0, 2, 13)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE(X2, X3, X4, X5, X6, X7, X8, X9)
+	LOAD_MSG(X8, X9, X10, X11, SI, 10, 3, 7, 9, 14, 6, 1, 4)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE_INV(X2, X3, X4, X5, X6, X7, X8, X9)
+
+	LOAD_MSG(X8, X9, X10, X11, SI, 7, 3, 13, 11, 9, 1, 12, 14)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE(X2, X3, X4, X5, X6, X7, X8, X9)
+	LOAD_MSG(X8, X9, X10, X11, SI, 2, 5, 4, 15, 6, 10, 0, 8)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE_INV(X2, X3, X4, X5, X6, X7, X8, X9)
+
+	LOAD_MSG(X8, X9, X10, X11, SI, 9, 5, 2, 10, 0, 7, 4, 15)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE(X2, X3, X4, X5, X6, X7, X8, X9)
+	LOAD_MSG(X8, X9, X10, X11, SI, 14, 11, 6, 3, 1, 12, 8, 13)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE_INV(X2, X3, X4, X5, X6, X7, X8, X9)
+
+	LOAD_MSG(X8, X9, X10, X11, SI, 2, 6, 0, 8, 12, 10, 11, 3)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE(X2, X3, X4, X5, X6, X7, X8, X9)
+	LOAD_MSG(X8, X9, X10, X11, SI, 4, 7, 15, 1, 13, 5, 14, 9)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE_INV(X2, X3, X4, X5, X6, X7, X8, X9)
+
+	LOAD_MSG(X8, X9, X10, X11, SI, 12, 1, 14, 4, 5, 15, 13, 10)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE(X2, X3, X4, X5, X6, X7, X8, X9)
+	LOAD_MSG(X8, X9, X10, X11, SI, 0, 6, 9, 8, 7, 3, 2, 11)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE_INV(X2, X3, X4, X5, X6, X7, X8, X9)
+
+	LOAD_MSG(X8, X9, X10, X11, SI, 13, 7, 12, 3, 11, 14, 1, 9)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE(X2, X3, X4, X5, X6, X7, X8, X9)
+	LOAD_MSG(X8, X9, X10, X11, SI, 5, 15, 8, 2, 0, 4, 6, 10)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE_INV(X2, X3, X4, X5, X6, X7, X8, X9)
+
+	LOAD_MSG(X8, X9, X10, X11, SI, 6, 14, 11, 0, 15, 9, 3, 8)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE(X2, X3, X4, X5, X6, X7, X8, X9)
+	LOAD_MSG(X8, X9, X10, X11, SI, 12, 13, 1, 10, 2, 7, 4, 5)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE_INV(X2, X3, X4, X5, X6, X7, X8, X9)
+
+	LOAD_MSG(X8, X9, X10, X11, SI, 10, 8, 7, 1, 2, 4, 6, 5)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE(X2, X3, X4, X5, X6, X7, X8, X9)
+	LOAD_MSG(X8, X9, X10, X11, SI, 15, 9, 3, 13, 11, 14, 12, 0)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X11, X13, X14)
+	SHUFFLE_INV(X2, X3, X4, X5, X6, X7, X8, X9)
+
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, Mem{Base: R10}.Offset(16), Mem{Base: R10}.Offset(32), Mem{Base: R10}.Offset(48), Mem{Base: R10}.Offset(64), X11, X13, X14)
+	SHUFFLE(X2, X3, X4, X5, X6, X7, X8, X9)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, Mem{Base: R10}.Offset(80), Mem{Base: R10}.Offset(96), Mem{Base: R10}.Offset(112), Mem{Base: R10}.Offset(128), X11, X13, X14)
+	SHUFFLE_INV(X2, X3, X4, X5, X6, X7, X8, X9)
+
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, Mem{Base: R10}.Offset(144), Mem{Base: R10}.Offset(160), Mem{Base: R10}.Offset(176), Mem{Base: R10}.Offset(192), X11, X13, X14)
+	SHUFFLE(X2, X3, X4, X5, X6, X7, X8, X9)
+	HALF_ROUND(X0, X1, X2, X3, X4, X5, X6, X7, Mem{Base: R10}.Offset(208), Mem{Base: R10}.Offset(224), Mem{Base: R10}.Offset(240), Mem{Base: R10}.Offset(256), X11, X13, X14)
+	SHUFFLE_INV(X2, X3, X4, X5, X6, X7, X8, X9)
+
+	MOVOU(Mem{Base: AX}.Offset(32), X10)
+	MOVOU(Mem{Base: AX}.Offset(48), X11)
+	PXOR(X0, X12)
+	PXOR(X1, X15)
+	PXOR(X2, X10)
+	PXOR(X3, X11)
+	PXOR(X4, X12)
+	PXOR(X5, X15)
+	PXOR(X6, X10)
+	PXOR(X7, X11)
+	MOVOU(X10, Mem{Base: AX}.Offset(32))
+	MOVOU(X11, Mem{Base: AX}.Offset(48))
+
+	LEAQ(Mem{Base: SI}.Offset(128), RSI)
+	SUBQ(Imm(128), RDI)
+	JNE(LabelRef("loop"))
+
+	MOVOU(X12, Mem{Base: AX}.Offset(0))
+	MOVOU(X15, Mem{Base: AX}.Offset(16))
+
+	MOVQ(R8, Mem{Base: BX}.Offset(0))
+	MOVQ(R9, Mem{Base: BX}.Offset(8))
+
+	RET()
+}
+
+// #~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~DATA SECTION~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~##
+
+func iv0_DATA() Mem {
+	if iv0_DATA_ptr != nil {
+		return *iv0_DATA_ptr
+	}
+
+	iv0 := GLOBL(ThatPeskyUnicodeDot+"iv0", NOPTR|RODATA)
+	iv0_DATA_ptr = &iv0
+	DATA(0x00, U64(0x6a09e667f3bcc908))
+	DATA(0x08, U64(0xbb67ae8584caa73b))
+	return iv0
+}
+
+func iv1_DATA() Mem {
+	if iv1_DATA_ptr != nil {
+		return *iv1_DATA_ptr
+	}
+
+	iv1 := GLOBL(ThatPeskyUnicodeDot+"iv1", NOPTR|RODATA)
+	iv1_DATA_ptr = &iv1
+	DATA(0x00, U64(0x3c6ef372fe94f82b))
+	DATA(0x08, U64(0xa54ff53a5f1d36f1))
+	return iv1
+}
+
+func iv2_DATA() Mem {
+	if iv2_DATA_ptr != nil {
+		return *iv2_DATA_ptr
+	}
+
+	iv2 := GLOBL(ThatPeskyUnicodeDot+"iv2", NOPTR|RODATA)
+	iv2_DATA_ptr = &iv2
+	DATA(0x00, U64(0x510e527fade682d1))
+	DATA(0x08, U64(0x9b05688c2b3e6c1f))
+	return iv2
+}
+
+func iv3_DATA() Mem {
+	if iv3_DATA_ptr != nil {
+		return *iv3_DATA_ptr
+	}
+
+	iv3 := GLOBL(ThatPeskyUnicodeDot+"iv3", NOPTR|RODATA)
+	iv3_DATA_ptr = &iv3
+	DATA(0x00, U64(0x1f83d9abfb41bd6b))
+	DATA(0x08, U64(0x5be0cd19137e2179))
+	return iv3
+}
+
+func c40_DATA() Mem {
+	if c40_DATA_ptr != nil {
+		return *c40_DATA_ptr
+	}
+
+	c40 := GLOBL(ThatPeskyUnicodeDot+"c40", NOPTR|RODATA)
+	c40_DATA_ptr = &c40
+	DATA(0x00, U64(0x0201000706050403))
+	DATA(0x08, U64(0x0a09080f0e0d0c0b))
+	return c40
+}
+
+func c48_DATA() Mem {
+	if c48_DATA_ptr != nil {
+		return *c48_DATA_ptr
+	}
+
+	c48 := GLOBL(ThatPeskyUnicodeDot+"c48", NOPTR|RODATA)
+	c48_DATA_ptr = &c48
+	DATA(0x00, U64(0x0100070605040302))
+	DATA(0x08, U64(0x09080f0e0d0c0b0a))
+	return c48
+}

blake2b/_asm/standard/go.mod

@@ -0,0 +1,15 @@
+module blake2b/_asm
+
+go 1.23
+
+require (
+	github.com/mmcloughlin/avo v0.6.0
+	golang.org/x/crypto v0.33.0
+)
+
+require (
+	golang.org/x/mod v0.20.0 // indirect
+	golang.org/x/sync v0.8.0 // indirect
+	golang.org/x/sys v0.30.0 // indirect
+	golang.org/x/tools v0.24.0 // indirect
+)

blake2b/_asm/standard/go.sum

@@ -0,0 +1,12 @@
+github.com/mmcloughlin/avo v0.6.0 h1:QH6FU8SKoTLaVs80GA8TJuLNkUYl4VokHKlPhVDg4YY=
+github.com/mmcloughlin/avo v0.6.0/go.mod h1:8CoAGaCSYXtCPR+8y18Y9aB/kxb8JSS6FRI7mSkvD+8=
+golang.org/x/crypto v0.33.0 h1:IOBPskki6Lysi0lo9qQvbxiQ+FvsCC/YWOecCHAixus=
+golang.org/x/crypto v0.33.0/go.mod h1:bVdXmD7IV/4GdElGPozy6U7lWdRXA4qyRVGJV57uQ5M=
+golang.org/x/mod v0.20.0 h1:utOm6MM3R3dnawAiJgn0y+xvuYRsm1RKM/4giyfDgV0=
+golang.org/x/mod v0.20.0/go.mod h1:hTbmBsO62+eylJbnUtE2MGJUyE7QWk4xUqPFrRgJ+7c=
+golang.org/x/sync v0.8.0 h1:3NFvSEYkUoMifnESzZl15y791HH1qU2xm6eCJU5ZPXQ=
+golang.org/x/sync v0.8.0/go.mod h1:Czt+wKu1gCyEFDUtn0jG5QVvpJ6rzVqr5aXyt9drQfk=
+golang.org/x/sys v0.30.0 h1:QjkSwP/36a20jFYWkSue1YwXzLmsV5Gfq7Eiy72C1uc=
+golang.org/x/sys v0.30.0/go.mod h1:/VUhepiaJMQUp4+oa/7Zr1D23ma6VTLIYjOOTFZPUcA=
+golang.org/x/tools v0.24.0 h1:J1shsA93PJUEVaUSaay7UXAyE8aimq3GW0pjlolpa24=
+golang.org/x/tools v0.24.0/go.mod h1:YhNqVBIfWHdzvTLs0d8LCuMhkKUgSUKldakyV7W/WDQ=

blake2s/_asm/blake2s_amd64_asm.go

@@ -0,0 +1,525 @@
+// Copyright 2024 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.
+
+package main
+
+import (
+	. "github.com/mmcloughlin/avo/build"
+	"github.com/mmcloughlin/avo/ir"
+	. "github.com/mmcloughlin/avo/operand"
+	. "github.com/mmcloughlin/avo/reg"
+	_ "golang.org/x/crypto/blake2s"
+)
+
+//go:generate go run . -out ../blake2s_amd64.s -pkg blake2s
+
+func main() {
+	Package("golang.org/x/crypto/blake2s")
+	ConstraintExpr("amd64,gc,!purego")
+	hashBlocksSSE2()
+	hashBlocksSSSE3()
+	hashBlocksSSE4()
+	Generate()
+}
+
+func ROTL_SSE2(n uint64, t, v VecPhysical) {
+	MOVO(v, t)
+	PSLLL(Imm(n), t)
+	PSRLL(Imm(32-n), v)
+	PXOR(t, v)
+}
+
+func ROTL_SSSE3(c, v VecPhysical) {
+	PSHUFB(c, v)
+}
+
+func ROUND_SSE2(v0, v1, v2, v3 VecPhysical, m0, m1, m2, m3 Mem, t VecPhysical) {
+	PADDL(m0, v0)
+	PADDL(v1, v0)
+	PXOR(v0, v3)
+	ROTL_SSE2(16, t, v3)
+	PADDL(v3, v2)
+	PXOR(v2, v1)
+	ROTL_SSE2(20, t, v1)
+	PADDL(m1, v0)
+	PADDL(v1, v0)
+	PXOR(v0, v3)
+	ROTL_SSE2(24, t, v3)
+	PADDL(v3, v2)
+	PXOR(v2, v1)
+	ROTL_SSE2(25, t, v1)
+	PSHUFL(Imm(0x39), v1, v1)
+	PSHUFL(Imm(0x4E), v2, v2)
+	PSHUFL(Imm(0x93), v3, v3)
+	PADDL(m2, v0)
+	PADDL(v1, v0)
+	PXOR(v0, v3)
+	ROTL_SSE2(16, t, v3)
+	PADDL(v3, v2)
+	PXOR(v2, v1)
+	ROTL_SSE2(20, t, v1)
+	PADDL(m3, v0)
+	PADDL(v1, v0)
+	PXOR(v0, v3)
+	ROTL_SSE2(24, t, v3)
+	PADDL(v3, v2)
+	PXOR(v2, v1)
+	ROTL_SSE2(25, t, v1)
+	PSHUFL(Imm(0x39), v3, v3)
+	PSHUFL(Imm(0x4E), v2, v2)
+	PSHUFL(Imm(0x93), v1, v1)
+}
+
+func ROUND_SSSE3(v0, v1, v2, v3 VecPhysical, m0, m1, m2, m3 Op, t, c16, c8 VecPhysical) {
+	PADDL(m0, v0)
+	PADDL(v1, v0)
+	PXOR(v0, v3)
+	ROTL_SSSE3(c16, v3)
+	PADDL(v3, v2)
+	PXOR(v2, v1)
+	ROTL_SSE2(20, t, v1)
+	PADDL(m1, v0)
+	PADDL(v1, v0)
+	PXOR(v0, v3)
+	ROTL_SSSE3(c8, v3)
+	PADDL(v3, v2)
+	PXOR(v2, v1)
+	ROTL_SSE2(25, t, v1)
+	PSHUFL(Imm(0x39), v1, v1)
+	PSHUFL(Imm(0x4E), v2, v2)
+	PSHUFL(Imm(0x93), v3, v3)
+	PADDL(m2, v0)
+	PADDL(v1, v0)
+	PXOR(v0, v3)
+	ROTL_SSSE3(c16, v3)
+	PADDL(v3, v2)
+	PXOR(v2, v1)
+	ROTL_SSE2(20, t, v1)
+	PADDL(m3, v0)
+	PADDL(v1, v0)
+	PXOR(v0, v3)
+	ROTL_SSSE3(c8, v3)
+	PADDL(v3, v2)
+	PXOR(v2, v1)
+	ROTL_SSE2(25, t, v1)
+	PSHUFL(Imm(0x39), v3, v3)
+	PSHUFL(Imm(0x4E), v2, v2)
+	PSHUFL(Imm(0x93), v1, v1)
+}
+
+func LOAD_MSG_SSE4(m0, m1, m2, m3 VecPhysical, src GPPhysical, i0, i1, i2, i3, i4, i5, i6, i7, i8, i9, i10, i11, i12, i13, i14, i15 int) {
+	// Hack to get Avo to emit a MOVL instruction with a VecPhysical as the destination
+	Instruction(&ir.Instruction{Opcode: "MOVL", Operands: []Op{Mem{Base: src}.Offset(i0 * 4), m0}})
+	PINSRD(Imm(1), Mem{Base: src}.Offset(i1*4), m0)
+	PINSRD(Imm(2), Mem{Base: src}.Offset(i2*4), m0)
+	PINSRD(Imm(3), Mem{Base: src}.Offset(i3*4), m0)
+	Instruction(&ir.Instruction{Opcode: "MOVL", Operands: []Op{Mem{Base: src}.Offset(i4 * 4), m1}})
+	PINSRD(Imm(1), Mem{Base: src}.Offset(i5*4), m1)
+	PINSRD(Imm(2), Mem{Base: src}.Offset(i6*4), m1)
+	PINSRD(Imm(3), Mem{Base: src}.Offset(i7*4), m1)
+	Instruction(&ir.Instruction{Opcode: "MOVL", Operands: []Op{Mem{Base: src}.Offset(i8 * 4), m2}})
+	PINSRD(Imm(1), Mem{Base: src}.Offset(i9*4), m2)
+	PINSRD(Imm(2), Mem{Base: src}.Offset(i10*4), m2)
+	PINSRD(Imm(3), Mem{Base: src}.Offset(i11*4), m2)
+	Instruction(&ir.Instruction{Opcode: "MOVL", Operands: []Op{Mem{Base: src}.Offset(i12 * 4), m3}})
+	PINSRD(Imm(1), Mem{Base: src}.Offset(i13*4), m3)
+	PINSRD(Imm(2), Mem{Base: src}.Offset(i14*4), m3)
+	PINSRD(Imm(3), Mem{Base: src}.Offset(i15*4), m3)
+}
+
+func PRECOMPUTE_MSG(dst GPPhysical, off int, src, R8, R9, R10, R11, R12, R13, R14, R15 GPPhysical) {
+	MOVQ(Mem{Base: src}.Offset(0*4), R8)
+	MOVQ(Mem{Base: src}.Offset(2*4), R9)
+	MOVQ(Mem{Base: src}.Offset(4*4), R10)
+	MOVQ(Mem{Base: src}.Offset(6*4), R11)
+	MOVQ(Mem{Base: src}.Offset(8*4), R12)
+	MOVQ(Mem{Base: src}.Offset(10*4), R13)
+	MOVQ(Mem{Base: src}.Offset(12*4), R14)
+	MOVQ(Mem{Base: src}.Offset(14*4), R15)
+
+	MOVL(R8L, Mem{Base: dst}.Offset(0*4+off+0))
+	MOVL(R8L, Mem{Base: dst}.Offset(9*4+off+64))
+	MOVL(R8L, Mem{Base: dst}.Offset(5*4+off+128))
+	MOVL(R8L, Mem{Base: dst}.Offset(14*4+off+192))
+	MOVL(R8L, Mem{Base: dst}.Offset(4*4+off+256))
+	MOVL(R8L, Mem{Base: dst}.Offset(2*4+off+320))
+	MOVL(R8L, Mem{Base: dst}.Offset(8*4+off+384))
+	MOVL(R8L, Mem{Base: dst}.Offset(12*4+off+448))
+	MOVL(R8L, Mem{Base: dst}.Offset(3*4+off+512))
+	MOVL(R8L, Mem{Base: dst}.Offset(15*4+off+576))
+	SHRQ(Imm(32), R8)
+	MOVL(R8L, Mem{Base: dst}.Offset(4*4+off+0))
+	MOVL(R8L, Mem{Base: dst}.Offset(8*4+off+64))
+	MOVL(R8L, Mem{Base: dst}.Offset(14*4+off+128))
+	MOVL(R8L, Mem{Base: dst}.Offset(5*4+off+192))
+	MOVL(R8L, Mem{Base: dst}.Offset(12*4+off+256))
+	MOVL(R8L, Mem{Base: dst}.Offset(11*4+off+320))
+	MOVL(R8L, Mem{Base: dst}.Offset(1*4+off+384))
+	MOVL(R8L, Mem{Base: dst}.Offset(6*4+off+448))
+	MOVL(R8L, Mem{Base: dst}.Offset(10*4+off+512))
+	MOVL(R8L, Mem{Base: dst}.Offset(3*4+off+576))
+
+	MOVL(R9L, Mem{Base: dst}.Offset(1*4+off+0))
+	MOVL(R9L, Mem{Base: dst}.Offset(13*4+off+64))
+	MOVL(R9L, Mem{Base: dst}.Offset(6*4+off+128))
+	MOVL(R9L, Mem{Base: dst}.Offset(8*4+off+192))
+	MOVL(R9L, Mem{Base: dst}.Offset(2*4+off+256))
+	MOVL(R9L, Mem{Base: dst}.Offset(0*4+off+320))
+	MOVL(R9L, Mem{Base: dst}.Offset(14*4+off+384))
+	MOVL(R9L, Mem{Base: dst}.Offset(11*4+off+448))
+	MOVL(R9L, Mem{Base: dst}.Offset(12*4+off+512))
+	MOVL(R9L, Mem{Base: dst}.Offset(4*4+off+576))
+	SHRQ(Imm(32), R9)
+	MOVL(R9L, Mem{Base: dst}.Offset(5*4+off+0))
+	MOVL(R9L, Mem{Base: dst}.Offset(15*4+off+64))
+	MOVL(R9L, Mem{Base: dst}.Offset(9*4+off+128))
+	MOVL(R9L, Mem{Base: dst}.Offset(1*4+off+192))
+	MOVL(R9L, Mem{Base: dst}.Offset(11*4+off+256))
+	MOVL(R9L, Mem{Base: dst}.Offset(7*4+off+320))
+	MOVL(R9L, Mem{Base: dst}.Offset(13*4+off+384))
+	MOVL(R9L, Mem{Base: dst}.Offset(3*4+off+448))
+	MOVL(R9L, Mem{Base: dst}.Offset(6*4+off+512))
+	MOVL(R9L, Mem{Base: dst}.Offset(10*4+off+576))
+
+	MOVL(R10L, Mem{Base: dst}.Offset(2*4+off+0))
+	MOVL(R10L, Mem{Base: dst}.Offset(1*4+off+64))
+	MOVL(R10L, Mem{Base: dst}.Offset(15*4+off+128))
+	MOVL(R10L, Mem{Base: dst}.Offset(10*4+off+192))
+	MOVL(R10L, Mem{Base: dst}.Offset(6*4+off+256))
+	MOVL(R10L, Mem{Base: dst}.Offset(8*4+off+320))
+	MOVL(R10L, Mem{Base: dst}.Offset(3*4+off+384))
+	MOVL(R10L, Mem{Base: dst}.Offset(13*4+off+448))
+	MOVL(R10L, Mem{Base: dst}.Offset(14*4+off+512))
+	MOVL(R10L, Mem{Base: dst}.Offset(5*4+off+576))
+	SHRQ(Imm(32), R10)
+	MOVL(R10L, Mem{Base: dst}.Offset(6*4+off+0))
+	MOVL(R10L, Mem{Base: dst}.Offset(11*4+off+64))
+	MOVL(R10L, Mem{Base: dst}.Offset(2*4+off+128))
+	MOVL(R10L, Mem{Base: dst}.Offset(9*4+off+192))
+	MOVL(R10L, Mem{Base: dst}.Offset(1*4+off+256))
+	MOVL(R10L, Mem{Base: dst}.Offset(13*4+off+320))
+	MOVL(R10L, Mem{Base: dst}.Offset(4*4+off+384))
+	MOVL(R10L, Mem{Base: dst}.Offset(8*4+off+448))
+	MOVL(R10L, Mem{Base: dst}.Offset(15*4+off+512))
+	MOVL(R10L, Mem{Base: dst}.Offset(7*4+off+576))
+
+	MOVL(R11L, Mem{Base: dst}.Offset(3*4+off+0))
+	MOVL(R11L, Mem{Base: dst}.Offset(7*4+off+64))
+	MOVL(R11L, Mem{Base: dst}.Offset(13*4+off+128))
+	MOVL(R11L, Mem{Base: dst}.Offset(12*4+off+192))
+	MOVL(R11L, Mem{Base: dst}.Offset(10*4+off+256))
+	MOVL(R11L, Mem{Base: dst}.Offset(1*4+off+320))
+	MOVL(R11L, Mem{Base: dst}.Offset(9*4+off+384))
+	MOVL(R11L, Mem{Base: dst}.Offset(14*4+off+448))
+	MOVL(R11L, Mem{Base: dst}.Offset(0*4+off+512))
+	MOVL(R11L, Mem{Base: dst}.Offset(6*4+off+576))
+	SHRQ(Imm(32), R11)
+	MOVL(R11L, Mem{Base: dst}.Offset(7*4+off+0))
+	MOVL(R11L, Mem{Base: dst}.Offset(14*4+off+64))
+	MOVL(R11L, Mem{Base: dst}.Offset(10*4+off+128))
+	MOVL(R11L, Mem{Base: dst}.Offset(0*4+off+192))
+	MOVL(R11L, Mem{Base: dst}.Offset(5*4+off+256))
+	MOVL(R11L, Mem{Base: dst}.Offset(9*4+off+320))
+	MOVL(R11L, Mem{Base: dst}.Offset(12*4+off+384))
+	MOVL(R11L, Mem{Base: dst}.Offset(1*4+off+448))
+	MOVL(R11L, Mem{Base: dst}.Offset(13*4+off+512))
+	MOVL(R11L, Mem{Base: dst}.Offset(2*4+off+576))
+
+	MOVL(R12L, Mem{Base: dst}.Offset(8*4+off+0))
+	MOVL(R12L, Mem{Base: dst}.Offset(5*4+off+64))
+	MOVL(R12L, Mem{Base: dst}.Offset(4*4+off+128))
+	MOVL(R12L, Mem{Base: dst}.Offset(15*4+off+192))
+	MOVL(R12L, Mem{Base: dst}.Offset(14*4+off+256))
+	MOVL(R12L, Mem{Base: dst}.Offset(3*4+off+320))
+	MOVL(R12L, Mem{Base: dst}.Offset(11*4+off+384))
+	MOVL(R12L, Mem{Base: dst}.Offset(10*4+off+448))
+	MOVL(R12L, Mem{Base: dst}.Offset(7*4+off+512))
+	MOVL(R12L, Mem{Base: dst}.Offset(1*4+off+576))
+	SHRQ(Imm(32), R12)
+	MOVL(R12L, Mem{Base: dst}.Offset(12*4+off+0))
+	MOVL(R12L, Mem{Base: dst}.Offset(2*4+off+64))
+	MOVL(R12L, Mem{Base: dst}.Offset(11*4+off+128))
+	MOVL(R12L, Mem{Base: dst}.Offset(4*4+off+192))
+	MOVL(R12L, Mem{Base: dst}.Offset(0*4+off+256))
+	MOVL(R12L, Mem{Base: dst}.Offset(15*4+off+320))
+	MOVL(R12L, Mem{Base: dst}.Offset(10*4+off+384))
+	MOVL(R12L, Mem{Base: dst}.Offset(7*4+off+448))
+	MOVL(R12L, Mem{Base: dst}.Offset(5*4+off+512))
+	MOVL(R12L, Mem{Base: dst}.Offset(9*4+off+576))
+
+	MOVL(R13L, Mem{Base: dst}.Offset(9*4+off+0))
+	MOVL(R13L, Mem{Base: dst}.Offset(4*4+off+64))
+	MOVL(R13L, Mem{Base: dst}.Offset(8*4+off+128))
+	MOVL(R13L, Mem{Base: dst}.Offset(13*4+off+192))
+	MOVL(R13L, Mem{Base: dst}.Offset(3*4+off+256))
+	MOVL(R13L, Mem{Base: dst}.Offset(5*4+off+320))
+	MOVL(R13L, Mem{Base: dst}.Offset(7*4+off+384))
+	MOVL(R13L, Mem{Base: dst}.Offset(15*4+off+448))
+	MOVL(R13L, Mem{Base: dst}.Offset(11*4+off+512))
+	MOVL(R13L, Mem{Base: dst}.Offset(0*4+off+576))
+	SHRQ(Imm(32), R13)
+	MOVL(R13L, Mem{Base: dst}.Offset(13*4+off+0))
+	MOVL(R13L, Mem{Base: dst}.Offset(10*4+off+64))
+	MOVL(R13L, Mem{Base: dst}.Offset(0*4+off+128))
+	MOVL(R13L, Mem{Base: dst}.Offset(3*4+off+192))
+	MOVL(R13L, Mem{Base: dst}.Offset(9*4+off+256))
+	MOVL(R13L, Mem{Base: dst}.Offset(6*4+off+320))
+	MOVL(R13L, Mem{Base: dst}.Offset(15*4+off+384))
+	MOVL(R13L, Mem{Base: dst}.Offset(4*4+off+448))
+	MOVL(R13L, Mem{Base: dst}.Offset(2*4+off+512))
+	MOVL(R13L, Mem{Base: dst}.Offset(12*4+off+576))
+
+	MOVL(R14L, Mem{Base: dst}.Offset(10*4+off+0))
+	MOVL(R14L, Mem{Base: dst}.Offset(12*4+off+64))
+	MOVL(R14L, Mem{Base: dst}.Offset(1*4+off+128))
+	MOVL(R14L, Mem{Base: dst}.Offset(6*4+off+192))
+	MOVL(R14L, Mem{Base: dst}.Offset(13*4+off+256))
+	MOVL(R14L, Mem{Base: dst}.Offset(4*4+off+320))
+	MOVL(R14L, Mem{Base: dst}.Offset(0*4+off+384))
+	MOVL(R14L, Mem{Base: dst}.Offset(2*4+off+448))
+	MOVL(R14L, Mem{Base: dst}.Offset(8*4+off+512))
+	MOVL(R14L, Mem{Base: dst}.Offset(14*4+off+576))
+	SHRQ(Imm(32), R14)
+	MOVL(R14L, Mem{Base: dst}.Offset(14*4+off+0))
+	MOVL(R14L, Mem{Base: dst}.Offset(3*4+off+64))
+	MOVL(R14L, Mem{Base: dst}.Offset(7*4+off+128))
+	MOVL(R14L, Mem{Base: dst}.Offset(2*4+off+192))
+	MOVL(R14L, Mem{Base: dst}.Offset(15*4+off+256))
+	MOVL(R14L, Mem{Base: dst}.Offset(12*4+off+320))
+	MOVL(R14L, Mem{Base: dst}.Offset(6*4+off+384))
+	MOVL(R14L, Mem{Base: dst}.Offset(0*4+off+448))
+	MOVL(R14L, Mem{Base: dst}.Offset(9*4+off+512))
+	MOVL(R14L, Mem{Base: dst}.Offset(11*4+off+576))
+
+	MOVL(R15L, Mem{Base: dst}.Offset(11*4+off+0))
+	MOVL(R15L, Mem{Base: dst}.Offset(0*4+off+64))
+	MOVL(R15L, Mem{Base: dst}.Offset(12*4+off+128))
+	MOVL(R15L, Mem{Base: dst}.Offset(7*4+off+192))
+	MOVL(R15L, Mem{Base: dst}.Offset(8*4+off+256))
+	MOVL(R15L, Mem{Base: dst}.Offset(14*4+off+320))
+	MOVL(R15L, Mem{Base: dst}.Offset(2*4+off+384))
+	MOVL(R15L, Mem{Base: dst}.Offset(5*4+off+448))
+	MOVL(R15L, Mem{Base: dst}.Offset(1*4+off+512))
+	MOVL(R15L, Mem{Base: dst}.Offset(13*4+off+576))
+	SHRQ(Imm(32), R15)
+	MOVL(R15L, Mem{Base: dst}.Offset(15*4+off+0))
+	MOVL(R15L, Mem{Base: dst}.Offset(6*4+off+64))
+	MOVL(R15L, Mem{Base: dst}.Offset(3*4+off+128))
+	MOVL(R15L, Mem{Base: dst}.Offset(11*4+off+192))
+	MOVL(R15L, Mem{Base: dst}.Offset(7*4+off+256))
+	MOVL(R15L, Mem{Base: dst}.Offset(10*4+off+320))
+	MOVL(R15L, Mem{Base: dst}.Offset(5*4+off+384))
+	MOVL(R15L, Mem{Base: dst}.Offset(9*4+off+448))
+	MOVL(R15L, Mem{Base: dst}.Offset(4*4+off+512))
+	MOVL(R15L, Mem{Base: dst}.Offset(8*4+off+576))
+}
+
+func BLAKE2s_SSE2() {
+	PRECOMPUTE_MSG(BP, 16, SI, R8, R9, R10, R11, R12, R13, R14, R15)
+	for i := 0; i < 10; i++ {
+		ROUND_SSE2(X4, X5, X6, X7, Mem{Base: BP}.Offset(16+64*i), Mem{Base: BP}.Offset(32+64*i), Mem{Base: BP}.Offset(48+64*i), Mem{Base: BP}.Offset(64+64*i), X8)
+	}
+}
+
+func BLAKE2s_SSSE3() {
+	PRECOMPUTE_MSG(BP, 16, SI, R8, R9, R10, R11, R12, R13, R14, R15)
+	for i := 0; i < 10; i++ {
+		ROUND_SSSE3(X4, X5, X6, X7, Mem{Base: BP}.Offset(16+64*i), Mem{Base: BP}.Offset(32+64*i), Mem{Base: BP}.Offset(48+64*i), Mem{Base: BP}.Offset(64+64*i), X8, X13, X14)
+	}
+}
+
+func BLAKE2s_SSE4() {
+	LOAD_MSG_SSE4(X8, X9, X10, X11, SI, 0, 2, 4, 6, 1, 3, 5, 7, 8, 10, 12, 14, 9, 11, 13, 15)
+	ROUND_SSSE3(X4, X5, X6, X7, X8, X9, X10, X11, X8, X13, X14)
+	LOAD_MSG_SSE4(X8, X9, X10, X11, SI, 14, 4, 9, 13, 10, 8, 15, 6, 1, 0, 11, 5, 12, 2, 7, 3)
+	ROUND_SSSE3(X4, X5, X6, X7, X8, X9, X10, X11, X8, X13, X14)
+	LOAD_MSG_SSE4(X8, X9, X10, X11, SI, 11, 12, 5, 15, 8, 0, 2, 13, 10, 3, 7, 9, 14, 6, 1, 4)
+	ROUND_SSSE3(X4, X5, X6, X7, X8, X9, X10, X11, X8, X13, X14)
+	LOAD_MSG_SSE4(X8, X9, X10, X11, SI, 7, 3, 13, 11, 9, 1, 12, 14, 2, 5, 4, 15, 6, 10, 0, 8)
+	ROUND_SSSE3(X4, X5, X6, X7, X8, X9, X10, X11, X8, X13, X14)
+	LOAD_MSG_SSE4(X8, X9, X10, X11, SI, 9, 5, 2, 10, 0, 7, 4, 15, 14, 11, 6, 3, 1, 12, 8, 13)
+	ROUND_SSSE3(X4, X5, X6, X7, X8, X9, X10, X11, X8, X13, X14)
+	LOAD_MSG_SSE4(X8, X9, X10, X11, SI, 2, 6, 0, 8, 12, 10, 11, 3, 4, 7, 15, 1, 13, 5, 14, 9)
+	ROUND_SSSE3(X4, X5, X6, X7, X8, X9, X10, X11, X8, X13, X14)
+	LOAD_MSG_SSE4(X8, X9, X10, X11, SI, 12, 1, 14, 4, 5, 15, 13, 10, 0, 6, 9, 8, 7, 3, 2, 11)
+	ROUND_SSSE3(X4, X5, X6, X7, X8, X9, X10, X11, X8, X13, X14)
+	LOAD_MSG_SSE4(X8, X9, X10, X11, SI, 13, 7, 12, 3, 11, 14, 1, 9, 5, 15, 8, 2, 0, 4, 6, 10)
+	ROUND_SSSE3(X4, X5, X6, X7, X8, X9, X10, X11, X8, X13, X14)
+	LOAD_MSG_SSE4(X8, X9, X10, X11, SI, 6, 14, 11, 0, 15, 9, 3, 8, 12, 13, 1, 10, 2, 7, 4, 5)
+	ROUND_SSSE3(X4, X5, X6, X7, X8, X9, X10, X11, X8, X13, X14)
+	LOAD_MSG_SSE4(X8, X9, X10, X11, SI, 10, 8, 7, 1, 2, 4, 6, 5, 15, 9, 3, 13, 11, 14, 12, 0)
+	ROUND_SSSE3(X4, X5, X6, X7, X8, X9, X10, X11, X8, X13, X14)
+}
+
+func HASH_BLOCKS(h, c, flag, blocks_base, blocks_len Mem, BLAKE2s_FUNC func()) {
+	MOVQ(h, RAX)
+	MOVQ(c, RBX)
+	MOVL(flag, ECX)
+	MOVQ(blocks_base, RSI)
+	MOVQ(blocks_len, RDX)
+
+	MOVQ(RSP, RBP)
+	ADDQ(Imm(15), RBP)
+	ANDQ(I32(^15), RBP)
+
+	MOVQ(Mem{Base: BX}.Offset(0), R9)
+	MOVQ(R9, Mem{Base: BP}.Offset(0))
+	MOVQ(RCX, Mem{Base: BP}.Offset(8))
+
+	MOVOU(Mem{Base: AX}.Offset(0), X0)
+	MOVOU(Mem{Base: AX}.Offset(16), X1)
+
+	iv0 := iv0_DATA()
+	iv1 := iv1_DATA()
+	MOVOU(iv0, X2)
+	MOVOU(iv1, X3)
+
+	counter := counter_DATA()
+	rol16 := rol16_DATA()
+	rol8 := rol8_DATA()
+	MOVOU(counter, X12)
+	MOVOU(rol16, X13)
+	MOVOU(rol8, X14)
+	MOVO(Mem{Base: BP}.Offset(0), X15)
+
+	Label("loop")
+	MOVO(X0, X4)
+	MOVO(X1, X5)
+	MOVO(X2, X6)
+	MOVO(X3, X7)
+
+	PADDQ(X12, X15)
+	PXOR(X15, X7)
+
+	BLAKE2s_FUNC()
+
+	PXOR(X4, X0)
+	PXOR(X5, X1)
+	PXOR(X6, X0)
+	PXOR(X7, X1)
+
+	LEAQ(Mem{Base: SI}.Offset(64), RSI)
+	SUBQ(Imm(64), RDX)
+	JNE(LabelRef("loop"))
+
+	MOVO(X15, Mem{Base: BP}.Offset(0))
+	MOVQ(Mem{Base: BP}.Offset(0), R9)
+	MOVQ(R9, Mem{Base: BX}.Offset(0))
+
+	MOVOU(X0, Mem{Base: AX}.Offset(0))
+	MOVOU(X1, Mem{Base: AX}.Offset(16))
+}
+
+func hashBlocksSSE2() {
+	Implement("hashBlocksSSE2")
+	Attributes(0)
+	AllocLocal(672) // frame = 656 + 16 byte alignment
+
+	h := NewParamAddr("h", 0)
+	c := NewParamAddr("c", 8)
+	flag := NewParamAddr("flag", 16)
+	blocks_base := NewParamAddr("blocks_base", 24)
+	blocks_len := NewParamAddr("blocks_len", 32)
+
+	HASH_BLOCKS(h, c, flag, blocks_base, blocks_len, BLAKE2s_SSE2)
+	RET()
+}
+
+func hashBlocksSSSE3() {
+	Implement("hashBlocksSSSE3")
+	Attributes(0)
+	AllocLocal(672) // frame = 656 + 16 byte alignment
+
+	h := NewParamAddr("h", 0)
+	c := NewParamAddr("c", 8)
+	flag := NewParamAddr("flag", 16)
+	blocks_base := NewParamAddr("blocks_base", 24)
+	blocks_len := NewParamAddr("blocks_len", 32)
+
+	HASH_BLOCKS(h, c, flag, blocks_base, blocks_len, BLAKE2s_SSSE3)
+	RET()
+}
+
+func hashBlocksSSE4() {
+	Implement("hashBlocksSSE4")
+	Attributes(0)
+	AllocLocal(32) // frame = 16 + 16 byte alignment
+
+	h := NewParamAddr("h", 0)
+	c := NewParamAddr("c", 8)
+	flag := NewParamAddr("flag", 16)
+	blocks_base := NewParamAddr("blocks_base", 24)
+	blocks_len := NewParamAddr("blocks_len", 32)
+
+	HASH_BLOCKS(h, c, flag, blocks_base, blocks_len, BLAKE2s_SSE4)
+	RET()
+}
+
+// ##~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~DATA SECTION~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~##
+
+var iv0_DATA_ptr, iv1_DATA_ptr, rol16_DATA_ptr, rol8_DATA_ptr, counter_DATA_ptr *Mem
+
+func iv0_DATA() Mem {
+	if iv0_DATA_ptr != nil {
+		return *iv0_DATA_ptr
+	}
+
+	iv0_DATA := GLOBL("iv0", NOPTR|RODATA)
+	iv0_DATA_ptr = &iv0_DATA
+	DATA(0x00, U32(0x6a09e667))
+	DATA(0x04, U32(0xbb67ae85))
+	DATA(0x08, U32(0x3c6ef372))
+	DATA(0x0c, U32(0xa54ff53a))
+	return iv0_DATA
+}
+
+func iv1_DATA() Mem {
+	if iv1_DATA_ptr != nil {
+		return *iv1_DATA_ptr
+	}
+
+	iv1_DATA := GLOBL("iv1", NOPTR|RODATA)
+	iv1_DATA_ptr = &iv1_DATA
+	DATA(0x00, U32(0x510e527f))
+	DATA(0x04, U32(0x9b05688c))
+	DATA(0x08, U32(0x1f83d9ab))
+	DATA(0x0c, U32(0x5be0cd19))
+	return iv1_DATA
+}
+
+func rol16_DATA() Mem {
+	if rol16_DATA_ptr != nil {
+		return *rol16_DATA_ptr
+	}
+
+	rol16_DATA := GLOBL("rol16", NOPTR|RODATA)
+	rol16_DATA_ptr = &rol16_DATA
+	DATA(0x00, U64(0x0504070601000302))
+	DATA(0x08, U64(0x0D0C0F0E09080B0A))
+	return rol16_DATA
+}
+
+func rol8_DATA() Mem {
+	if rol8_DATA_ptr != nil {
+		return *rol8_DATA_ptr
+	}
+
+	rol8_DATA := GLOBL("rol8", NOPTR|RODATA)
+	rol8_DATA_ptr = &rol8_DATA
+	DATA(0x00, U64(0x0407060500030201))
+	DATA(0x08, U64(0x0C0F0E0D080B0A09))
+	return rol8_DATA
+}
+
+func counter_DATA() Mem {
+	if counter_DATA_ptr != nil {
+		return *counter_DATA_ptr
+	}
+
+	counter_DATA := GLOBL("counter", NOPTR|RODATA)
+	counter_DATA_ptr = &counter_DATA
+	DATA(0x00, U64(0x0000000000000040))
+	DATA(0x08, U64(0x0000000000000000))
+	return counter_DATA
+}

blake2s/_asm/go.mod

@@ -0,0 +1,15 @@
+module blake2s/_asm
+
+go 1.23
+
+require (
+	github.com/mmcloughlin/avo v0.6.0
+	golang.org/x/crypto v0.33.0
+)
+
+require (
+	golang.org/x/mod v0.20.0 // indirect
+	golang.org/x/sync v0.8.0 // indirect
+	golang.org/x/sys v0.30.0 // indirect
+	golang.org/x/tools v0.24.0 // indirect
+)

blake2s/_asm/go.sum

@@ -0,0 +1,12 @@
+github.com/mmcloughlin/avo v0.6.0 h1:QH6FU8SKoTLaVs80GA8TJuLNkUYl4VokHKlPhVDg4YY=
+github.com/mmcloughlin/avo v0.6.0/go.mod h1:8CoAGaCSYXtCPR+8y18Y9aB/kxb8JSS6FRI7mSkvD+8=
+golang.org/x/crypto v0.33.0 h1:IOBPskki6Lysi0lo9qQvbxiQ+FvsCC/YWOecCHAixus=
+golang.org/x/crypto v0.33.0/go.mod h1:bVdXmD7IV/4GdElGPozy6U7lWdRXA4qyRVGJV57uQ5M=
+golang.org/x/mod v0.20.0 h1:utOm6MM3R3dnawAiJgn0y+xvuYRsm1RKM/4giyfDgV0=
+golang.org/x/mod v0.20.0/go.mod h1:hTbmBsO62+eylJbnUtE2MGJUyE7QWk4xUqPFrRgJ+7c=
+golang.org/x/sync v0.8.0 h1:3NFvSEYkUoMifnESzZl15y791HH1qU2xm6eCJU5ZPXQ=
+golang.org/x/sync v0.8.0/go.mod h1:Czt+wKu1gCyEFDUtn0jG5QVvpJ6rzVqr5aXyt9drQfk=
+golang.org/x/sys v0.30.0 h1:QjkSwP/36a20jFYWkSue1YwXzLmsV5Gfq7Eiy72C1uc=
+golang.org/x/sys v0.30.0/go.mod h1:/VUhepiaJMQUp4+oa/7Zr1D23ma6VTLIYjOOTFZPUcA=
+golang.org/x/tools v0.24.0 h1:J1shsA93PJUEVaUSaay7UXAyE8aimq3GW0pjlolpa24=
+golang.org/x/tools v0.24.0/go.mod h1:YhNqVBIfWHdzvTLs0d8LCuMhkKUgSUKldakyV7W/WDQ=

bn256/gfp12.go

@@ -5,7 +5,7 @@
 package bn256
 
 // For details of the algorithms used, see "Multiplication and Squaring on
-// Pairing-Friendly Fields, Devegili et al.
+// Pairing-Friendly Fields", Devegili et al.
 // http://eprint.iacr.org/2006/471.pdf.
 
 import (

bn256/gfp2.go

@@ -5,7 +5,7 @@
 package bn256
 
 // For details of the algorithms used, see "Multiplication and Squaring on
-// Pairing-Friendly Fields, Devegili et al.
+// Pairing-Friendly Fields", Devegili et al.
 // http://eprint.iacr.org/2006/471.pdf.
 
 import (

bn256/gfp6.go

@@ -5,7 +5,7 @@
 package bn256
 
 // For details of the algorithms used, see "Multiplication and Squaring on
-// Pairing-Friendly Fields, Devegili et al.
+// Pairing-Friendly Fields", Devegili et al.
 // http://eprint.iacr.org/2006/471.pdf.
 
 import (

chacha20/chacha_noasm.go

@@ -2,7 +2,7 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
-//go:build (!arm64 && !s390x && !ppc64le) || !gc || purego
+//go:build (!arm64 && !s390x && !ppc64 && !ppc64le) || !gc || purego
 
 package chacha20
 

chacha20/chacha_ppc64x.go

@@ -2,7 +2,7 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
-//go:build gc && !purego
+//go:build gc && !purego && (ppc64 || ppc64le)
 
 package chacha20
 

chacha20/chacha_ppc64x.s

@@ -19,7 +19,7 @@
 // The differences in this and the original implementation are
 // due to the calling conventions and initialization of constants.
 
-//go:build gc && !purego
+//go:build gc && !purego && (ppc64 || ppc64le)
 
 #include "textflag.h"
 
@@ -36,32 +36,68 @@
 // for VPERMXOR
 #define MASK  R18
 
-DATA consts<>+0x00(SB)/8, $0x3320646e61707865
-DATA consts<>+0x08(SB)/8, $0x6b20657479622d32
-DATA consts<>+0x10(SB)/8, $0x0000000000000001
-DATA consts<>+0x18(SB)/8, $0x0000000000000000
-DATA consts<>+0x20(SB)/8, $0x0000000000000004
-DATA consts<>+0x28(SB)/8, $0x0000000000000000
-DATA consts<>+0x30(SB)/8, $0x0a0b08090e0f0c0d
-DATA consts<>+0x38(SB)/8, $0x0203000106070405
-DATA consts<>+0x40(SB)/8, $0x090a0b080d0e0f0c
-DATA consts<>+0x48(SB)/8, $0x0102030005060704
-DATA consts<>+0x50(SB)/8, $0x6170786561707865
-DATA consts<>+0x58(SB)/8, $0x6170786561707865
-DATA consts<>+0x60(SB)/8, $0x3320646e3320646e
-DATA consts<>+0x68(SB)/8, $0x3320646e3320646e
-DATA consts<>+0x70(SB)/8, $0x79622d3279622d32
-DATA consts<>+0x78(SB)/8, $0x79622d3279622d32
-DATA consts<>+0x80(SB)/8, $0x6b2065746b206574
-DATA consts<>+0x88(SB)/8, $0x6b2065746b206574
-DATA consts<>+0x90(SB)/8, $0x0000000100000000
-DATA consts<>+0x98(SB)/8, $0x0000000300000002
-DATA consts<>+0xa0(SB)/8, $0x5566774411223300
-DATA consts<>+0xa8(SB)/8, $0xddeeffcc99aabb88
-DATA consts<>+0xb0(SB)/8, $0x6677445522330011
-DATA consts<>+0xb8(SB)/8, $0xeeffccddaabb8899
+DATA consts<>+0x00(SB)/4, $0x61707865
+DATA consts<>+0x04(SB)/4, $0x3320646e
+DATA consts<>+0x08(SB)/4, $0x79622d32
+DATA consts<>+0x0c(SB)/4, $0x6b206574
+DATA consts<>+0x10(SB)/4, $0x00000001
+DATA consts<>+0x14(SB)/4, $0x00000000
+DATA consts<>+0x18(SB)/4, $0x00000000
+DATA consts<>+0x1c(SB)/4, $0x00000000
+DATA consts<>+0x20(SB)/4, $0x00000004
+DATA consts<>+0x24(SB)/4, $0x00000000
+DATA consts<>+0x28(SB)/4, $0x00000000
+DATA consts<>+0x2c(SB)/4, $0x00000000
+DATA consts<>+0x30(SB)/4, $0x0e0f0c0d
+DATA consts<>+0x34(SB)/4, $0x0a0b0809
+DATA consts<>+0x38(SB)/4, $0x06070405
+DATA consts<>+0x3c(SB)/4, $0x02030001
+DATA consts<>+0x40(SB)/4, $0x0d0e0f0c
+DATA consts<>+0x44(SB)/4, $0x090a0b08
+DATA consts<>+0x48(SB)/4, $0x05060704
+DATA consts<>+0x4c(SB)/4, $0x01020300
+DATA consts<>+0x50(SB)/4, $0x61707865
+DATA consts<>+0x54(SB)/4, $0x61707865
+DATA consts<>+0x58(SB)/4, $0x61707865
+DATA consts<>+0x5c(SB)/4, $0x61707865
+DATA consts<>+0x60(SB)/4, $0x3320646e
+DATA consts<>+0x64(SB)/4, $0x3320646e
+DATA consts<>+0x68(SB)/4, $0x3320646e
+DATA consts<>+0x6c(SB)/4, $0x3320646e
+DATA consts<>+0x70(SB)/4, $0x79622d32
+DATA consts<>+0x74(SB)/4, $0x79622d32
+DATA consts<>+0x78(SB)/4, $0x79622d32
+DATA consts<>+0x7c(SB)/4, $0x79622d32
+DATA consts<>+0x80(SB)/4, $0x6b206574
+DATA consts<>+0x84(SB)/4, $0x6b206574
+DATA consts<>+0x88(SB)/4, $0x6b206574
+DATA consts<>+0x8c(SB)/4, $0x6b206574
+DATA consts<>+0x90(SB)/4, $0x00000000
+DATA consts<>+0x94(SB)/4, $0x00000001
+DATA consts<>+0x98(SB)/4, $0x00000002
+DATA consts<>+0x9c(SB)/4, $0x00000003
+DATA consts<>+0xa0(SB)/4, $0x11223300
+DATA consts<>+0xa4(SB)/4, $0x55667744
+DATA consts<>+0xa8(SB)/4, $0x99aabb88
+DATA consts<>+0xac(SB)/4, $0xddeeffcc
+DATA consts<>+0xb0(SB)/4, $0x22330011
+DATA consts<>+0xb4(SB)/4, $0x66774455
+DATA consts<>+0xb8(SB)/4, $0xaabb8899
+DATA consts<>+0xbc(SB)/4, $0xeeffccdd
 GLOBL consts<>(SB), RODATA, $0xc0
 
+#ifdef GOARCH_ppc64
+#define BE_XXBRW_INIT() \
+		LVSL (R0)(R0), V24 \
+		VSPLTISB $3, V25   \
+		VXOR V24, V25, V24 \
+
+#define BE_XXBRW(vr) VPERM vr, vr, V24, vr
+#else
+#define BE_XXBRW_INIT()
+#define BE_XXBRW(vr)
+#endif
+
 //func chaCha20_ctr32_vsx(out, inp *byte, len int, key *[8]uint32, counter *uint32)
 TEXT ·chaCha20_ctr32_vsx(SB),NOSPLIT,$64-40
 	MOVD out+0(FP), OUT
@@ -94,6 +130,8 @@ TEXT ·chaCha20_ctr32_vsx(SB),NOSPLIT,$64-40
 	// Clear V27
 	VXOR V27, V27, V27
 
+	BE_XXBRW_INIT()
+
 	// V28
 	LXVW4X (CONSTBASE)(R11), VS60
 
@@ -299,6 +337,11 @@ loop_vsx:
 	VADDUWM V8, V18, V8
 	VADDUWM V12, V19, V12
 
+	BE_XXBRW(V0)
+	BE_XXBRW(V4)
+	BE_XXBRW(V8)
+	BE_XXBRW(V12)
+
 	CMPU LEN, $64
 	BLT tail_vsx
 
@@ -327,6 +370,11 @@ loop_vsx:
 	VADDUWM V9, V18, V8
 	VADDUWM V13, V19, V12
 
+	BE_XXBRW(V0)
+	BE_XXBRW(V4)
+	BE_XXBRW(V8)
+	BE_XXBRW(V12)
+
 	CMPU  LEN, $64
 	BLT   tail_vsx
 
@@ -334,8 +382,8 @@ loop_vsx:
 	LXVW4X (INP)(R8), VS60
 	LXVW4X (INP)(R9), VS61
 	LXVW4X (INP)(R10), VS62
-	VXOR   V27, V0, V27
 
+	VXOR V27, V0, V27
 	VXOR V28, V4, V28
 	VXOR V29, V8, V29
 	VXOR V30, V12, V30
@@ -354,6 +402,11 @@ loop_vsx:
 	VADDUWM V10, V18, V8
 	VADDUWM V14, V19, V12
 
+	BE_XXBRW(V0)
+	BE_XXBRW(V4)
+	BE_XXBRW(V8)
+	BE_XXBRW(V12)
+
 	CMPU LEN, $64
 	BLT  tail_vsx
 
@@ -381,6 +434,11 @@ loop_vsx:
 	VADDUWM V11, V18, V8
 	VADDUWM V15, V19, V12
 
+	BE_XXBRW(V0)
+	BE_XXBRW(V4)
+	BE_XXBRW(V8)
+	BE_XXBRW(V12)
+
 	CMPU  LEN, $64
 	BLT   tail_vsx
 
@@ -408,9 +466,9 @@ loop_vsx:
 
 done_vsx:
 	// Increment counter by number of 64 byte blocks
-	MOVD (CNT), R14
+	MOVWZ (CNT), R14
 	ADD  BLOCKS, R14
-	MOVD R14, (CNT)
+	MOVWZ R14, (CNT)
 	RET
 
 tail_vsx:

chacha20poly1305/_asm/go.mod

@@ -0,0 +1,15 @@
+module chacha20poly1305/_asm
+
+go 1.23
+
+require (
+	github.com/mmcloughlin/avo v0.6.0
+	golang.org/x/crypto v0.33.0
+)
+
+require (
+	golang.org/x/mod v0.20.0 // indirect
+	golang.org/x/sync v0.8.0 // indirect
+	golang.org/x/sys v0.30.0 // indirect
+	golang.org/x/tools v0.24.0 // indirect
+)

chacha20poly1305/_asm/go.sum

@@ -0,0 +1,12 @@
+github.com/mmcloughlin/avo v0.6.0 h1:QH6FU8SKoTLaVs80GA8TJuLNkUYl4VokHKlPhVDg4YY=
+github.com/mmcloughlin/avo v0.6.0/go.mod h1:8CoAGaCSYXtCPR+8y18Y9aB/kxb8JSS6FRI7mSkvD+8=
+golang.org/x/crypto v0.33.0 h1:IOBPskki6Lysi0lo9qQvbxiQ+FvsCC/YWOecCHAixus=
+golang.org/x/crypto v0.33.0/go.mod h1:bVdXmD7IV/4GdElGPozy6U7lWdRXA4qyRVGJV57uQ5M=
+golang.org/x/mod v0.20.0 h1:utOm6MM3R3dnawAiJgn0y+xvuYRsm1RKM/4giyfDgV0=
+golang.org/x/mod v0.20.0/go.mod h1:hTbmBsO62+eylJbnUtE2MGJUyE7QWk4xUqPFrRgJ+7c=
+golang.org/x/sync v0.8.0 h1:3NFvSEYkUoMifnESzZl15y791HH1qU2xm6eCJU5ZPXQ=
+golang.org/x/sync v0.8.0/go.mod h1:Czt+wKu1gCyEFDUtn0jG5QVvpJ6rzVqr5aXyt9drQfk=
+golang.org/x/sys v0.30.0 h1:QjkSwP/36a20jFYWkSue1YwXzLmsV5Gfq7Eiy72C1uc=
+golang.org/x/sys v0.30.0/go.mod h1:/VUhepiaJMQUp4+oa/7Zr1D23ma6VTLIYjOOTFZPUcA=
+golang.org/x/tools v0.24.0 h1:J1shsA93PJUEVaUSaay7UXAyE8aimq3GW0pjlolpa24=
+golang.org/x/tools v0.24.0/go.mod h1:YhNqVBIfWHdzvTLs0d8LCuMhkKUgSUKldakyV7W/WDQ=

cryptobyte/asn1.go

@@ -234,7 +234,7 @@ func (b *Builder) AddASN1(tag asn1.Tag, f BuilderContinuation) {
 	// Identifiers with the low five bits set indicate high-tag-number format
 	// (two or more octets), which we don't support.
 	if tag&0x1f == 0x1f {
-		b.err = fmt.Errorf("cryptobyte: high-tag number identifier octects not supported: 0x%x", tag)
+		b.err = fmt.Errorf("cryptobyte: high-tag number identifier octets not supported: 0x%x", tag)
 		return
 	}
 	b.AddUint8(uint8(tag))

go.mod

@@ -1,11 +1,11 @@
 module golang.org/x/crypto
 
-go 1.20
+go 1.23.0
 
 require (
 	golang.org/x/net v0.21.0 // tagx:ignore
-	golang.org/x/sys v0.23.0
-	golang.org/x/term v0.23.0
+	golang.org/x/sys v0.32.0
+	golang.org/x/term v0.31.0
 )
 
-require golang.org/x/text v0.17.0 // indirect
+require golang.org/x/text v0.24.0 // indirect

go.sum

@@ -1,8 +1,8 @@
 golang.org/x/net v0.21.0 h1:AQyQV4dYCvJ7vGmJyKki9+PBdyvhkSd8EIx/qb0AYv4=
 golang.org/x/net v0.21.0/go.mod h1:bIjVDfnllIU7BJ2DNgfnXvpSvtn8VRwhlsaeUTyUS44=
-golang.org/x/sys v0.23.0 h1:YfKFowiIMvtgl1UERQoTPPToxltDeZfbj4H7dVUCwmM=
-golang.org/x/sys v0.23.0/go.mod h1:/VUhepiaJMQUp4+oa/7Zr1D23ma6VTLIYjOOTFZPUcA=
-golang.org/x/term v0.23.0 h1:F6D4vR+EHoL9/sWAWgAR1H2DcHr4PareCbAaCo1RpuU=
-golang.org/x/term v0.23.0/go.mod h1:DgV24QBUrK6jhZXl+20l6UWznPlwAHm1Q1mGHtydmSk=
-golang.org/x/text v0.17.0 h1:XtiM5bkSOt+ewxlOE/aE/AKEHibwj/6gvWMl9Rsh0Qc=
-golang.org/x/text v0.17.0/go.mod h1:BuEKDfySbSR4drPmRPG/7iBdf8hvFMuRexcpahXilzY=
+golang.org/x/sys v0.32.0 h1:s77OFDvIQeibCmezSnk/q6iAfkdiQaJi4VzroCFrN20=
+golang.org/x/sys v0.32.0/go.mod h1:BJP2sWEmIv4KK5OTEluFJCKSidICx8ciO85XgH3Ak8k=
+golang.org/x/term v0.31.0 h1:erwDkOK1Msy6offm1mOgvspSkslFnIGsFnxOKoufg3o=
+golang.org/x/term v0.31.0/go.mod h1:R4BeIy7D95HzImkxGkTW1UQTtP54tio2RyHz7PwK0aw=
+golang.org/x/text v0.24.0 h1:dd5Bzh4yt5KYA8f9CJHCP4FB4D51c2c6JvN37xJJkJ0=
+golang.org/x/text v0.24.0/go.mod h1:L8rBsPeo2pSS+xqN0d5u2ikmjtmoJbDBT1b7nHvFCdU=

internal/poly1305/_asm/go.mod

@@ -0,0 +1,15 @@
+module internal/poly1305/_asm
+
+go 1.23
+
+require (
+	github.com/mmcloughlin/avo v0.6.0
+	golang.org/x/crypto v0.33.0
+)
+
+require (
+	golang.org/x/mod v0.20.0 // indirect
+	golang.org/x/sync v0.8.0 // indirect
+	golang.org/x/sys v0.30.0 // indirect
+	golang.org/x/tools v0.24.0 // indirect
+)

internal/poly1305/_asm/go.sum

@@ -0,0 +1,12 @@
+github.com/mmcloughlin/avo v0.6.0 h1:QH6FU8SKoTLaVs80GA8TJuLNkUYl4VokHKlPhVDg4YY=
+github.com/mmcloughlin/avo v0.6.0/go.mod h1:8CoAGaCSYXtCPR+8y18Y9aB/kxb8JSS6FRI7mSkvD+8=
+golang.org/x/crypto v0.33.0 h1:IOBPskki6Lysi0lo9qQvbxiQ+FvsCC/YWOecCHAixus=
+golang.org/x/crypto v0.33.0/go.mod h1:bVdXmD7IV/4GdElGPozy6U7lWdRXA4qyRVGJV57uQ5M=
+golang.org/x/mod v0.20.0 h1:utOm6MM3R3dnawAiJgn0y+xvuYRsm1RKM/4giyfDgV0=
+golang.org/x/mod v0.20.0/go.mod h1:hTbmBsO62+eylJbnUtE2MGJUyE7QWk4xUqPFrRgJ+7c=
+golang.org/x/sync v0.8.0 h1:3NFvSEYkUoMifnESzZl15y791HH1qU2xm6eCJU5ZPXQ=
+golang.org/x/sync v0.8.0/go.mod h1:Czt+wKu1gCyEFDUtn0jG5QVvpJ6rzVqr5aXyt9drQfk=
+golang.org/x/sys v0.30.0 h1:QjkSwP/36a20jFYWkSue1YwXzLmsV5Gfq7Eiy72C1uc=
+golang.org/x/sys v0.30.0/go.mod h1:/VUhepiaJMQUp4+oa/7Zr1D23ma6VTLIYjOOTFZPUcA=
+golang.org/x/tools v0.24.0 h1:J1shsA93PJUEVaUSaay7UXAyE8aimq3GW0pjlolpa24=
+golang.org/x/tools v0.24.0/go.mod h1:YhNqVBIfWHdzvTLs0d8LCuMhkKUgSUKldakyV7W/WDQ=

internal/poly1305/_asm/sum_amd64_asm.go

@@ -0,0 +1,126 @@
+// Copyright 2024 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.
+
+package main
+
+import (
+	. "github.com/mmcloughlin/avo/build"
+	. "github.com/mmcloughlin/avo/operand"
+	. "github.com/mmcloughlin/avo/reg"
+	_ "golang.org/x/crypto/sha3"
+)
+
+//go:generate go run . -out ../sum_amd64.s -pkg poly1305
+
+func main() {
+	Package("golang.org/x/crypto/internal/poly1305")
+	ConstraintExpr("gc,!purego")
+	update()
+	Generate()
+}
+
+func update() {
+	Implement("update")
+
+	Load(Param("state"), RDI)
+	MOVQ(NewParamAddr("msg_base", 8), RSI)
+	MOVQ(NewParamAddr("msg_len", 16), R15)
+
+	MOVQ(Mem{Base: DI}.Offset(0), R8)   // h0
+	MOVQ(Mem{Base: DI}.Offset(8), R9)   // h1
+	MOVQ(Mem{Base: DI}.Offset(16), R10) // h2
+	MOVQ(Mem{Base: DI}.Offset(24), R11) // r0
+	MOVQ(Mem{Base: DI}.Offset(32), R12) // r1
+
+	CMPQ(R15, Imm(16))
+	JB(LabelRef("bytes_between_0_and_15"))
+
+	Label("loop")
+	POLY1305_ADD(RSI, R8, R9, R10)
+
+	Label("multiply")
+	POLY1305_MUL(R8, R9, R10, R11, R12, RBX, RCX, R13, R14)
+	SUBQ(Imm(16), R15)
+	CMPQ(R15, Imm(16))
+	JAE(LabelRef("loop"))
+
+	Label("bytes_between_0_and_15")
+	TESTQ(R15, R15)
+	JZ(LabelRef("done"))
+	MOVQ(U32(1), RBX)
+	XORQ(RCX, RCX)
+	XORQ(R13, R13)
+	ADDQ(R15, RSI)
+
+	Label("flush_buffer")
+	SHLQ(Imm(8), RBX, RCX)
+	SHLQ(Imm(8), RBX)
+	MOVB(Mem{Base: SI}.Offset(-1), R13B)
+	XORQ(R13, RBX)
+	DECQ(RSI)
+	DECQ(R15)
+	JNZ(LabelRef("flush_buffer"))
+
+	ADDQ(RBX, R8)
+	ADCQ(RCX, R9)
+	ADCQ(Imm(0), R10)
+	MOVQ(U32(16), R15)
+	JMP(LabelRef("multiply"))
+
+	Label("done")
+	MOVQ(R8, Mem{Base: DI}.Offset(0))
+	MOVQ(R9, Mem{Base: DI}.Offset(8))
+	MOVQ(R10, Mem{Base: DI}.Offset(16))
+	RET()
+}
+
+func POLY1305_ADD(msg, h0, h1, h2 GPPhysical) {
+	ADDQ(Mem{Base: msg}.Offset(0), h0)
+	ADCQ(Mem{Base: msg}.Offset(8), h1)
+	ADCQ(Imm(1), h2)
+	LEAQ(Mem{Base: msg}.Offset(16), msg)
+}
+
+func POLY1305_MUL(h0, h1, h2, r0, r1, t0, t1, t2, t3 GPPhysical) {
+	MOVQ(r0, RAX)
+	MULQ(h0)
+	MOVQ(RAX, t0)
+	MOVQ(RDX, t1)
+	MOVQ(r0, RAX)
+	MULQ(h1)
+	ADDQ(RAX, t1)
+	ADCQ(Imm(0), RDX)
+	MOVQ(r0, t2)
+	IMULQ(h2, t2)
+	ADDQ(RDX, t2)
+
+	MOVQ(r1, RAX)
+	MULQ(h0)
+	ADDQ(RAX, t1)
+	ADCQ(Imm(0), RDX)
+	MOVQ(RDX, h0)
+	MOVQ(r1, t3)
+	IMULQ(h2, t3)
+	MOVQ(r1, RAX)
+	MULQ(h1)
+	ADDQ(RAX, t2)
+	ADCQ(RDX, t3)
+	ADDQ(h0, t2)
+	ADCQ(Imm(0), t3)
+
+	MOVQ(t0, h0)
+	MOVQ(t1, h1)
+	MOVQ(t2, h2)
+	ANDQ(Imm(3), h2)
+	MOVQ(t2, t0)
+	ANDQ(I32(-4), t0)
+	ADDQ(t0, h0)
+	ADCQ(t3, h1)
+	ADCQ(Imm(0), h2)
+	SHRQ(Imm(2), t3, t2)
+	SHRQ(Imm(2), t3)
+	ADDQ(t2, h0)
+	ADCQ(t3, h1)
+	ADCQ(Imm(0), h2)
+}

internal/poly1305/mac_noasm.go

@@ -2,7 +2,7 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
-//go:build (!amd64 && !ppc64le && !s390x) || !gc || purego
+//go:build (!amd64 && !loong64 && !ppc64le && !ppc64 && !s390x) || !gc || purego
 
 package poly1305
 

internal/poly1305/sum_amd64.s

@@ -1,108 +1,93 @@
-// Copyright 2012 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.
+// Code generated by command: go run sum_amd64_asm.go -out ../sum_amd64.s -pkg poly1305. DO NOT EDIT.
 
 //go:build gc && !purego
 
-#include "textflag.h"
-
-#define POLY1305_ADD(msg, h0, h1, h2) \
-	ADDQ 0(msg), h0;  \
-	ADCQ 8(msg), h1;  \
-	ADCQ $1, h2;      \
-	LEAQ 16(msg), msg
-
-#define POLY1305_MUL(h0, h1, h2, r0, r1, t0, t1, t2, t3) \
-	MOVQ  r0, AX;                  \
-	MULQ  h0;                      \
-	MOVQ  AX, t0;                  \
-	MOVQ  DX, t1;                  \
-	MOVQ  r0, AX;                  \
-	MULQ  h1;                      \
-	ADDQ  AX, t1;                  \
-	ADCQ  $0, DX;                  \
-	MOVQ  r0, t2;                  \
-	IMULQ h2, t2;                  \
-	ADDQ  DX, t2;                  \
-	                               \
-	MOVQ  r1, AX;                  \
-	MULQ  h0;                      \
-	ADDQ  AX, t1;                  \
-	ADCQ  $0, DX;                  \
-	MOVQ  DX, h0;                  \
-	MOVQ  r1, t3;                  \
-	IMULQ h2, t3;                  \
-	MOVQ  r1, AX;                  \
-	MULQ  h1;                      \
-	ADDQ  AX, t2;                  \
-	ADCQ  DX, t3;                  \
-	ADDQ  h0, t2;                  \
-	ADCQ  $0, t3;                  \
-	                               \
-	MOVQ  t0, h0;                  \
-	MOVQ  t1, h1;                  \
-	MOVQ  t2, h2;                  \
-	ANDQ  $3, h2;                  \
-	MOVQ  t2, t0;                  \
-	ANDQ  $0xFFFFFFFFFFFFFFFC, t0; \
-	ADDQ  t0, h0;                  \
-	ADCQ  t3, h1;                  \
-	ADCQ  $0, h2;                  \
-	SHRQ  $2, t3, t2;              \
-	SHRQ  $2, t3;                  \
-	ADDQ  t2, h0;                  \
-	ADCQ  t3, h1;                  \
-	ADCQ  $0, h2
-
-// func update(state *[7]uint64, msg []byte)
+// func update(state *macState, msg []byte)
 TEXT ·update(SB), $0-32
 	MOVQ state+0(FP), DI
 	MOVQ msg_base+8(FP), SI
 	MOVQ msg_len+16(FP), R15
-
-	MOVQ 0(DI), R8   // h0
-	MOVQ 8(DI), R9   // h1
-	MOVQ 16(DI), R10 // h2
-	MOVQ 24(DI), R11 // r0
-	MOVQ 32(DI), R12 // r1
-
-	CMPQ R15, $16
+	MOVQ (DI), R8
+	MOVQ 8(DI), R9
+	MOVQ 16(DI), R10
+	MOVQ 24(DI), R11
+	MOVQ 32(DI), R12
+	CMPQ R15, $0x10
 	JB   bytes_between_0_and_15
 
 loop:
-	POLY1305_ADD(SI, R8, R9, R10)
+	ADDQ (SI), R8
+	ADCQ 8(SI), R9
+	ADCQ $0x01, R10
+	LEAQ 16(SI), SI
 
 multiply:
-	POLY1305_MUL(R8, R9, R10, R11, R12, BX, CX, R13, R14)
-	SUBQ $16, R15
-	CMPQ R15, $16
-	JAE  loop
+	MOVQ  R11, AX
+	MULQ  R8
+	MOVQ  AX, BX
+	MOVQ  DX, CX
+	MOVQ  R11, AX
+	MULQ  R9
+	ADDQ  AX, CX
+	ADCQ  $0x00, DX
+	MOVQ  R11, R13
+	IMULQ R10, R13
+	ADDQ  DX, R13
+	MOVQ  R12, AX
+	MULQ  R8
+	ADDQ  AX, CX
+	ADCQ  $0x00, DX
+	MOVQ  DX, R8
+	MOVQ  R12, R14
+	IMULQ R10, R14
+	MOVQ  R12, AX
+	MULQ  R9
+	ADDQ  AX, R13
+	ADCQ  DX, R14
+	ADDQ  R8, R13
+	ADCQ  $0x00, R14
+	MOVQ  BX, R8
+	MOVQ  CX, R9
+	MOVQ  R13, R10
+	ANDQ  $0x03, R10
+	MOVQ  R13, BX
+	ANDQ  $-4, BX
+	ADDQ  BX, R8
+	ADCQ  R14, R9
+	ADCQ  $0x00, R10
+	SHRQ  $0x02, R14, R13
+	SHRQ  $0x02, R14
+	ADDQ  R13, R8
+	ADCQ  R14, R9
+	ADCQ  $0x00, R10
+	SUBQ  $0x10, R15
+	CMPQ  R15, $0x10
+	JAE   loop
 
 bytes_between_0_and_15:
 	TESTQ R15, R15
 	JZ    done
-	MOVQ  $1, BX
+	MOVQ  $0x00000001, BX
 	XORQ  CX, CX
 	XORQ  R13, R13
 	ADDQ  R15, SI
 
 flush_buffer:
-	SHLQ $8, BX, CX
-	SHLQ $8, BX
+	SHLQ $0x08, BX, CX
+	SHLQ $0x08, BX
 	MOVB -1(SI), R13
 	XORQ R13, BX
 	DECQ SI
 	DECQ R15
 	JNZ  flush_buffer
-
 	ADDQ BX, R8
 	ADCQ CX, R9
-	ADCQ $0, R10
-	MOVQ $16, R15
+	ADCQ $0x00, R10
+	MOVQ $0x00000010, R15
 	JMP  multiply
 
 done:
-	MOVQ R8, 0(DI)
+	MOVQ R8, (DI)
 	MOVQ R9, 8(DI)
 	MOVQ R10, 16(DI)
 	RET

internal/poly1305/sum_asm.go

@@ -2,7 +2,7 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
-//go:build gc && !purego
+//go:build gc && !purego && (amd64 || loong64 || ppc64 || ppc64le)
 
 package poly1305
 

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

internal/poly1305/sum_ppc64le.go

@@ -1,47 +0,0 @@
-// 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
-
-package poly1305
-
-//go:noescape
-func update(state *macState, msg []byte)
-
-// mac is a wrapper for macGeneric that redirects calls that would have gone to
-// updateGeneric to update.
-//
-// Its Write and Sum methods are otherwise identical to the macGeneric ones, but
-// using function pointers would carry a major performance cost.
-type mac struct{ macGeneric }
-
-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 < TagSize {
-			h.offset += n
-			return nn, nil
-		}
-		p = p[n:]
-		h.offset = 0
-		update(&h.macState, h.buffer[:])
-	}
-	if n := len(p) - (len(p) % TagSize); n > 0 {
-		update(&h.macState, p[:n])
-		p = p[n:]
-	}
-	if len(p) > 0 {
-		h.offset += copy(h.buffer[h.offset:], p)
-	}
-	return nn, nil
-}
-
-func (h *mac) Sum(out *[16]byte) {
-	state := h.macState
-	if h.offset > 0 {
-		update(&state, h.buffer[:h.offset])
-	}
-	finalize(out, &state.h, &state.s)
-}

internal/poly1305/sum_ppc64x.s

@@ -2,15 +2,25 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
-//go:build gc && !purego
+//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) \
-	MOVD (msg), t0;  \
-	MOVD 8(msg), t1; \
+	LE_MOVD (msg)( R0), t0; \
+	LE_MOVD (msg)(R24), t1; \
 	MOVD $1, t2;     \
 	ADDC t0, h0, h0; \
 	ADDE t1, h1, h1; \
@@ -50,10 +60,6 @@
 	ADDE   t3, h1, h1;  \
 	ADDZE  h2
 
-DATA ·poly1305Mask<>+0x00(SB)/8, $0x0FFFFFFC0FFFFFFF
-DATA ·poly1305Mask<>+0x08(SB)/8, $0x0FFFFFFC0FFFFFFC
-GLOBL ·poly1305Mask<>(SB), RODATA, $16
-
 // func update(state *[7]uint64, msg []byte)
 TEXT ·update(SB), $0-32
 	MOVD state+0(FP), R3
@@ -66,6 +72,8 @@ TEXT ·update(SB), $0-32
 	MOVD 24(R3), R11 // r0
 	MOVD 32(R3), R12 // r1
 
+	MOVD $8, R24
+
 	CMP R5, $16
 	BLT bytes_between_0_and_15
 
@@ -94,7 +102,7 @@ flush_buffer:
 
 	// Greater than 8 -- load the rightmost remaining bytes in msg
 	// and put into R17 (h1)
-	MOVD (R4)(R21), R17
+	LE_MOVD (R4)(R21), R17
 	MOVD $16, R22
 
 	// Find the offset to those bytes
@@ -118,7 +126,7 @@ just1:
 	BLT less8
 
 	// Exactly 8
-	MOVD (R4), R16
+	LE_MOVD (R4), R16
 
 	CMP R17, $0
 
@@ -133,7 +141,7 @@ less8:
 	MOVD  $0, R22   // shift count
 	CMP   R5, $4
 	BLT   less4
-	MOVWZ (R4), R16
+	LE_MOVWZ (R4), R16
 	ADD   $4, R4
 	ADD   $-4, R5
 	MOVD  $32, R22
@@ -141,7 +149,7 @@ less8:
 less4:
 	CMP   R5, $2
 	BLT   less2
-	MOVHZ (R4), R21
+	LE_MOVHZ (R4), R21
 	SLD   R22, R21, R21
 	OR    R16, R21, R16
 	ADD   $16, R22

openpgp/clearsign/clearsign_test.go

@@ -123,24 +123,12 @@ func TestSigning(t *testing.T) {
 	}
 }
 
-// We use this to make test keys, so that they aren't all the same.
-type quickRand byte
-
-func (qr *quickRand) Read(p []byte) (int, error) {
-	for i := range p {
-		p[i] = byte(*qr)
-	}
-	*qr++
-	return len(p), nil
-}
-
 func TestMultiSign(t *testing.T) {
 	if testing.Short() {
 		t.Skip("skipping long test in -short mode")
 	}
 
-	zero := quickRand(0)
-	config := packet.Config{Rand: &zero}
+	var config packet.Config
 
 	for nKeys := 0; nKeys < 4; nKeys++ {
 	nextTest:

pkcs12/crypto.go

@@ -26,7 +26,7 @@ type pbeCipher interface {
 	create(key []byte) (cipher.Block, error)
 	// deriveKey returns a key derived from the given password and salt.
 	deriveKey(salt, password []byte, iterations int) []byte
-	// deriveKey returns an IV derived from the given password and salt.
+	// deriveIV returns an IV derived from the given password and salt.
 	deriveIV(salt, password []byte, iterations int) []byte
 }
 

salsa20/salsa/_asm/go.mod

@@ -0,0 +1,14 @@
+module salsa20/salsa/_asm
+
+go 1.23
+
+require (
+	github.com/mmcloughlin/avo v0.6.0
+	golang.org/x/crypto v0.33.0
+)
+
+require (
+	golang.org/x/mod v0.20.0 // indirect
+	golang.org/x/sync v0.8.0 // indirect
+	golang.org/x/tools v0.24.0 // indirect
+)

salsa20/salsa/_asm/go.sum

@@ -0,0 +1,10 @@
+github.com/mmcloughlin/avo v0.6.0 h1:QH6FU8SKoTLaVs80GA8TJuLNkUYl4VokHKlPhVDg4YY=
+github.com/mmcloughlin/avo v0.6.0/go.mod h1:8CoAGaCSYXtCPR+8y18Y9aB/kxb8JSS6FRI7mSkvD+8=
+golang.org/x/crypto v0.33.0 h1:IOBPskki6Lysi0lo9qQvbxiQ+FvsCC/YWOecCHAixus=
+golang.org/x/crypto v0.33.0/go.mod h1:bVdXmD7IV/4GdElGPozy6U7lWdRXA4qyRVGJV57uQ5M=
+golang.org/x/mod v0.20.0 h1:utOm6MM3R3dnawAiJgn0y+xvuYRsm1RKM/4giyfDgV0=
+golang.org/x/mod v0.20.0/go.mod h1:hTbmBsO62+eylJbnUtE2MGJUyE7QWk4xUqPFrRgJ+7c=
+golang.org/x/sync v0.8.0 h1:3NFvSEYkUoMifnESzZl15y791HH1qU2xm6eCJU5ZPXQ=
+golang.org/x/sync v0.8.0/go.mod h1:Czt+wKu1gCyEFDUtn0jG5QVvpJ6rzVqr5aXyt9drQfk=
+golang.org/x/tools v0.24.0 h1:J1shsA93PJUEVaUSaay7UXAyE8aimq3GW0pjlolpa24=
+golang.org/x/tools v0.24.0/go.mod h1:YhNqVBIfWHdzvTLs0d8LCuMhkKUgSUKldakyV7W/WDQ=

salsa20/salsa/_asm/salsa20_amd64_asm.go

@@ -0,0 +1,932 @@
+// Copyright 2024 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 code was translated into a form compatible with 6a from the public
+// domain sources in SUPERCOP: https://bench.cr.yp.to/supercop.html
+
+package main
+
+import (
+	. "github.com/mmcloughlin/avo/build"
+	"github.com/mmcloughlin/avo/ir"
+	. "github.com/mmcloughlin/avo/operand"
+	. "github.com/mmcloughlin/avo/reg"
+	_ "golang.org/x/crypto/salsa20/salsa"
+)
+
+//go:generate go run . -out ../salsa20_amd64.s -pkg salsa
+
+func main() {
+	Package("golang.org/x/crypto/salsa20/salsa")
+	ConstraintExpr("amd64,!purego,gc")
+	salsa2020XORKeyStream()
+	Generate()
+}
+
+func salsa2020XORKeyStream() {
+	Implement("salsa2020XORKeyStream")
+	Attributes(0)
+	AllocLocal(456) // frame = 424 + 32 byte alignment
+	Comment("This needs up to 64 bytes at 360(R12); hence the non-obvious frame size.")
+
+	Load(Param("out"), RDI)
+	Load(Param("in"), RSI)
+	Load(Param("n"), RDX)
+	Load(Param("nonce"), RCX)
+	Load(Param("key"), R8)
+
+	MOVQ(RSP, R12)
+	ADDQ(Imm(31), R12)
+	ANDQ(I32(^31), R12)
+
+	MOVQ(RDX, R9)
+	MOVQ(RCX, RDX)
+	MOVQ(R8, R10)
+	CMPQ(R9, Imm(0))
+	JBE(LabelRef("DONE"))
+
+	START()
+	BYTESATLEAST256()
+	MAINLOOP1()
+	BYTESBETWEEN1AND255()
+	NOCOPY()
+	MAINLOOP2()
+
+	Label("BYTESATLEAST64")
+	Label("DONE")
+	RET()
+	Label("BYTESATLEAST65")
+	SUBQ(Imm(64), R9)
+	ADDQ(Imm(64), RDI)
+	ADDQ(Imm(64), RSI)
+	JMP(LabelRef("BYTESBETWEEN1AND255"))
+}
+
+func START() {
+	Label("START")
+	MOVL(Mem{Base: R10}.Offset(20), ECX)
+	MOVL(Mem{Base: R10}.Offset(0), R8L)
+	MOVL(Mem{Base: EDX}.Offset(0), EAX)
+	MOVL(Mem{Base: R10}.Offset(16), R11L)
+	MOVL(ECX, Mem{Base: R12}.Offset(0))
+	MOVL(R8L, Mem{Base: R12}.Offset(4))
+	MOVL(EAX, Mem{Base: R12}.Offset(8))
+	MOVL(R11L, Mem{Base: R12}.Offset(12))
+	MOVL(Mem{Base: EDX}.Offset(8), ECX)
+	MOVL(Mem{Base: R10}.Offset(24), R8L)
+	MOVL(Mem{Base: R10}.Offset(4), EAX)
+	MOVL(Mem{Base: EDX}.Offset(4), R11L)
+	MOVL(ECX, Mem{Base: R12}.Offset(16))
+	MOVL(R8L, Mem{Base: R12}.Offset(20))
+	MOVL(EAX, Mem{Base: R12}.Offset(24))
+	MOVL(R11L, Mem{Base: R12}.Offset(28))
+	MOVL(Mem{Base: EDX}.Offset(12), ECX)
+	MOVL(Mem{Base: R10}.Offset(12), EDX)
+	MOVL(Mem{Base: R10}.Offset(28), R8L)
+	MOVL(Mem{Base: R10}.Offset(8), EAX)
+	MOVL(EDX, Mem{Base: R12}.Offset(32))
+	MOVL(ECX, Mem{Base: R12}.Offset(36))
+	MOVL(R8L, Mem{Base: R12}.Offset(40))
+	MOVL(EAX, Mem{Base: R12}.Offset(44))
+	MOVQ(Imm(1634760805), RDX)
+	MOVQ(Imm(857760878), RCX)
+	MOVQ(Imm(2036477234), R8)
+	MOVQ(Imm(1797285236), RAX)
+	MOVL(EDX, Mem{Base: R12}.Offset(48))
+	MOVL(ECX, Mem{Base: R12}.Offset(52))
+	MOVL(R8L, Mem{Base: R12}.Offset(56))
+	MOVL(EAX, Mem{Base: R12}.Offset(60))
+	CMPQ(R9, U32(256))
+	JB(LabelRef("BYTESBETWEEN1AND255"))
+	MOVOA(Mem{Base: R12}.Offset(48), X0)
+	PSHUFL(Imm(0x55), X0, X1)
+	PSHUFL(Imm(0xAA), X0, X2)
+	PSHUFL(Imm(0xFF), X0, X3)
+	PSHUFL(Imm(0x00), X0, X0)
+	MOVOA(X1, Mem{Base: R12}.Offset(64))
+	MOVOA(X2, Mem{Base: R12}.Offset(80))
+	MOVOA(X3, Mem{Base: R12}.Offset(96))
+	MOVOA(X0, Mem{Base: R12}.Offset(112))
+	MOVOA(Mem{Base: R12}.Offset(0), X0)
+	PSHUFL(Imm(0xAA), X0, X1)
+	PSHUFL(Imm(0xFF), X0, X2)
+	PSHUFL(Imm(0x00), X0, X3)
+	PSHUFL(Imm(0x55), X0, X0)
+	MOVOA(X1, Mem{Base: R12}.Offset(128))
+	MOVOA(X2, Mem{Base: R12}.Offset(144))
+	MOVOA(X3, Mem{Base: R12}.Offset(160))
+	MOVOA(X0, Mem{Base: R12}.Offset(176))
+	MOVOA(Mem{Base: R12}.Offset(16), X0)
+	PSHUFL(Imm(0xFF), X0, X1)
+	PSHUFL(Imm(0x55), X0, X2)
+	PSHUFL(Imm(0xAA), X0, X0)
+	MOVOA(X1, Mem{Base: R12}.Offset(192))
+	MOVOA(X2, Mem{Base: R12}.Offset(208))
+	MOVOA(X0, Mem{Base: R12}.Offset(224))
+	MOVOA(Mem{Base: R12}.Offset(32), X0)
+	PSHUFL(Imm(0x00), X0, X1)
+	PSHUFL(Imm(0xAA), X0, X2)
+	PSHUFL(Imm(0xFF), X0, X0)
+	MOVOA(X1, Mem{Base: R12}.Offset(240))
+	MOVOA(X2, Mem{Base: R12}.Offset(256))
+	MOVOA(X0, Mem{Base: R12}.Offset(272))
+
+}
+
+func BYTESATLEAST256() {
+	Label("BYTESATLEAST256")
+	MOVL(Mem{Base: R12}.Offset(16), EDX)
+	MOVL(Mem{Base: R12}.Offset(36), ECX)
+	MOVL(EDX, Mem{Base: R12}.Offset(288))
+	MOVL(ECX, Mem{Base: R12}.Offset(304))
+	SHLQ(Imm(32), RCX)
+	ADDQ(RCX, RDX)
+	ADDQ(Imm(1), RDX)
+	MOVQ(RDX, RCX)
+	SHRQ(Imm(32), RCX)
+	MOVL(EDX, Mem{Base: R12}.Offset(292))
+	MOVL(ECX, Mem{Base: R12}.Offset(308))
+	ADDQ(Imm(1), RDX)
+	MOVQ(RDX, RCX)
+	SHRQ(Imm(32), RCX)
+	MOVL(EDX, Mem{Base: R12}.Offset(296))
+	MOVL(ECX, Mem{Base: R12}.Offset(312))
+	ADDQ(Imm(1), RDX)
+	MOVQ(RDX, RCX)
+	SHRQ(Imm(32), RCX)
+	MOVL(EDX, Mem{Base: R12}.Offset(300))
+	MOVL(ECX, Mem{Base: R12}.Offset(316))
+	ADDQ(Imm(1), RDX)
+	MOVQ(RDX, RCX)
+	SHRQ(Imm(32), RCX)
+	MOVL(EDX, Mem{Base: R12}.Offset(16))
+	MOVL(ECX, Mem{Base: R12}.Offset(36))
+	MOVQ(R9, Mem{Base: R12}.Offset(352))
+	MOVQ(U32(20), RDX)
+	MOVOA(Mem{Base: R12}.Offset(64), X0)
+	MOVOA(Mem{Base: R12}.Offset(80), X1)
+	MOVOA(Mem{Base: R12}.Offset(96), X2)
+	MOVOA(Mem{Base: R12}.Offset(256), X3)
+	MOVOA(Mem{Base: R12}.Offset(272), X4)
+	MOVOA(Mem{Base: R12}.Offset(128), X5)
+	MOVOA(Mem{Base: R12}.Offset(144), X6)
+	MOVOA(Mem{Base: R12}.Offset(176), X7)
+	MOVOA(Mem{Base: R12}.Offset(192), X8)
+	MOVOA(Mem{Base: R12}.Offset(208), X9)
+	MOVOA(Mem{Base: R12}.Offset(224), X10)
+	MOVOA(Mem{Base: R12}.Offset(304), X11)
+	MOVOA(Mem{Base: R12}.Offset(112), X12)
+	MOVOA(Mem{Base: R12}.Offset(160), X13)
+	MOVOA(Mem{Base: R12}.Offset(240), X14)
+	MOVOA(Mem{Base: R12}.Offset(288), X15)
+}
+
+func MAINLOOP1() {
+	Label("MAINLOOP1")
+	MOVOA(X1, Mem{Base: R12}.Offset(320))
+	MOVOA(X2, Mem{Base: R12}.Offset(336))
+	MOVOA(X13, X1)
+	PADDL(X12, X1)
+	MOVOA(X1, X2)
+	PSLLL(Imm(7), X1)
+	PXOR(X1, X14)
+	PSRLL(Imm(25), X2)
+	PXOR(X2, X14)
+	MOVOA(X7, X1)
+	PADDL(X0, X1)
+	MOVOA(X1, X2)
+	PSLLL(Imm(7), X1)
+	PXOR(X1, X11)
+	PSRLL(Imm(25), X2)
+	PXOR(X2, X11)
+	MOVOA(X12, X1)
+	PADDL(X14, X1)
+	MOVOA(X1, X2)
+	PSLLL(Imm(9), X1)
+	PXOR(X1, X15)
+	PSRLL(Imm(23), X2)
+	PXOR(X2, X15)
+	MOVOA(X0, X1)
+	PADDL(X11, X1)
+	MOVOA(X1, X2)
+	PSLLL(Imm(9), X1)
+	PXOR(X1, X9)
+	PSRLL(Imm(23), X2)
+	PXOR(X2, X9)
+	MOVOA(X14, X1)
+	PADDL(X15, X1)
+	MOVOA(X1, X2)
+	PSLLL(Imm(13), X1)
+	PXOR(X1, X13)
+	PSRLL(Imm(19), X2)
+	PXOR(X2, X13)
+	MOVOA(X11, X1)
+	PADDL(X9, X1)
+	MOVOA(X1, X2)
+	PSLLL(Imm(13), X1)
+	PXOR(X1, X7)
+	PSRLL(Imm(19), X2)
+	PXOR(X2, X7)
+	MOVOA(X15, X1)
+	PADDL(X13, X1)
+	MOVOA(X1, X2)
+	PSLLL(Imm(18), X1)
+	PXOR(X1, X12)
+	PSRLL(Imm(14), X2)
+	PXOR(X2, X12)
+	MOVOA(Mem{Base: R12}.Offset(320), X1)
+	MOVOA(X12, Mem{Base: R12}.Offset(320))
+	MOVOA(X9, X2)
+	PADDL(X7, X2)
+	MOVOA(X2, X12)
+	PSLLL(Imm(18), X2)
+	PXOR(X2, X0)
+	PSRLL(Imm(14), X12)
+	PXOR(X12, X0)
+	MOVOA(X5, X2)
+	PADDL(X1, X2)
+	MOVOA(X2, X12)
+	PSLLL(Imm(7), X2)
+	PXOR(X2, X3)
+	PSRLL(Imm(25), X12)
+	PXOR(X12, X3)
+	MOVOA(Mem{Base: R12}.Offset(336), X2)
+	MOVOA(X0, Mem{Base: R12}.Offset(336))
+	MOVOA(X6, X0)
+	PADDL(X2, X0)
+	MOVOA(X0, X12)
+	PSLLL(Imm(7), X0)
+	PXOR(X0, X4)
+	PSRLL(Imm(25), X12)
+	PXOR(X12, X4)
+	MOVOA(X1, X0)
+	PADDL(X3, X0)
+	MOVOA(X0, X12)
+	PSLLL(Imm(9), X0)
+	PXOR(X0, X10)
+	PSRLL(Imm(23), X12)
+	PXOR(X12, X10)
+	MOVOA(X2, X0)
+	PADDL(X4, X0)
+	MOVOA(X0, X12)
+	PSLLL(Imm(9), X0)
+	PXOR(X0, X8)
+	PSRLL(Imm(23), X12)
+	PXOR(X12, X8)
+	MOVOA(X3, X0)
+	PADDL(X10, X0)
+	MOVOA(X0, X12)
+	PSLLL(Imm(13), X0)
+	PXOR(X0, X5)
+	PSRLL(Imm(19), X12)
+	PXOR(X12, X5)
+	MOVOA(X4, X0)
+	PADDL(X8, X0)
+	MOVOA(X0, X12)
+	PSLLL(Imm(13), X0)
+	PXOR(X0, X6)
+	PSRLL(Imm(19), X12)
+	PXOR(X12, X6)
+	MOVOA(X10, X0)
+	PADDL(X5, X0)
+	MOVOA(X0, X12)
+	PSLLL(Imm(18), X0)
+	PXOR(X0, X1)
+	PSRLL(Imm(14), X12)
+	PXOR(X12, X1)
+	MOVOA(Mem{Base: R12}.Offset(320), X0)
+	MOVOA(X1, Mem{Base: R12}.Offset(320))
+	MOVOA(X4, X1)
+	PADDL(X0, X1)
+	MOVOA(X1, X12)
+	PSLLL(Imm(7), X1)
+	PXOR(X1, X7)
+	PSRLL(Imm(25), X12)
+	PXOR(X12, X7)
+	MOVOA(X8, X1)
+	PADDL(X6, X1)
+	MOVOA(X1, X12)
+	PSLLL(Imm(18), X1)
+	PXOR(X1, X2)
+	PSRLL(Imm(14), X12)
+	PXOR(X12, X2)
+	MOVOA(Mem{Base: R12}.Offset(336), X12)
+	MOVOA(X2, Mem{Base: R12}.Offset(336))
+	MOVOA(X14, X1)
+	PADDL(X12, X1)
+	MOVOA(X1, X2)
+	PSLLL(Imm(7), X1)
+	PXOR(X1, X5)
+	PSRLL(Imm(25), X2)
+	PXOR(X2, X5)
+	MOVOA(X0, X1)
+	PADDL(X7, X1)
+	MOVOA(X1, X2)
+	PSLLL(Imm(9), X1)
+	PXOR(X1, X10)
+	PSRLL(Imm(23), X2)
+	PXOR(X2, X10)
+	MOVOA(X12, X1)
+	PADDL(X5, X1)
+	MOVOA(X1, X2)
+	PSLLL(Imm(9), X1)
+	PXOR(X1, X8)
+	PSRLL(Imm(23), X2)
+	PXOR(X2, X8)
+	MOVOA(X7, X1)
+	PADDL(X10, X1)
+	MOVOA(X1, X2)
+	PSLLL(Imm(13), X1)
+	PXOR(X1, X4)
+	PSRLL(Imm(19), X2)
+	PXOR(X2, X4)
+	MOVOA(X5, X1)
+	PADDL(X8, X1)
+	MOVOA(X1, X2)
+	PSLLL(Imm(13), X1)
+	PXOR(X1, X14)
+	PSRLL(Imm(19), X2)
+	PXOR(X2, X14)
+	MOVOA(X10, X1)
+	PADDL(X4, X1)
+	MOVOA(X1, X2)
+	PSLLL(Imm(18), X1)
+	PXOR(X1, X0)
+	PSRLL(Imm(14), X2)
+	PXOR(X2, X0)
+	MOVOA(Mem{Base: R12}.Offset(320), X1)
+	MOVOA(X0, Mem{Base: R12}.Offset(320))
+	MOVOA(X8, X0)
+	PADDL(X14, X0)
+	MOVOA(X0, X2)
+	PSLLL(Imm(18), X0)
+	PXOR(X0, X12)
+	PSRLL(Imm(14), X2)
+	PXOR(X2, X12)
+	MOVOA(X11, X0)
+	PADDL(X1, X0)
+	MOVOA(X0, X2)
+	PSLLL(Imm(7), X0)
+	PXOR(X0, X6)
+	PSRLL(Imm(25), X2)
+	PXOR(X2, X6)
+	MOVOA(Mem{Base: R12}.Offset(336), X2)
+	MOVOA(X12, Mem{Base: R12}.Offset(336))
+	MOVOA(X3, X0)
+	PADDL(X2, X0)
+	MOVOA(X0, X12)
+	PSLLL(Imm(7), X0)
+	PXOR(X0, X13)
+	PSRLL(Imm(25), X12)
+	PXOR(X12, X13)
+	MOVOA(X1, X0)
+	PADDL(X6, X0)
+	MOVOA(X0, X12)
+	PSLLL(Imm(9), X0)
+	PXOR(X0, X15)
+	PSRLL(Imm(23), X12)
+	PXOR(X12, X15)
+	MOVOA(X2, X0)
+	PADDL(X13, X0)
+	MOVOA(X0, X12)
+	PSLLL(Imm(9), X0)
+	PXOR(X0, X9)
+	PSRLL(Imm(23), X12)
+	PXOR(X12, X9)
+	MOVOA(X6, X0)
+	PADDL(X15, X0)
+	MOVOA(X0, X12)
+	PSLLL(Imm(13), X0)
+	PXOR(X0, X11)
+	PSRLL(Imm(19), X12)
+	PXOR(X12, X11)
+	MOVOA(X13, X0)
+	PADDL(X9, X0)
+	MOVOA(X0, X12)
+	PSLLL(Imm(13), X0)
+	PXOR(X0, X3)
+	PSRLL(Imm(19), X12)
+	PXOR(X12, X3)
+	MOVOA(X15, X0)
+	PADDL(X11, X0)
+	MOVOA(X0, X12)
+	PSLLL(Imm(18), X0)
+	PXOR(X0, X1)
+	PSRLL(Imm(14), X12)
+	PXOR(X12, X1)
+	MOVOA(X9, X0)
+	PADDL(X3, X0)
+	MOVOA(X0, X12)
+	PSLLL(Imm(18), X0)
+	PXOR(X0, X2)
+	PSRLL(Imm(14), X12)
+	PXOR(X12, X2)
+	MOVOA(Mem{Base: R12}.Offset(320), X12)
+	MOVOA(Mem{Base: R12}.Offset(336), X0)
+	SUBQ(Imm(2), RDX)
+	JA(LabelRef("MAINLOOP1"))
+	PADDL(Mem{Base: R12}.Offset(112), X12)
+	PADDL(Mem{Base: R12}.Offset(176), X7)
+	PADDL(Mem{Base: R12}.Offset(224), X10)
+	PADDL(Mem{Base: R12}.Offset(272), X4)
+	MOVD(X12, EDX)
+	MOVD(X7, ECX)
+	MOVD(X10, R8)
+	MOVD(X4, R9)
+	PSHUFL(Imm(0x39), X12, X12)
+	PSHUFL(Imm(0x39), X7, X7)
+	PSHUFL(Imm(0x39), X10, X10)
+	PSHUFL(Imm(0x39), X4, X4)
+	XORL(Mem{Base: SI}.Offset(0), EDX)
+	XORL(Mem{Base: SI}.Offset(4), ECX)
+	XORL(Mem{Base: SI}.Offset(8), R8L)
+	XORL(Mem{Base: SI}.Offset(12), R9L)
+	MOVL(EDX, Mem{Base: DI}.Offset(0))
+	MOVL(ECX, Mem{Base: DI}.Offset(4))
+	MOVL(R8L, Mem{Base: DI}.Offset(8))
+	MOVL(R9L, Mem{Base: DI}.Offset(12))
+	MOVD(X12, EDX)
+	MOVD(X7, ECX)
+	MOVD(X10, R8)
+	MOVD(X4, R9)
+	PSHUFL(Imm(0x39), X12, X12)
+	PSHUFL(Imm(0x39), X7, X7)
+	PSHUFL(Imm(0x39), X10, X10)
+	PSHUFL(Imm(0x39), X4, X4)
+	XORL(Mem{Base: SI}.Offset(64), EDX)
+	XORL(Mem{Base: SI}.Offset(68), ECX)
+	XORL(Mem{Base: SI}.Offset(72), R8L)
+	XORL(Mem{Base: SI}.Offset(76), R9L)
+	MOVL(EDX, Mem{Base: DI}.Offset(64))
+	MOVL(ECX, Mem{Base: DI}.Offset(68))
+	MOVL(R8L, Mem{Base: DI}.Offset(72))
+	MOVL(R9L, Mem{Base: DI}.Offset(76))
+	MOVD(X12, EDX)
+	MOVD(X7, ECX)
+	MOVD(X10, R8)
+	MOVD(X4, R9)
+	PSHUFL(Imm(0x39), X12, X12)
+	PSHUFL(Imm(0x39), X7, X7)
+	PSHUFL(Imm(0x39), X10, X10)
+	PSHUFL(Imm(0x39), X4, X4)
+	XORL(Mem{Base: SI}.Offset(128), EDX)
+	XORL(Mem{Base: SI}.Offset(132), ECX)
+	XORL(Mem{Base: SI}.Offset(136), R8L)
+	XORL(Mem{Base: SI}.Offset(140), R9L)
+	MOVL(EDX, Mem{Base: DI}.Offset(128))
+	MOVL(ECX, Mem{Base: DI}.Offset(132))
+	MOVL(R8L, Mem{Base: DI}.Offset(136))
+	MOVL(R9L, Mem{Base: DI}.Offset(140))
+	MOVD(X12, EDX)
+	MOVD(X7, ECX)
+	MOVD(X10, R8)
+	MOVD(X4, R9)
+	XORL(Mem{Base: SI}.Offset(192), EDX)
+	XORL(Mem{Base: SI}.Offset(196), ECX)
+	XORL(Mem{Base: SI}.Offset(200), R8L)
+	XORL(Mem{Base: SI}.Offset(204), R9L)
+	MOVL(EDX, Mem{Base: DI}.Offset(192))
+	MOVL(ECX, Mem{Base: DI}.Offset(196))
+	MOVL(R8L, Mem{Base: DI}.Offset(200))
+	MOVL(R9L, Mem{Base: DI}.Offset(204))
+	PADDL(Mem{Base: R12}.Offset(240), X14)
+	PADDL(Mem{Base: R12}.Offset(64), X0)
+	PADDL(Mem{Base: R12}.Offset(128), X5)
+	PADDL(Mem{Base: R12}.Offset(192), X8)
+	MOVD(X14, EDX)
+	MOVD(X0, ECX)
+	MOVD(X5, R8)
+	MOVD(X8, R9)
+	PSHUFL(Imm(0x39), X14, X14)
+	PSHUFL(Imm(0x39), X0, X0)
+	PSHUFL(Imm(0x39), X5, X5)
+	PSHUFL(Imm(0x39), X8, X8)
+	XORL(Mem{Base: SI}.Offset(16), EDX)
+	XORL(Mem{Base: SI}.Offset(20), ECX)
+	XORL(Mem{Base: SI}.Offset(24), R8L)
+	XORL(Mem{Base: SI}.Offset(28), R9L)
+	MOVL(EDX, Mem{Base: DI}.Offset(16))
+	MOVL(ECX, Mem{Base: DI}.Offset(20))
+	MOVL(R8L, Mem{Base: DI}.Offset(24))
+	MOVL(R9L, Mem{Base: DI}.Offset(28))
+	MOVD(X14, EDX)
+	MOVD(X0, ECX)
+	MOVD(X5, R8)
+	MOVD(X8, R9)
+	PSHUFL(Imm(0x39), X14, X14)
+	PSHUFL(Imm(0x39), X0, X0)
+	PSHUFL(Imm(0x39), X5, X5)
+	PSHUFL(Imm(0x39), X8, X8)
+	XORL(Mem{Base: SI}.Offset(80), EDX)
+	XORL(Mem{Base: SI}.Offset(84), ECX)
+	XORL(Mem{Base: SI}.Offset(88), R8L)
+	XORL(Mem{Base: SI}.Offset(92), R9L)
+	MOVL(EDX, Mem{Base: DI}.Offset(80))
+	MOVL(ECX, Mem{Base: DI}.Offset(84))
+	MOVL(R8L, Mem{Base: DI}.Offset(88))
+	MOVL(R9L, Mem{Base: DI}.Offset(92))
+	MOVD(X14, EDX)
+	MOVD(X0, ECX)
+	MOVD(X5, R8)
+	MOVD(X8, R9)
+	PSHUFL(Imm(0x39), X14, X14)
+	PSHUFL(Imm(0x39), X0, X0)
+	PSHUFL(Imm(0x39), X5, X5)
+	PSHUFL(Imm(0x39), X8, X8)
+	XORL(Mem{Base: SI}.Offset(144), EDX)
+	XORL(Mem{Base: SI}.Offset(148), ECX)
+	XORL(Mem{Base: SI}.Offset(152), R8L)
+	XORL(Mem{Base: SI}.Offset(156), R9L)
+	MOVL(EDX, Mem{Base: DI}.Offset(144))
+	MOVL(ECX, Mem{Base: DI}.Offset(148))
+	MOVL(R8L, Mem{Base: DI}.Offset(152))
+	MOVL(R9L, Mem{Base: DI}.Offset(156))
+	MOVD(X14, EDX)
+	MOVD(X0, ECX)
+	MOVD(X5, R8)
+	MOVD(X8, R9)
+	XORL(Mem{Base: SI}.Offset(208), EDX)
+	XORL(Mem{Base: SI}.Offset(212), ECX)
+	XORL(Mem{Base: SI}.Offset(216), R8L)
+	XORL(Mem{Base: SI}.Offset(220), R9L)
+	MOVL(EDX, Mem{Base: DI}.Offset(208))
+	MOVL(ECX, Mem{Base: DI}.Offset(212))
+	MOVL(R8L, Mem{Base: DI}.Offset(216))
+	MOVL(R9L, Mem{Base: DI}.Offset(220))
+	PADDL(Mem{Base: R12}.Offset(288), X15)
+	PADDL(Mem{Base: R12}.Offset(304), X11)
+	PADDL(Mem{Base: R12}.Offset(80), X1)
+	PADDL(Mem{Base: R12}.Offset(144), X6)
+	MOVD(X15, EDX)
+	MOVD(X11, ECX)
+	MOVD(X1, R8)
+	MOVD(X6, R9)
+	PSHUFL(Imm(0x39), X15, X15)
+	PSHUFL(Imm(0x39), X11, X11)
+	PSHUFL(Imm(0x39), X1, X1)
+	PSHUFL(Imm(0x39), X6, X6)
+	XORL(Mem{Base: SI}.Offset(32), EDX)
+	XORL(Mem{Base: SI}.Offset(36), ECX)
+	XORL(Mem{Base: SI}.Offset(40), R8L)
+	XORL(Mem{Base: SI}.Offset(44), R9L)
+	MOVL(EDX, Mem{Base: DI}.Offset(32))
+	MOVL(ECX, Mem{Base: DI}.Offset(36))
+	MOVL(R8L, Mem{Base: DI}.Offset(40))
+	MOVL(R9L, Mem{Base: DI}.Offset(44))
+	MOVD(X15, EDX)
+	MOVD(X11, ECX)
+	MOVD(X1, R8)
+	MOVD(X6, R9)
+	PSHUFL(Imm(0x39), X15, X15)
+	PSHUFL(Imm(0x39), X11, X11)
+	PSHUFL(Imm(0x39), X1, X1)
+	PSHUFL(Imm(0x39), X6, X6)
+	XORL(Mem{Base: SI}.Offset(96), EDX)
+	XORL(Mem{Base: SI}.Offset(100), ECX)
+	XORL(Mem{Base: SI}.Offset(104), R8L)
+	XORL(Mem{Base: SI}.Offset(108), R9L)
+	MOVL(EDX, Mem{Base: DI}.Offset(96))
+	MOVL(ECX, Mem{Base: DI}.Offset(100))
+	MOVL(R8L, Mem{Base: DI}.Offset(104))
+	MOVL(R9L, Mem{Base: DI}.Offset(108))
+	MOVD(X15, EDX)
+	MOVD(X11, ECX)
+	MOVD(X1, R8)
+	MOVD(X6, R9)
+	PSHUFL(Imm(0x39), X15, X15)
+	PSHUFL(Imm(0x39), X11, X11)
+	PSHUFL(Imm(0x39), X1, X1)
+	PSHUFL(Imm(0x39), X6, X6)
+	XORL(Mem{Base: SI}.Offset(160), EDX)
+	XORL(Mem{Base: SI}.Offset(164), ECX)
+	XORL(Mem{Base: SI}.Offset(168), R8L)
+	XORL(Mem{Base: SI}.Offset(172), R9L)
+	MOVL(EDX, Mem{Base: DI}.Offset(160))
+	MOVL(ECX, Mem{Base: DI}.Offset(164))
+	MOVL(R8L, Mem{Base: DI}.Offset(168))
+	MOVL(R9L, Mem{Base: DI}.Offset(172))
+	MOVD(X15, EDX)
+	MOVD(X11, ECX)
+	MOVD(X1, R8)
+	MOVD(X6, R9)
+	XORL(Mem{Base: SI}.Offset(224), EDX)
+	XORL(Mem{Base: SI}.Offset(228), ECX)
+	XORL(Mem{Base: SI}.Offset(232), R8L)
+	XORL(Mem{Base: SI}.Offset(236), R9L)
+	MOVL(EDX, Mem{Base: DI}.Offset(224))
+	MOVL(ECX, Mem{Base: DI}.Offset(228))
+	MOVL(R8L, Mem{Base: DI}.Offset(232))
+	MOVL(R9L, Mem{Base: DI}.Offset(236))
+	PADDL(Mem{Base: R12}.Offset(160), X13)
+	PADDL(Mem{Base: R12}.Offset(208), X9)
+	PADDL(Mem{Base: R12}.Offset(256), X3)
+	PADDL(Mem{Base: R12}.Offset(96), X2)
+	MOVD(X13, EDX)
+	MOVD(X9, ECX)
+	MOVD(X3, R8)
+	MOVD(X2, R9)
+	PSHUFL(Imm(0x39), X13, X13)
+	PSHUFL(Imm(0x39), X9, X9)
+	PSHUFL(Imm(0x39), X3, X3)
+	PSHUFL(Imm(0x39), X2, X2)
+	XORL(Mem{Base: SI}.Offset(48), EDX)
+	XORL(Mem{Base: SI}.Offset(52), ECX)
+	XORL(Mem{Base: SI}.Offset(56), R8L)
+	XORL(Mem{Base: SI}.Offset(60), R9L)
+	MOVL(EDX, Mem{Base: DI}.Offset(48))
+	MOVL(ECX, Mem{Base: DI}.Offset(52))
+	MOVL(R8L, Mem{Base: DI}.Offset(56))
+	MOVL(R9L, Mem{Base: DI}.Offset(60))
+	MOVD(X13, EDX)
+	MOVD(X9, ECX)
+	MOVD(X3, R8)
+	MOVD(X2, R9)
+	PSHUFL(Imm(0x39), X13, X13)
+	PSHUFL(Imm(0x39), X9, X9)
+	PSHUFL(Imm(0x39), X3, X3)
+	PSHUFL(Imm(0x39), X2, X2)
+	XORL(Mem{Base: SI}.Offset(112), EDX)
+	XORL(Mem{Base: SI}.Offset(116), ECX)
+	XORL(Mem{Base: SI}.Offset(120), R8L)
+	XORL(Mem{Base: SI}.Offset(124), R9L)
+	MOVL(EDX, Mem{Base: DI}.Offset(112))
+	MOVL(ECX, Mem{Base: DI}.Offset(116))
+	MOVL(R8L, Mem{Base: DI}.Offset(120))
+	MOVL(R9L, Mem{Base: DI}.Offset(124))
+	MOVD(X13, EDX)
+	MOVD(X9, ECX)
+	MOVD(X3, R8)
+	MOVD(X2, R9)
+	PSHUFL(Imm(0x39), X13, X13)
+	PSHUFL(Imm(0x39), X9, X9)
+	PSHUFL(Imm(0x39), X3, X3)
+	PSHUFL(Imm(0x39), X2, X2)
+	XORL(Mem{Base: SI}.Offset(176), EDX)
+	XORL(Mem{Base: SI}.Offset(180), ECX)
+	XORL(Mem{Base: SI}.Offset(184), R8L)
+	XORL(Mem{Base: SI}.Offset(188), R9L)
+	MOVL(EDX, Mem{Base: DI}.Offset(176))
+	MOVL(ECX, Mem{Base: DI}.Offset(180))
+	MOVL(R8L, Mem{Base: DI}.Offset(184))
+	MOVL(R9L, Mem{Base: DI}.Offset(188))
+	MOVD(X13, EDX)
+	MOVD(X9, ECX)
+	MOVD(X3, R8)
+	MOVD(X2, R9)
+	XORL(Mem{Base: SI}.Offset(240), EDX)
+	XORL(Mem{Base: SI}.Offset(244), ECX)
+	XORL(Mem{Base: SI}.Offset(248), R8L)
+	XORL(Mem{Base: SI}.Offset(252), R9L)
+	MOVL(EDX, Mem{Base: DI}.Offset(240))
+	MOVL(ECX, Mem{Base: DI}.Offset(244))
+	MOVL(R8L, Mem{Base: DI}.Offset(248))
+	MOVL(R9L, Mem{Base: DI}.Offset(252))
+	MOVQ(Mem{Base: R12}.Offset(352), R9)
+	SUBQ(U32(256), R9)
+	ADDQ(U32(256), RSI)
+	ADDQ(U32(256), RDI)
+	CMPQ(R9, U32(256))
+	JAE(LabelRef("BYTESATLEAST256"))
+	CMPQ(R9, Imm(0))
+	JBE(LabelRef("DONE"))
+}
+
+func BYTESBETWEEN1AND255() {
+	Label("BYTESBETWEEN1AND255")
+	CMPQ(R9, Imm(64))
+	JAE(LabelRef("NOCOPY"))
+	MOVQ(RDI, RDX)
+	LEAQ(Mem{Base: R12}.Offset(360), RDI)
+	MOVQ(R9, RCX)
+	// Hack to get Avo to emit:
+	// 	REP; MOVSB
+	Instruction(&ir.Instruction{Opcode: "REP; MOVSB"})
+	LEAQ(Mem{Base: R12}.Offset(360), RDI)
+	LEAQ(Mem{Base: R12}.Offset(360), RSI)
+}
+
+func NOCOPY() {
+	Label("NOCOPY")
+	MOVQ(R9, Mem{Base: R12}.Offset(352))
+	MOVOA(Mem{Base: R12}.Offset(48), X0)
+	MOVOA(Mem{Base: R12}.Offset(0), X1)
+	MOVOA(Mem{Base: R12}.Offset(16), X2)
+	MOVOA(Mem{Base: R12}.Offset(32), X3)
+	MOVOA(X1, X4)
+	MOVQ(U32(20), RCX)
+}
+
+func MAINLOOP2() {
+	Label("MAINLOOP2")
+	PADDL(X0, X4)
+	MOVOA(X0, X5)
+	MOVOA(X4, X6)
+	PSLLL(Imm(7), X4)
+	PSRLL(Imm(25), X6)
+	PXOR(X4, X3)
+	PXOR(X6, X3)
+	PADDL(X3, X5)
+	MOVOA(X3, X4)
+	MOVOA(X5, X6)
+	PSLLL(Imm(9), X5)
+	PSRLL(Imm(23), X6)
+	PXOR(X5, X2)
+	PSHUFL(Imm(0x93), X3, X3)
+	PXOR(X6, X2)
+	PADDL(X2, X4)
+	MOVOA(X2, X5)
+	MOVOA(X4, X6)
+	PSLLL(Imm(13), X4)
+	PSRLL(Imm(19), X6)
+	PXOR(X4, X1)
+	PSHUFL(Imm(0x4E), X2, X2)
+	PXOR(X6, X1)
+	PADDL(X1, X5)
+	MOVOA(X3, X4)
+	MOVOA(X5, X6)
+	PSLLL(Imm(18), X5)
+	PSRLL(Imm(14), X6)
+	PXOR(X5, X0)
+	PSHUFL(Imm(0x39), X1, X1)
+	PXOR(X6, X0)
+	PADDL(X0, X4)
+	MOVOA(X0, X5)
+	MOVOA(X4, X6)
+	PSLLL(Imm(7), X4)
+	PSRLL(Imm(25), X6)
+	PXOR(X4, X1)
+	PXOR(X6, X1)
+	PADDL(X1, X5)
+	MOVOA(X1, X4)
+	MOVOA(X5, X6)
+	PSLLL(Imm(9), X5)
+	PSRLL(Imm(23), X6)
+	PXOR(X5, X2)
+	PSHUFL(Imm(0x93), X1, X1)
+	PXOR(X6, X2)
+	PADDL(X2, X4)
+	MOVOA(X2, X5)
+	MOVOA(X4, X6)
+	PSLLL(Imm(13), X4)
+	PSRLL(Imm(19), X6)
+	PXOR(X4, X3)
+	PSHUFL(Imm(0x4E), X2, X2)
+	PXOR(X6, X3)
+	PADDL(X3, X5)
+	MOVOA(X1, X4)
+	MOVOA(X5, X6)
+	PSLLL(Imm(18), X5)
+	PSRLL(Imm(14), X6)
+	PXOR(X5, X0)
+	PSHUFL(Imm(0x39), X3, X3)
+	PXOR(X6, X0)
+	PADDL(X0, X4)
+	MOVOA(X0, X5)
+	MOVOA(X4, X6)
+	PSLLL(Imm(7), X4)
+	PSRLL(Imm(25), X6)
+	PXOR(X4, X3)
+	PXOR(X6, X3)
+	PADDL(X3, X5)
+	MOVOA(X3, X4)
+	MOVOA(X5, X6)
+	PSLLL(Imm(9), X5)
+	PSRLL(Imm(23), X6)
+	PXOR(X5, X2)
+	PSHUFL(Imm(0x93), X3, X3)
+	PXOR(X6, X2)
+	PADDL(X2, X4)
+	MOVOA(X2, X5)
+	MOVOA(X4, X6)
+	PSLLL(Imm(13), X4)
+	PSRLL(Imm(19), X6)
+	PXOR(X4, X1)
+	PSHUFL(Imm(0x4E), X2, X2)
+	PXOR(X6, X1)
+	PADDL(X1, X5)
+	MOVOA(X3, X4)
+	MOVOA(X5, X6)
+	PSLLL(Imm(18), X5)
+	PSRLL(Imm(14), X6)
+	PXOR(X5, X0)
+	PSHUFL(Imm(0x39), X1, X1)
+	PXOR(X6, X0)
+	PADDL(X0, X4)
+	MOVOA(X0, X5)
+	MOVOA(X4, X6)
+	PSLLL(Imm(7), X4)
+	PSRLL(Imm(25), X6)
+	PXOR(X4, X1)
+	PXOR(X6, X1)
+	PADDL(X1, X5)
+	MOVOA(X1, X4)
+	MOVOA(X5, X6)
+	PSLLL(Imm(9), X5)
+	PSRLL(Imm(23), X6)
+	PXOR(X5, X2)
+	PSHUFL(Imm(0x93), X1, X1)
+	PXOR(X6, X2)
+	PADDL(X2, X4)
+	MOVOA(X2, X5)
+	MOVOA(X4, X6)
+	PSLLL(Imm(13), X4)
+	PSRLL(Imm(19), X6)
+	PXOR(X4, X3)
+	PSHUFL(Imm(0x4E), X2, X2)
+	PXOR(X6, X3)
+	SUBQ(Imm(4), RCX)
+	PADDL(X3, X5)
+	MOVOA(X1, X4)
+	MOVOA(X5, X6)
+	PSLLL(Imm(18), X5)
+	PXOR(X7, X7)
+	PSRLL(Imm(14), X6)
+	PXOR(X5, X0)
+	PSHUFL(Imm(0x39), X3, X3)
+	PXOR(X6, X0)
+	JA(LabelRef("MAINLOOP2"))
+	PADDL(Mem{Base: R12}.Offset(48), X0)
+	PADDL(Mem{Base: R12}.Offset(0), X1)
+	PADDL(Mem{Base: R12}.Offset(16), X2)
+	PADDL(Mem{Base: R12}.Offset(32), X3)
+	MOVD(X0, ECX)
+	MOVD(X1, R8)
+	MOVD(X2, R9)
+	MOVD(X3, EAX)
+	PSHUFL(Imm(0x39), X0, X0)
+	PSHUFL(Imm(0x39), X1, X1)
+	PSHUFL(Imm(0x39), X2, X2)
+	PSHUFL(Imm(0x39), X3, X3)
+	XORL(Mem{Base: SI}.Offset(0), ECX)
+	XORL(Mem{Base: SI}.Offset(48), R8L)
+	XORL(Mem{Base: SI}.Offset(32), R9L)
+	XORL(Mem{Base: SI}.Offset(16), EAX)
+	MOVL(ECX, Mem{Base: DI}.Offset(0))
+	MOVL(R8L, Mem{Base: DI}.Offset(48))
+	MOVL(R9L, Mem{Base: DI}.Offset(32))
+	MOVL(EAX, Mem{Base: DI}.Offset(16))
+	MOVD(X0, ECX)
+	MOVD(X1, R8)
+	MOVD(X2, R9)
+	MOVD(X3, EAX)
+	PSHUFL(Imm(0x39), X0, X0)
+	PSHUFL(Imm(0x39), X1, X1)
+	PSHUFL(Imm(0x39), X2, X2)
+	PSHUFL(Imm(0x39), X3, X3)
+	XORL(Mem{Base: SI}.Offset(20), ECX)
+	XORL(Mem{Base: SI}.Offset(4), R8L)
+	XORL(Mem{Base: SI}.Offset(52), R9L)
+	XORL(Mem{Base: SI}.Offset(36), EAX)
+	MOVL(ECX, Mem{Base: DI}.Offset(20))
+	MOVL(R8L, Mem{Base: DI}.Offset(4))
+	MOVL(R9L, Mem{Base: DI}.Offset(52))
+	MOVL(EAX, Mem{Base: DI}.Offset(36))
+	MOVD(X0, ECX)
+	MOVD(X1, R8)
+	MOVD(X2, R9)
+	MOVD(X3, EAX)
+	PSHUFL(Imm(0x39), X0, X0)
+	PSHUFL(Imm(0x39), X1, X1)
+	PSHUFL(Imm(0x39), X2, X2)
+	PSHUFL(Imm(0x39), X3, X3)
+	XORL(Mem{Base: SI}.Offset(40), ECX)
+	XORL(Mem{Base: SI}.Offset(24), R8L)
+	XORL(Mem{Base: SI}.Offset(8), R9L)
+	XORL(Mem{Base: SI}.Offset(56), EAX)
+	MOVL(ECX, Mem{Base: DI}.Offset(40))
+	MOVL(R8L, Mem{Base: DI}.Offset(24))
+	MOVL(R9L, Mem{Base: DI}.Offset(8))
+	MOVL(EAX, Mem{Base: DI}.Offset(56))
+	MOVD(X0, ECX)
+	MOVD(X1, R8)
+	MOVD(X2, R9)
+	MOVD(X3, EAX)
+	XORL(Mem{Base: SI}.Offset(60), ECX)
+	XORL(Mem{Base: SI}.Offset(44), R8L)
+	XORL(Mem{Base: SI}.Offset(28), R9L)
+	XORL(Mem{Base: SI}.Offset(12), EAX)
+	MOVL(ECX, Mem{Base: DI}.Offset(60))
+	MOVL(R8L, Mem{Base: DI}.Offset(44))
+	MOVL(R9L, Mem{Base: DI}.Offset(28))
+	MOVL(EAX, Mem{Base: DI}.Offset(12))
+	MOVQ(Mem{Base: R12}.Offset(352), R9)
+	MOVL(Mem{Base: R12}.Offset(16), ECX)
+	MOVL(Mem{Base: R12}.Offset(36), R8L)
+	ADDQ(Imm(1), RCX)
+	SHLQ(Imm(32), R8)
+	ADDQ(R8, RCX)
+	MOVQ(RCX, R8)
+	SHRQ(Imm(32), R8)
+	MOVL(ECX, Mem{Base: R12}.Offset(16))
+	MOVL(R8L, Mem{Base: R12}.Offset(36))
+	CMPQ(R9, Imm(64))
+	JA(LabelRef("BYTESATLEAST65"))
+	JAE(LabelRef("BYTESATLEAST64"))
+	MOVQ(RDI, RSI)
+	MOVQ(RDX, RDI)
+	MOVQ(R9, RCX)
+	// Hack to get Avo to emit:
+	// 	REP; MOVSB
+	Instruction(&ir.Instruction{Opcode: "REP; MOVSB"})
+}

sha3/_asm/go.mod

@@ -4,12 +4,12 @@ go 1.22
 
 require (
 	github.com/mmcloughlin/avo v0.6.0
-	golang.org/x/crypto v0.25.0
+	golang.org/x/crypto v0.33.0
 )
 
 require (
 	golang.org/x/mod v0.19.0 // indirect
 	golang.org/x/sync v0.7.0 // indirect
-	golang.org/x/sys v0.22.0 // indirect
+	golang.org/x/sys v0.30.0 // indirect
 	golang.org/x/tools v0.23.0 // indirect
 )

sha3/_asm/go.sum

@@ -1,12 +1,12 @@
 github.com/mmcloughlin/avo v0.6.0 h1:QH6FU8SKoTLaVs80GA8TJuLNkUYl4VokHKlPhVDg4YY=
 github.com/mmcloughlin/avo v0.6.0/go.mod h1:8CoAGaCSYXtCPR+8y18Y9aB/kxb8JSS6FRI7mSkvD+8=
-golang.org/x/crypto v0.25.0 h1:ypSNr+bnYL2YhwoMt2zPxHFmbAN1KZs/njMG3hxUp30=
-golang.org/x/crypto v0.25.0/go.mod h1:T+wALwcMOSE0kXgUAnPAHqTLW+XHgcELELW8VaDgm/M=
+golang.org/x/crypto v0.33.0 h1:IOBPskki6Lysi0lo9qQvbxiQ+FvsCC/YWOecCHAixus=
+golang.org/x/crypto v0.33.0/go.mod h1:bVdXmD7IV/4GdElGPozy6U7lWdRXA4qyRVGJV57uQ5M=
 golang.org/x/mod v0.19.0 h1:fEdghXQSo20giMthA7cd28ZC+jts4amQ3YMXiP5oMQ8=
 golang.org/x/mod v0.19.0/go.mod h1:hTbmBsO62+eylJbnUtE2MGJUyE7QWk4xUqPFrRgJ+7c=
 golang.org/x/sync v0.7.0 h1:YsImfSBoP9QPYL0xyKJPq0gcaJdG3rInoqxTWbfQu9M=
 golang.org/x/sync v0.7.0/go.mod h1:Czt+wKu1gCyEFDUtn0jG5QVvpJ6rzVqr5aXyt9drQfk=
-golang.org/x/sys v0.22.0 h1:RI27ohtqKCnwULzJLqkv897zojh5/DwS/ENaMzUOaWI=
-golang.org/x/sys v0.22.0/go.mod h1:/VUhepiaJMQUp4+oa/7Zr1D23ma6VTLIYjOOTFZPUcA=
+golang.org/x/sys v0.30.0 h1:QjkSwP/36a20jFYWkSue1YwXzLmsV5Gfq7Eiy72C1uc=
+golang.org/x/sys v0.30.0/go.mod h1:/VUhepiaJMQUp4+oa/7Zr1D23ma6VTLIYjOOTFZPUcA=
 golang.org/x/tools v0.23.0 h1:SGsXPZ+2l4JsgaCKkx+FQ9YZ5XEtA1GZYuoDjenLjvg=
 golang.org/x/tools v0.23.0/go.mod h1:pnu6ufv6vQkll6szChhK3C3L/ruaIv5eBeztNG8wtsI=

sha3/doc.go

@@ -5,6 +5,10 @@
 // Package sha3 implements the SHA-3 fixed-output-length hash functions and
 // the SHAKE variable-output-length hash functions defined by FIPS-202.
 //
+// All types in this package also implement [encoding.BinaryMarshaler],
+// [encoding.BinaryAppender] and [encoding.BinaryUnmarshaler] to marshal and
+// unmarshal the internal state of the hash.
+//
 // Both types of hash function use the "sponge" construction and the Keccak
 // permutation. For a detailed specification see http://keccak.noekeon.org/
 //

sha3/hashes.go

@@ -48,33 +48,52 @@ func init() {
 	crypto.RegisterHash(crypto.SHA3_512, New512)
 }
 
+const (
+	dsbyteSHA3   = 0b00000110
+	dsbyteKeccak = 0b00000001
+	dsbyteShake  = 0b00011111
+	dsbyteCShake = 0b00000100
+
+	// rateK[c] is the rate in bytes for Keccak[c] where c is the capacity in
+	// bits. Given the sponge size is 1600 bits, the rate is 1600 - c bits.
+	rateK256  = (1600 - 256) / 8
+	rateK448  = (1600 - 448) / 8
+	rateK512  = (1600 - 512) / 8
+	rateK768  = (1600 - 768) / 8
+	rateK1024 = (1600 - 1024) / 8
+)
+
 func new224Generic() *state {
-	return &state{rate: 144, outputLen: 28, dsbyte: 0x06}
+	return &state{rate: rateK448, outputLen: 28, dsbyte: dsbyteSHA3}
 }
 
 func new256Generic() *state {
-	return &state{rate: 136, outputLen: 32, dsbyte: 0x06}
+	return &state{rate: rateK512, outputLen: 32, dsbyte: dsbyteSHA3}
 }
 
 func new384Generic() *state {
-	return &state{rate: 104, outputLen: 48, dsbyte: 0x06}
+	return &state{rate: rateK768, outputLen: 48, dsbyte: dsbyteSHA3}
 }
 
 func new512Generic() *state {
-	return &state{rate: 72, outputLen: 64, dsbyte: 0x06}
+	return &state{rate: rateK1024, outputLen: 64, dsbyte: dsbyteSHA3}
 }
 
 // NewLegacyKeccak256 creates a new Keccak-256 hash.
 //
 // Only use this function if you require compatibility with an existing cryptosystem
 // that uses non-standard padding. All other users should use New256 instead.
-func NewLegacyKeccak256() hash.Hash { return &state{rate: 136, outputLen: 32, dsbyte: 0x01} }
+func NewLegacyKeccak256() hash.Hash {
+	return &state{rate: rateK512, outputLen: 32, dsbyte: dsbyteKeccak}
+}
 
 // NewLegacyKeccak512 creates a new Keccak-512 hash.
 //
 // Only use this function if you require compatibility with an existing cryptosystem
 // that uses non-standard padding. All other users should use New512 instead.
-func NewLegacyKeccak512() hash.Hash { return &state{rate: 72, outputLen: 64, dsbyte: 0x01} }
+func NewLegacyKeccak512() hash.Hash {
+	return &state{rate: rateK1024, outputLen: 64, dsbyte: dsbyteKeccak}
+}
 
 // Sum224 returns the SHA3-224 digest of the data.
 func Sum224(data []byte) (digest [28]byte) {

sha3/sha3.go

@@ -4,6 +4,15 @@
 
 package sha3
 
+import (
+	"crypto/subtle"
+	"encoding/binary"
+	"errors"
+	"unsafe"
+
+	"golang.org/x/sys/cpu"
+)
+
 // spongeDirection indicates the direction bytes are flowing through the sponge.
 type spongeDirection int
 
@@ -14,16 +23,13 @@ const (
 	spongeSqueezing
 )
 
-const (
-	// maxRate is the maximum size of the internal buffer. SHAKE-256
-	// currently needs the largest buffer.
-	maxRate = 168
-)
-
 type state struct {
-	// Generic sponge components.
-	a    [25]uint64 // main state of the hash
-	rate int        // the number of bytes of state to use
+	a [1600 / 8]byte // main state of the hash
+
+	// a[n:rate] is the buffer. If absorbing, it's the remaining space to XOR
+	// into before running the permutation. If squeezing, it's the remaining
+	// output to produce before running the permutation.
+	n, rate int
 
 	// dsbyte contains the "domain separation" bits and the first bit of
 	// the padding. Sections 6.1 and 6.2 of [1] separate the outputs of the
@@ -39,10 +45,6 @@ type state struct {
 	//      Extendable-Output Functions (May 2014)"
 	dsbyte byte
 
-	i, n    int // storage[i:n] is the buffer, i is only used while squeezing
-	storage [maxRate]byte
-
-	// Specific to SHA-3 and SHAKE.
 	outputLen int             // the default output size in bytes
 	state     spongeDirection // whether the sponge is absorbing or squeezing
 }
@@ -61,7 +63,7 @@ func (d *state) Reset() {
 		d.a[i] = 0
 	}
 	d.state = spongeAbsorbing
-	d.i, d.n = 0, 0
+	d.n = 0
 }
 
 func (d *state) clone() *state {
@@ -69,22 +71,25 @@ func (d *state) clone() *state {
 	return &ret
 }
 
-// permute applies the KeccakF-1600 permutation. It handles
-// any input-output buffering.
+// permute applies the KeccakF-1600 permutation.
 func (d *state) permute() {
-	switch d.state {
-	case spongeAbsorbing:
-		// If we're absorbing, we need to xor the input into the state
-		// before applying the permutation.
-		xorIn(d, d.storage[:d.rate])
-		d.n = 0
-		keccakF1600(&d.a)
-	case spongeSqueezing:
-		// If we're squeezing, we need to apply the permutation before
-		// copying more output.
-		keccakF1600(&d.a)
-		d.i = 0
-		copyOut(d, d.storage[:d.rate])
+	var a *[25]uint64
+	if cpu.IsBigEndian {
+		a = new([25]uint64)
+		for i := range a {
+			a[i] = binary.LittleEndian.Uint64(d.a[i*8:])
+		}
+	} else {
+		a = (*[25]uint64)(unsafe.Pointer(&d.a))
+	}
+
+	keccakF1600(a)
+	d.n = 0
+
+	if cpu.IsBigEndian {
+		for i := range a {
+			binary.LittleEndian.PutUint64(d.a[i*8:], a[i])
+		}
 	}
 }
 
@@ -92,53 +97,36 @@ func (d *state) permute() {
 // the multi-bitrate 10..1 padding rule, and permutes the state.
 func (d *state) padAndPermute() {
 	// Pad with this instance's domain-separator bits. We know that there's
-	// at least one byte of space in d.buf because, if it were full,
+	// at least one byte of space in the sponge because, if it were full,
 	// permute would have been called to empty it. dsbyte also contains the
 	// first one bit for the padding. See the comment in the state struct.
-	d.storage[d.n] = d.dsbyte
-	d.n++
-	for d.n < d.rate {
-		d.storage[d.n] = 0
-		d.n++
-	}
+	d.a[d.n] ^= d.dsbyte
 	// This adds the final one bit for the padding. Because of the way that
 	// bits are numbered from the LSB upwards, the final bit is the MSB of
 	// the last byte.
-	d.storage[d.rate-1] ^= 0x80
+	d.a[d.rate-1] ^= 0x80
 	// Apply the permutation
 	d.permute()
 	d.state = spongeSqueezing
-	d.n = d.rate
-	copyOut(d, d.storage[:d.rate])
 }
 
 // Write absorbs more data into the hash's state. It panics if any
 // output has already been read.
-func (d *state) Write(p []byte) (written int, err error) {
+func (d *state) Write(p []byte) (n int, err error) {
 	if d.state != spongeAbsorbing {
 		panic("sha3: Write after Read")
 	}
-	written = len(p)
+
+	n = len(p)
 
 	for len(p) > 0 {
-		if d.n == 0 && len(p) >= d.rate {
-			// The fast path; absorb a full "rate" bytes of input and apply the permutation.
-			xorIn(d, p[:d.rate])
-			p = p[d.rate:]
-			keccakF1600(&d.a)
-		} else {
-			// The slow path; buffer the input until we can fill the sponge, and then xor it in.
-			todo := d.rate - d.n
-			if todo > len(p) {
-				todo = len(p)
-			}
-			d.n += copy(d.storage[d.n:], p[:todo])
-			p = p[todo:]
+		x := subtle.XORBytes(d.a[d.n:d.rate], d.a[d.n:d.rate], p)
+		d.n += x
+		p = p[x:]
 
-			// If the sponge is full, apply the permutation.
-			if d.n == d.rate {
-				d.permute()
-			}
+		// If the sponge is full, apply the permutation.
+		if d.n == d.rate {
+			d.permute()
 		}
 	}
 
@@ -156,14 +144,14 @@ func (d *state) Read(out []byte) (n int, err error) {
 
 	// Now, do the squeezing.
 	for len(out) > 0 {
-		n := copy(out, d.storage[d.i:d.n])
-		d.i += n
-		out = out[n:]
-
 		// Apply the permutation if we've squeezed the sponge dry.
-		if d.i == d.rate {
+		if d.n == d.rate {
 			d.permute()
 		}
+
+		x := copy(out, d.a[d.n:d.rate])
+		d.n += x
+		out = out[x:]
 	}
 
 	return
@@ -183,3 +171,74 @@ func (d *state) Sum(in []byte) []byte {
 	dup.Read(hash)
 	return append(in, hash...)
 }
+
+const (
+	magicSHA3   = "sha\x08"
+	magicShake  = "sha\x09"
+	magicCShake = "sha\x0a"
+	magicKeccak = "sha\x0b"
+	// magic || rate || main state || n || sponge direction
+	marshaledSize = len(magicSHA3) + 1 + 200 + 1 + 1
+)
+
+func (d *state) MarshalBinary() ([]byte, error) {
+	return d.AppendBinary(make([]byte, 0, marshaledSize))
+}
+
+func (d *state) AppendBinary(b []byte) ([]byte, error) {
+	switch d.dsbyte {
+	case dsbyteSHA3:
+		b = append(b, magicSHA3...)
+	case dsbyteShake:
+		b = append(b, magicShake...)
+	case dsbyteCShake:
+		b = append(b, magicCShake...)
+	case dsbyteKeccak:
+		b = append(b, magicKeccak...)
+	default:
+		panic("unknown dsbyte")
+	}
+	// rate is at most 168, and n is at most rate.
+	b = append(b, byte(d.rate))
+	b = append(b, d.a[:]...)
+	b = append(b, byte(d.n), byte(d.state))
+	return b, nil
+}
+
+func (d *state) UnmarshalBinary(b []byte) error {
+	if len(b) != marshaledSize {
+		return errors.New("sha3: invalid hash state")
+	}
+
+	magic := string(b[:len(magicSHA3)])
+	b = b[len(magicSHA3):]
+	switch {
+	case magic == magicSHA3 && d.dsbyte == dsbyteSHA3:
+	case magic == magicShake && d.dsbyte == dsbyteShake:
+	case magic == magicCShake && d.dsbyte == dsbyteCShake:
+	case magic == magicKeccak && d.dsbyte == dsbyteKeccak:
+	default:
+		return errors.New("sha3: invalid hash state identifier")
+	}
+
+	rate := int(b[0])
+	b = b[1:]
+	if rate != d.rate {
+		return errors.New("sha3: invalid hash state function")
+	}
+
+	copy(d.a[:], b)
+	b = b[len(d.a):]
+
+	n, state := int(b[0]), spongeDirection(b[1])
+	if n > d.rate {
+		return errors.New("sha3: invalid hash state")
+	}
+	d.n = n
+	if state != spongeAbsorbing && state != spongeSqueezing {
+		return errors.New("sha3: invalid hash state")
+	}
+	d.state = state
+
+	return nil
+}

sha3/sha3_test.go

@@ -13,10 +13,12 @@ package sha3
 import (
 	"bytes"
 	"compress/flate"
+	"encoding"
 	"encoding/hex"
 	"encoding/json"
 	"fmt"
 	"hash"
+	"io"
 	"math/rand"
 	"os"
 	"strings"
@@ -375,6 +377,153 @@ func TestClone(t *testing.T) {
 	}
 }
 
+func TestCSHAKEAccumulated(t *testing.T) {
+	// Generated with pycryptodome@3.20.0
+	//
+	//    from Crypto.Hash import cSHAKE128
+	//    rng = cSHAKE128.new()
+	//    acc = cSHAKE128.new()
+	//    for n in range(200):
+	//        N = rng.read(n)
+	//        for s in range(200):
+	//            S = rng.read(s)
+	//            c = cSHAKE128.cSHAKE_XOF(data=None, custom=S, capacity=256, function=N)
+	//            c.update(rng.read(100))
+	//            acc.update(c.read(200))
+	//            c = cSHAKE128.cSHAKE_XOF(data=None, custom=S, capacity=256, function=N)
+	//            c.update(rng.read(168))
+	//            acc.update(c.read(200))
+	//            c = cSHAKE128.cSHAKE_XOF(data=None, custom=S, capacity=256, function=N)
+	//            c.update(rng.read(200))
+	//            acc.update(c.read(200))
+	//    print(acc.read(32).hex())
+	//
+	// and with @noble/hashes@v1.5.0
+	//
+	//    import { bytesToHex } from "@noble/hashes/utils";
+	//    import { cshake128 } from "@noble/hashes/sha3-addons";
+	//    const rng = cshake128.create();
+	//    const acc = cshake128.create();
+	//    for (let n = 0; n < 200; n++) {
+	//        const N = rng.xof(n);
+	//        for (let s = 0; s < 200; s++) {
+	//            const S = rng.xof(s);
+	//            let c = cshake128.create({ NISTfn: N, personalization: S });
+	//            c.update(rng.xof(100));
+	//            acc.update(c.xof(200));
+	//            c = cshake128.create({ NISTfn: N, personalization: S });
+	//            c.update(rng.xof(168));
+	//            acc.update(c.xof(200));
+	//            c = cshake128.create({ NISTfn: N, personalization: S });
+	//            c.update(rng.xof(200));
+	//            acc.update(c.xof(200));
+	//        }
+	//    }
+	//    console.log(bytesToHex(acc.xof(32)));
+	//
+	t.Run("cSHAKE128", func(t *testing.T) {
+		testCSHAKEAccumulated(t, NewCShake128, rateK256,
+			"bb14f8657c6ec5403d0b0e2ef3d3393497e9d3b1a9a9e8e6c81dbaa5fd809252")
+	})
+	t.Run("cSHAKE256", func(t *testing.T) {
+		testCSHAKEAccumulated(t, NewCShake256, rateK512,
+			"0baaf9250c6e25f0c14ea5c7f9bfde54c8a922c8276437db28f3895bdf6eeeef")
+	})
+}
+
+func testCSHAKEAccumulated(t *testing.T, newCShake func(N, S []byte) ShakeHash, rate int64, exp string) {
+	rnd := newCShake(nil, nil)
+	acc := newCShake(nil, nil)
+	for n := 0; n < 200; n++ {
+		N := make([]byte, n)
+		rnd.Read(N)
+		for s := 0; s < 200; s++ {
+			S := make([]byte, s)
+			rnd.Read(S)
+
+			c := newCShake(N, S)
+			io.CopyN(c, rnd, 100 /* < rate */)
+			io.CopyN(acc, c, 200)
+
+			c.Reset()
+			io.CopyN(c, rnd, rate)
+			io.CopyN(acc, c, 200)
+
+			c.Reset()
+			io.CopyN(c, rnd, 200 /* > rate */)
+			io.CopyN(acc, c, 200)
+		}
+	}
+	if got := hex.EncodeToString(acc.Sum(nil)[:32]); got != exp {
+		t.Errorf("got %s, want %s", got, exp)
+	}
+}
+
+func TestCSHAKELargeS(t *testing.T) {
+	if testing.Short() {
+		t.Skip("skipping test in short mode.")
+	}
+
+	// See https://go.dev/issue/66232.
+	const s = (1<<32)/8 + 1000 // s * 8 > 2^32
+	S := make([]byte, s)
+	rnd := NewShake128()
+	rnd.Read(S)
+	c := NewCShake128(nil, S)
+	io.CopyN(c, rnd, 1000)
+
+	// Generated with pycryptodome@3.20.0
+	//
+	//    from Crypto.Hash import cSHAKE128
+	//    rng = cSHAKE128.new()
+	//    S = rng.read(536871912)
+	//    c = cSHAKE128.new(custom=S)
+	//    c.update(rng.read(1000))
+	//    print(c.read(32).hex())
+	//
+	exp := "2cb9f237767e98f2614b8779cf096a52da9b3a849280bbddec820771ae529cf0"
+	if got := hex.EncodeToString(c.Sum(nil)); got != exp {
+		t.Errorf("got %s, want %s", got, exp)
+	}
+}
+
+func TestMarshalUnmarshal(t *testing.T) {
+	t.Run("SHA3-224", func(t *testing.T) { testMarshalUnmarshal(t, New224()) })
+	t.Run("SHA3-256", func(t *testing.T) { testMarshalUnmarshal(t, New256()) })
+	t.Run("SHA3-384", func(t *testing.T) { testMarshalUnmarshal(t, New384()) })
+	t.Run("SHA3-512", func(t *testing.T) { testMarshalUnmarshal(t, New512()) })
+	t.Run("SHAKE128", func(t *testing.T) { testMarshalUnmarshal(t, NewShake128()) })
+	t.Run("SHAKE256", func(t *testing.T) { testMarshalUnmarshal(t, NewShake256()) })
+	t.Run("cSHAKE128", func(t *testing.T) { testMarshalUnmarshal(t, NewCShake128([]byte("N"), []byte("S"))) })
+	t.Run("cSHAKE256", func(t *testing.T) { testMarshalUnmarshal(t, NewCShake256([]byte("N"), []byte("S"))) })
+	t.Run("Keccak-256", func(t *testing.T) { testMarshalUnmarshal(t, NewLegacyKeccak256()) })
+	t.Run("Keccak-512", func(t *testing.T) { testMarshalUnmarshal(t, NewLegacyKeccak512()) })
+}
+
+// TODO(filippo): move this to crypto/internal/cryptotest.
+func testMarshalUnmarshal(t *testing.T, h hash.Hash) {
+	buf := make([]byte, 200)
+	rand.Read(buf)
+	n := rand.Intn(200)
+	h.Write(buf)
+	want := h.Sum(nil)
+	h.Reset()
+	h.Write(buf[:n])
+	b, err := h.(encoding.BinaryMarshaler).MarshalBinary()
+	if err != nil {
+		t.Errorf("MarshalBinary: %v", err)
+	}
+	h.Write(bytes.Repeat([]byte{0}, 200))
+	if err := h.(encoding.BinaryUnmarshaler).UnmarshalBinary(b); err != nil {
+		t.Errorf("UnmarshalBinary: %v", err)
+	}
+	h.Write(buf[n:])
+	got := h.Sum(nil)
+	if !bytes.Equal(got, want) {
+		t.Errorf("got %x, want %x", got, want)
+	}
+}
+
 // BenchmarkPermutationFunction measures the speed of the permutation function
 // with no input data.
 func BenchmarkPermutationFunction(b *testing.B) {

sha3/shake.go

@@ -16,9 +16,12 @@ package sha3
 // [2] https://doi.org/10.6028/NIST.SP.800-185
 
 import (
+	"bytes"
 	"encoding/binary"
+	"errors"
 	"hash"
 	"io"
+	"math/bits"
 )
 
 // ShakeHash defines the interface to hash functions that support
@@ -50,44 +53,36 @@ type cshakeState struct {
 	initBlock []byte
 }
 
-// Consts for configuring initial SHA-3 state
-const (
-	dsbyteShake  = 0x1f
-	dsbyteCShake = 0x04
-	rate128      = 168
-	rate256      = 136
-)
-
-func bytepad(input []byte, w int) []byte {
-	// leftEncode always returns max 9 bytes
-	buf := make([]byte, 0, 9+len(input)+w)
-	buf = append(buf, leftEncode(uint64(w))...)
-	buf = append(buf, input...)
-	padlen := w - (len(buf) % w)
-	return append(buf, make([]byte, padlen)...)
+func bytepad(data []byte, rate int) []byte {
+	out := make([]byte, 0, 9+len(data)+rate-1)
+	out = append(out, leftEncode(uint64(rate))...)
+	out = append(out, data...)
+	if padlen := rate - len(out)%rate; padlen < rate {
+		out = append(out, make([]byte, padlen)...)
+	}
+	return out
 }
 
-func leftEncode(value uint64) []byte {
-	var b [9]byte
-	binary.BigEndian.PutUint64(b[1:], value)
-	// Trim all but last leading zero bytes
-	i := byte(1)
-	for i < 8 && b[i] == 0 {
-		i++
+func leftEncode(x uint64) []byte {
+	// Let n be the smallest positive integer for which 2^(8n) > x.
+	n := (bits.Len64(x) + 7) / 8
+	if n == 0 {
+		n = 1
 	}
-	// Prepend number of encoded bytes
-	b[i-1] = 9 - i
-	return b[i-1:]
+	// Return n || x with n as a byte and x an n bytes in big-endian order.
+	b := make([]byte, 9)
+	binary.BigEndian.PutUint64(b[1:], x)
+	b = b[9-n-1:]
+	b[0] = byte(n)
+	return b
 }
 
 func newCShake(N, S []byte, rate, outputLen int, dsbyte byte) ShakeHash {
 	c := cshakeState{state: &state{rate: rate, outputLen: outputLen, dsbyte: dsbyte}}
-
-	// leftEncode returns max 9 bytes
-	c.initBlock = make([]byte, 0, 9*2+len(N)+len(S))
-	c.initBlock = append(c.initBlock, leftEncode(uint64(len(N)*8))...)
+	c.initBlock = make([]byte, 0, 9+len(N)+9+len(S)) // leftEncode returns max 9 bytes
+	c.initBlock = append(c.initBlock, leftEncode(uint64(len(N))*8)...)
 	c.initBlock = append(c.initBlock, N...)
-	c.initBlock = append(c.initBlock, leftEncode(uint64(len(S)*8))...)
+	c.initBlock = append(c.initBlock, leftEncode(uint64(len(S))*8)...)
 	c.initBlock = append(c.initBlock, S...)
 	c.Write(bytepad(c.initBlock, c.rate))
 	return &c
@@ -111,6 +106,30 @@ func (c *state) Clone() ShakeHash {
 	return c.clone()
 }
 
+func (c *cshakeState) MarshalBinary() ([]byte, error) {
+	return c.AppendBinary(make([]byte, 0, marshaledSize+len(c.initBlock)))
+}
+
+func (c *cshakeState) AppendBinary(b []byte) ([]byte, error) {
+	b, err := c.state.AppendBinary(b)
+	if err != nil {
+		return nil, err
+	}
+	b = append(b, c.initBlock...)
+	return b, nil
+}
+
+func (c *cshakeState) UnmarshalBinary(b []byte) error {
+	if len(b) <= marshaledSize {
+		return errors.New("sha3: invalid hash state")
+	}
+	if err := c.state.UnmarshalBinary(b[:marshaledSize]); err != nil {
+		return err
+	}
+	c.initBlock = bytes.Clone(b[marshaledSize:])
+	return nil
+}
+
 // NewShake128 creates a new SHAKE128 variable-output-length ShakeHash.
 // Its generic security strength is 128 bits against all attacks if at
 // least 32 bytes of its output are used.
@@ -126,11 +145,11 @@ func NewShake256() ShakeHash {
 }
 
 func newShake128Generic() *state {
-	return &state{rate: rate128, outputLen: 32, dsbyte: dsbyteShake}
+	return &state{rate: rateK256, outputLen: 32, dsbyte: dsbyteShake}
 }
 
 func newShake256Generic() *state {
-	return &state{rate: rate256, outputLen: 64, dsbyte: dsbyteShake}
+	return &state{rate: rateK512, outputLen: 64, dsbyte: dsbyteShake}
 }
 
 // NewCShake128 creates a new instance of cSHAKE128 variable-output-length ShakeHash,
@@ -143,7 +162,7 @@ func NewCShake128(N, S []byte) ShakeHash {
 	if len(N) == 0 && len(S) == 0 {
 		return NewShake128()
 	}
-	return newCShake(N, S, rate128, 32, dsbyteCShake)
+	return newCShake(N, S, rateK256, 32, dsbyteCShake)
 }
 
 // NewCShake256 creates a new instance of cSHAKE256 variable-output-length ShakeHash,
@@ -156,7 +175,7 @@ func NewCShake256(N, S []byte) ShakeHash {
 	if len(N) == 0 && len(S) == 0 {
 		return NewShake256()
 	}
-	return newCShake(N, S, rate256, 64, dsbyteCShake)
+	return newCShake(N, S, rateK512, 64, dsbyteCShake)
 }
 
 // ShakeSum128 writes an arbitrary-length digest of data into hash.

sha3/xor.go

@@ -1,40 +0,0 @@
-// Copyright 2015 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.
-
-package sha3
-
-import (
-	"crypto/subtle"
-	"encoding/binary"
-	"unsafe"
-
-	"golang.org/x/sys/cpu"
-)
-
-// xorIn xors the bytes in buf into the state.
-func xorIn(d *state, buf []byte) {
-	if cpu.IsBigEndian {
-		for i := 0; len(buf) >= 8; i++ {
-			a := binary.LittleEndian.Uint64(buf)
-			d.a[i] ^= a
-			buf = buf[8:]
-		}
-	} else {
-		ab := (*[25 * 64 / 8]byte)(unsafe.Pointer(&d.a))
-		subtle.XORBytes(ab[:], ab[:], buf)
-	}
-}
-
-// copyOut copies uint64s to a byte buffer.
-func copyOut(d *state, b []byte) {
-	if cpu.IsBigEndian {
-		for i := 0; len(b) >= 8; i++ {
-			binary.LittleEndian.PutUint64(b, d.a[i])
-			b = b[8:]
-		}
-	} else {
-		ab := (*[25 * 64 / 8]byte)(unsafe.Pointer(&d.a))
-		copy(b, ab[:])
-	}
-}

ssh/agent/keyring.go

@@ -175,6 +175,15 @@ func (r *keyring) Add(key AddedKey) error {
 		p.expire = &t
 	}
 
+	// If we already have a Signer with the same public key, replace it with the
+	// new one.
+	for idx, k := range r.keys {
+		if bytes.Equal(k.signer.PublicKey().Marshal(), p.signer.PublicKey().Marshal()) {
+			r.keys[idx] = p
+			return nil
+		}
+	}
+
 	r.keys = append(r.keys, p)
 
 	return nil

ssh/agent/keyring_test.go

@@ -29,6 +29,10 @@ func validateListedKeys(t *testing.T, a Agent, expectedKeys []string) {
 		t.Fatalf("failed to list keys: %v", err)
 		return
 	}
+	if len(listedKeys) != len(expectedKeys) {
+		t.Fatalf("expeted %d key, got %d", len(expectedKeys), len(listedKeys))
+		return
+	}
 	actualKeys := make(map[string]bool)
 	for _, key := range listedKeys {
 		actualKeys[key.Comment] = true
@@ -74,3 +78,45 @@ func TestKeyringAddingAndRemoving(t *testing.T) {
 	}
 	validateListedKeys(t, k, []string{})
 }
+
+func TestAddDuplicateKey(t *testing.T) {
+	keyNames := []string{"rsa", "user"}
+
+	k := NewKeyring()
+	for _, keyName := range keyNames {
+		addTestKey(t, k, keyName)
+	}
+	validateListedKeys(t, k, keyNames)
+	// Add the keys again.
+	for _, keyName := range keyNames {
+		addTestKey(t, k, keyName)
+	}
+	validateListedKeys(t, k, keyNames)
+	// Add an existing key with an updated comment.
+	keyName := keyNames[0]
+	addedKey := AddedKey{
+		PrivateKey: testPrivateKeys[keyName],
+		Comment:    "comment updated",
+	}
+	err := k.Add(addedKey)
+	if err != nil {
+		t.Fatalf("failed to add key %q: %v", keyName, err)
+	}
+	// Check the that key is found and the comment was updated.
+	keys, err := k.List()
+	if err != nil {
+		t.Fatalf("failed to list keys: %v", err)
+	}
+	if len(keys) != len(keyNames) {
+		t.Fatalf("expected %d keys, got %d", len(keyNames), len(keys))
+	}
+	isFound := false
+	for _, key := range keys {
+		if key.Comment == addedKey.Comment {
+			isFound = true
+		}
+	}
+	if !isFound {
+		t.Fatal("key with the updated comment not found")
+	}
+}

ssh/certs_test.go

@@ -15,14 +15,12 @@ import (
 	"reflect"
 	"testing"
 	"time"
+
+	"golang.org/x/crypto/ssh/testdata"
 )
 
-// Cert generated by ssh-keygen 6.0p1 Debian-4.
-// % ssh-keygen -s ca-key -I test user-key
-const exampleSSHCert = `ssh-rsa-cert-v01@openssh.com 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`
-
 func TestParseCert(t *testing.T) {
-	authKeyBytes := []byte(exampleSSHCert)
+	authKeyBytes := bytes.TrimSuffix(testdata.SSHCertificates["rsa"], []byte(" host.example.com\n"))
 
 	key, _, _, rest, err := ParseAuthorizedKey(authKeyBytes)
 	if err != nil {
@@ -103,7 +101,7 @@ func TestParseCertWithOptions(t *testing.T) {
 }
 
 func TestValidateCert(t *testing.T) {
-	key, _, _, _, err := ParseAuthorizedKey([]byte(exampleSSHCert))
+	key, _, _, _, err := ParseAuthorizedKey(testdata.SSHCertificates["rsa-user-testcertificate"])
 	if err != nil {
 		t.Fatalf("ParseAuthorizedKey: %v", err)
 	}
@@ -116,7 +114,7 @@ func TestValidateCert(t *testing.T) {
 		return bytes.Equal(k.Marshal(), validCert.SignatureKey.Marshal())
 	}
 
-	if err := checker.CheckCert("user", validCert); err != nil {
+	if err := checker.CheckCert("testcertificate", validCert); err != nil {
 		t.Errorf("Unable to validate certificate: %v", err)
 	}
 	invalidCert := &Certificate{
@@ -125,7 +123,7 @@ func TestValidateCert(t *testing.T) {
 		ValidBefore:  CertTimeInfinity,
 		Signature:    &Signature{},
 	}
-	if err := checker.CheckCert("user", invalidCert); err == nil {
+	if err := checker.CheckCert("testcertificate", invalidCert); err == nil {
 		t.Error("Invalid cert signature passed validation")
 	}
 }

ssh/client_auth.go

@@ -555,6 +555,7 @@ func (cb KeyboardInteractiveChallenge) auth(session []byte, user string, c packe
 	}
 
 	gotMsgExtInfo := false
+	gotUserAuthInfoRequest := false
 	for {
 		packet, err := c.readPacket()
 		if err != nil {
@@ -585,6 +586,9 @@ func (cb KeyboardInteractiveChallenge) auth(session []byte, user string, c packe
 			if msg.PartialSuccess {
 				return authPartialSuccess, msg.Methods, nil
 			}
+			if !gotUserAuthInfoRequest {
+				return authFailure, msg.Methods, unexpectedMessageError(msgUserAuthInfoRequest, packet[0])
+			}
 			return authFailure, msg.Methods, nil
 		case msgUserAuthSuccess:
 			return authSuccess, nil, nil
@@ -596,6 +600,7 @@ func (cb KeyboardInteractiveChallenge) auth(session []byte, user string, c packe
 		if err := Unmarshal(packet, &msg); err != nil {
 			return authFailure, nil, err
 		}
+		gotUserAuthInfoRequest = true
 
 		// Manually unpack the prompt/echo pairs.
 		rest := msg.Prompts

ssh/client_auth_test.go

@@ -38,7 +38,7 @@ func tryAuth(t *testing.T, config *ClientConfig) error {
 	return err
 }
 
-// tryAuth runs a handshake with a given config against an SSH server
+// tryAuthWithGSSAPIWithMICConfig runs a handshake with a given config against an SSH server
 // with a given GSSAPIWithMICConfig and config serverConfig. Returns both client and server side errors.
 func tryAuthWithGSSAPIWithMICConfig(t *testing.T, clientConfig *ClientConfig, gssAPIWithMICConfig *GSSAPIWithMICConfig) error {
 	err, _ := tryAuthBothSides(t, clientConfig, gssAPIWithMICConfig)
@@ -641,17 +641,28 @@ func TestClientAuthMaxAuthTries(t *testing.T) {
 		defer c1.Close()
 		defer c2.Close()
 
-		go newServer(c1, serverConfig)
-		_, _, _, err = NewClientConn(c2, "", clientConfig)
-		if tries > 2 {
-			if err == nil {
+		errCh := make(chan error, 1)
+
+		go func() {
+			_, err := newServer(c1, serverConfig)
+			errCh <- err
+		}()
+		_, _, _, cliErr := NewClientConn(c2, "", clientConfig)
+		srvErr := <-errCh
+
+		if tries > serverConfig.MaxAuthTries {
+			if cliErr == nil {
 				t.Fatalf("client: got no error, want %s", expectedErr)
-			} else if err.Error() != expectedErr.Error() {
+			} else if cliErr.Error() != expectedErr.Error() {
 				t.Fatalf("client: got %s, want %s", err, expectedErr)
 			}
+			var authErr *ServerAuthError
+			if !errors.As(srvErr, &authErr) {
+				t.Errorf("expected ServerAuthError, got: %v", srvErr)
+			}
 		} else {
-			if err != nil {
-				t.Fatalf("client: got %s, want no error", err)
+			if cliErr != nil {
+				t.Fatalf("client: got %s, want no error", cliErr)
 			}
 		}
 	}
@@ -1282,3 +1293,92 @@ func TestCertAuthOpenSSHCompat(t *testing.T) {
 		t.Fatalf("unable to dial remote side: %s", err)
 	}
 }
+
+func TestKeyboardInteractiveAuthEarlyFail(t *testing.T) {
+	const maxAuthTries = 2
+
+	c1, c2, err := netPipe()
+	if err != nil {
+		t.Fatalf("netPipe: %v", err)
+	}
+	defer c1.Close()
+	defer c2.Close()
+
+	// Start testserver
+	serverConfig := &ServerConfig{
+		MaxAuthTries: maxAuthTries,
+		KeyboardInteractiveCallback: func(c ConnMetadata,
+			client KeyboardInteractiveChallenge) (*Permissions, error) {
+			// Fail keyboard-interactive authentication early before
+			// any prompt is sent to client.
+			return nil, errors.New("keyboard-interactive auth failed")
+		},
+		PasswordCallback: func(c ConnMetadata,
+			pass []byte) (*Permissions, error) {
+			if string(pass) == clientPassword {
+				return nil, nil
+			}
+			return nil, errors.New("password auth failed")
+		},
+	}
+	serverConfig.AddHostKey(testSigners["rsa"])
+
+	serverDone := make(chan struct{})
+	go func() {
+		defer func() { serverDone <- struct{}{} }()
+		conn, chans, reqs, err := NewServerConn(c2, serverConfig)
+		if err != nil {
+			return
+		}
+		_ = conn.Close()
+
+		discarderDone := make(chan struct{})
+		go func() {
+			defer func() { discarderDone <- struct{}{} }()
+			DiscardRequests(reqs)
+		}()
+		for newChannel := range chans {
+			newChannel.Reject(Prohibited,
+				"testserver not accepting requests")
+		}
+
+		<-discarderDone
+	}()
+
+	// Connect to testserver, expect KeyboardInteractive() to be not called,
+	// PasswordCallback() to be called and connection to succeed.
+	passwordCallbackCalled := false
+	clientConfig := &ClientConfig{
+		User: "testuser",
+		Auth: []AuthMethod{
+			RetryableAuthMethod(KeyboardInteractive(func(name,
+				instruction string, questions []string,
+				echos []bool) ([]string, error) {
+				t.Errorf("unexpected call to KeyboardInteractive()")
+				return []string{clientPassword}, nil
+			}), maxAuthTries),
+			RetryableAuthMethod(PasswordCallback(func() (secret string,
+				err error) {
+				t.Logf("PasswordCallback()")
+				passwordCallbackCalled = true
+				return clientPassword, nil
+			}), maxAuthTries),
+		},
+		HostKeyCallback: InsecureIgnoreHostKey(),
+	}
+
+	conn, _, _, err := NewClientConn(c1, "", clientConfig)
+	if err != nil {
+		t.Errorf("unexpected NewClientConn() error: %v", err)
+	}
+	if conn != nil {
+		conn.Close()
+	}
+
+	// Wait for server to finish.
+	<-serverDone
+
+	if !passwordCallbackCalled {
+		t.Errorf("expected PasswordCallback() to be called")
+	}
+}

ssh/handshake.go

@@ -5,7 +5,6 @@
 package ssh
 
 import (
-	"crypto/rand"
 	"errors"
 	"fmt"
 	"io"
@@ -25,6 +24,11 @@ const debugHandshake = false
 // quickly.
 const chanSize = 16
 
+// maxPendingPackets sets the maximum number of packets to queue while waiting
+// for KEX to complete. This limits the total pending data to maxPendingPackets
+// * maxPacket bytes, which is ~16.8MB.
+const maxPendingPackets = 64
+
 // keyingTransport is a packet based transport that supports key
 // changes. It need not be thread-safe. It should pass through
 // msgNewKeys in both directions.
@@ -73,13 +77,22 @@ type handshakeTransport struct {
 	incoming  chan []byte
 	readError error
 
-	mu               sync.Mutex
-	writeError       error
-	sentInitPacket   []byte
-	sentInitMsg      *kexInitMsg
-	pendingPackets   [][]byte // Used when a key exchange is in progress.
+	mu sync.Mutex
+	// Condition for the above mutex. It is used to notify a completed key
+	// exchange or a write failure. Writes can wait for this condition while a
+	// key exchange is in progress.
+	writeCond      *sync.Cond
+	writeError     error
+	sentInitPacket []byte
+	sentInitMsg    *kexInitMsg
+	// Used to queue writes when a key exchange is in progress. The length is
+	// limited by pendingPacketsSize. Once full, writes will block until the key
+	// exchange is completed or an error occurs. If not empty, it is emptied
+	// all at once when the key exchange is completed in kexLoop.
+	pendingPackets   [][]byte
 	writePacketsLeft uint32
 	writeBytesLeft   int64
+	userAuthComplete bool // whether the user authentication phase is complete
 
 	// If the read loop wants to schedule a kex, it pings this
 	// channel, and the write loop will send out a kex
@@ -133,6 +146,7 @@ func newHandshakeTransport(conn keyingTransport, config *Config, clientVersion,
 
 		config: config,
 	}
+	t.writeCond = sync.NewCond(&t.mu)
 	t.resetReadThresholds()
 	t.resetWriteThresholds()
 
@@ -259,6 +273,7 @@ func (t *handshakeTransport) recordWriteError(err error) {
 	defer t.mu.Unlock()
 	if t.writeError == nil && err != nil {
 		t.writeError = err
+		t.writeCond.Broadcast()
 	}
 }
 
@@ -362,6 +377,8 @@ write:
 			}
 		}
 		t.pendingPackets = t.pendingPackets[:0]
+		// Unblock writePacket if waiting for KEX.
+		t.writeCond.Broadcast()
 		t.mu.Unlock()
 	}
 
@@ -483,7 +500,7 @@ func (t *handshakeTransport) sendKexInit() error {
 		CompressionClientServer: supportedCompressions,
 		CompressionServerClient: supportedCompressions,
 	}
-	io.ReadFull(rand.Reader, msg.Cookie[:])
+	io.ReadFull(t.config.Rand, msg.Cookie[:])
 
 	// We mutate the KexAlgos slice, in order to add the kex-strict extension algorithm,
 	// and possibly to add the ext-info extension algorithm. Since the slice may be the
@@ -552,26 +569,44 @@ func (t *handshakeTransport) sendKexInit() error {
 	return nil
 }
 
+var errSendBannerPhase = errors.New("ssh: SendAuthBanner outside of authentication phase")
+
 func (t *handshakeTransport) writePacket(p []byte) error {
+	t.mu.Lock()
+	defer t.mu.Unlock()
+
 	switch p[0] {
 	case msgKexInit:
 		return errors.New("ssh: only handshakeTransport can send kexInit")
 	case msgNewKeys:
 		return errors.New("ssh: only handshakeTransport can send newKeys")
+	case msgUserAuthBanner:
+		if t.userAuthComplete {
+			return errSendBannerPhase
+		}
+	case msgUserAuthSuccess:
+		t.userAuthComplete = true
 	}
 
-	t.mu.Lock()
-	defer t.mu.Unlock()
 	if t.writeError != nil {
 		return t.writeError
 	}
 
 	if t.sentInitMsg != nil {
-		// Copy the packet so the writer can reuse the buffer.
-		cp := make([]byte, len(p))
-		copy(cp, p)
-		t.pendingPackets = append(t.pendingPackets, cp)
-		return nil
+		if len(t.pendingPackets) < maxPendingPackets {
+			// Copy the packet so the writer can reuse the buffer.
+			cp := make([]byte, len(p))
+			copy(cp, p)
+			t.pendingPackets = append(t.pendingPackets, cp)
+			return nil
+		}
+		for t.sentInitMsg != nil {
+			// Block and wait for KEX to complete or an error.
+			t.writeCond.Wait()
+			if t.writeError != nil {
+				return t.writeError
+			}
+		}
 	}
 
 	if t.writeBytesLeft > 0 {
@@ -588,6 +623,7 @@ func (t *handshakeTransport) writePacket(p []byte) error {
 
 	if err := t.pushPacket(p); err != nil {
 		t.writeError = err
+		t.writeCond.Broadcast()
 	}
 
 	return nil

ssh/handshake_test.go

@@ -539,6 +539,226 @@ func TestDisconnect(t *testing.T) {
 	}
 }
 
+type mockKeyingTransport struct {
+	packetConn
+	kexInitAllowed chan struct{}
+	kexInitSent    chan struct{}
+}
+
+func (n *mockKeyingTransport) prepareKeyChange(*algorithms, *kexResult) error {
+	return nil
+}
+
+func (n *mockKeyingTransport) writePacket(packet []byte) error {
+	if packet[0] == msgKexInit {
+		<-n.kexInitAllowed
+		n.kexInitSent <- struct{}{}
+	}
+	return n.packetConn.writePacket(packet)
+}
+
+func (n *mockKeyingTransport) readPacket() ([]byte, error) {
+	return n.packetConn.readPacket()
+}
+
+func (n *mockKeyingTransport) setStrictMode() error { return nil }
+
+func (n *mockKeyingTransport) setInitialKEXDone() {}
+
+func TestHandshakePendingPacketsWait(t *testing.T) {
+	a, b := memPipe()
+
+	trS := &mockKeyingTransport{
+		packetConn:     a,
+		kexInitAllowed: make(chan struct{}, 2),
+		kexInitSent:    make(chan struct{}, 2),
+	}
+	// Allow the first KEX.
+	trS.kexInitAllowed <- struct{}{}
+
+	trC := &mockKeyingTransport{
+		packetConn:     b,
+		kexInitAllowed: make(chan struct{}, 2),
+		kexInitSent:    make(chan struct{}, 2),
+	}
+	// Allow the first KEX.
+	trC.kexInitAllowed <- struct{}{}
+
+	clientConf := &ClientConfig{
+		HostKeyCallback: InsecureIgnoreHostKey(),
+	}
+	clientConf.SetDefaults()
+
+	v := []byte("version")
+	client := newClientTransport(trC, v, v, clientConf, "addr", nil)
+
+	serverConf := &ServerConfig{}
+	serverConf.AddHostKey(testSigners["ecdsa"])
+	serverConf.AddHostKey(testSigners["rsa"])
+	serverConf.SetDefaults()
+	server := newServerTransport(trS, v, v, serverConf)
+
+	if err := server.waitSession(); err != nil {
+		t.Fatalf("server.waitSession: %v", err)
+	}
+	if err := client.waitSession(); err != nil {
+		t.Fatalf("client.waitSession: %v", err)
+	}
+
+	<-trC.kexInitSent
+	<-trS.kexInitSent
+
+	// Allow and request new KEX server side.
+	trS.kexInitAllowed <- struct{}{}
+	server.requestKeyExchange()
+	// Wait until the KEX init is sent.
+	<-trS.kexInitSent
+	// The client is not allowed to respond to the KEX, so writes will be
+	// blocked on the server side once the packets queue is full.
+	for i := 0; i < maxPendingPackets; i++ {
+		p := []byte{msgRequestSuccess, byte(i)}
+		if err := server.writePacket(p); err != nil {
+			t.Errorf("unexpected write error: %v", err)
+		}
+	}
+	// The packets queue is now full, the next write will block.
+	server.mu.Lock()
+	if len(server.pendingPackets) != maxPendingPackets {
+		t.Errorf("unexpected pending packets size; got: %d, want: %d", len(server.pendingPackets), maxPendingPackets)
+	}
+	server.mu.Unlock()
+
+	writeDone := make(chan struct{})
+	go func() {
+		defer close(writeDone)
+
+		p := []byte{msgRequestSuccess, byte(65)}
+		// This write will block until KEX completes.
+		err := server.writePacket(p)
+		if err != nil {
+			t.Errorf("unexpected write error: %v", err)
+		}
+	}()
+
+	// Consume packets on the client side
+	readDone := make(chan bool)
+	go func() {
+		defer close(readDone)
+
+		for {
+			if _, err := client.readPacket(); err != nil {
+				if err != io.EOF {
+					t.Errorf("unexpected read error: %v", err)
+				}
+				break
+			}
+		}
+	}()
+
+	// Allow the client to reply to the KEX and so unblock the write goroutine.
+	trC.kexInitAllowed <- struct{}{}
+	<-trC.kexInitSent
+	<-writeDone
+	// Close the client to unblock the read goroutine.
+	client.Close()
+	<-readDone
+	server.Close()
+}
+
+func TestHandshakePendingPacketsError(t *testing.T) {
+	a, b := memPipe()
+
+	trS := &mockKeyingTransport{
+		packetConn:     a,
+		kexInitAllowed: make(chan struct{}, 2),
+		kexInitSent:    make(chan struct{}, 2),
+	}
+	// Allow the first KEX.
+	trS.kexInitAllowed <- struct{}{}
+
+	trC := &mockKeyingTransport{
+		packetConn:     b,
+		kexInitAllowed: make(chan struct{}, 2),
+		kexInitSent:    make(chan struct{}, 2),
+	}
+	// Allow the first KEX.
+	trC.kexInitAllowed <- struct{}{}
+
+	clientConf := &ClientConfig{
+		HostKeyCallback: InsecureIgnoreHostKey(),
+	}
+	clientConf.SetDefaults()
+
+	v := []byte("version")
+	client := newClientTransport(trC, v, v, clientConf, "addr", nil)
+
+	serverConf := &ServerConfig{}
+	serverConf.AddHostKey(testSigners["ecdsa"])
+	serverConf.AddHostKey(testSigners["rsa"])
+	serverConf.SetDefaults()
+	server := newServerTransport(trS, v, v, serverConf)
+
+	if err := server.waitSession(); err != nil {
+		t.Fatalf("server.waitSession: %v", err)
+	}
+	if err := client.waitSession(); err != nil {
+		t.Fatalf("client.waitSession: %v", err)
+	}
+
+	<-trC.kexInitSent
+	<-trS.kexInitSent
+
+	// Allow and request new KEX server side.
+	trS.kexInitAllowed <- struct{}{}
+	server.requestKeyExchange()
+	// Wait until the KEX init is sent.
+	<-trS.kexInitSent
+	// The client is not allowed to respond to the KEX, so writes will be
+	// blocked on the server side once the packets queue is full.
+	for i := 0; i < maxPendingPackets; i++ {
+		p := []byte{msgRequestSuccess, byte(i)}
+		if err := server.writePacket(p); err != nil {
+			t.Errorf("unexpected write error: %v", err)
+		}
+	}
+	// The packets queue is now full, the next write will block.
+	writeDone := make(chan struct{})
+	go func() {
+		defer close(writeDone)
+
+		p := []byte{msgRequestSuccess, byte(65)}
+		// This write will block until KEX completes.
+		err := server.writePacket(p)
+		if err != io.EOF {
+			t.Errorf("unexpected write error: %v", err)
+		}
+	}()
+
+	// Consume packets on the client side
+	readDone := make(chan bool)
+	go func() {
+		defer close(readDone)
+
+		for {
+			if _, err := client.readPacket(); err != nil {
+				if err != io.EOF {
+					t.Errorf("unexpected read error: %v", err)
+				}
+				break
+			}
+		}
+	}()
+
+	// Close the server to unblock the write after an error
+	server.Close()
+	<-writeDone
+	// Unblock the pending write and close the client to unblock the read
+	// goroutine.
+	trC.kexInitAllowed <- struct{}{}
+	client.Close()
+	<-readDone
+}
+
 func TestHandshakeRekeyDefault(t *testing.T) {
 	clientConf := &ClientConfig{
 		Config: Config{

ssh/knownhosts/knownhosts.go

@@ -302,8 +302,8 @@ func (k *KnownKey) String() string {
 // applications can offer an interactive prompt to the user.
 type KeyError struct {
 	// Want holds the accepted host keys. For each key algorithm,
-	// there can be one hostkey.  If Want is empty, the host is
-	// unknown. If Want is non-empty, there was a mismatch, which
+	// there can be multiple hostkeys.  If Want is empty, the host
+	// is unknown. If Want is non-empty, there was a mismatch, which
 	// can signify a MITM attack.
 	Want []KnownKey
 }
@@ -358,34 +358,20 @@ func (db *hostKeyDB) checkAddr(a addr, remoteKey ssh.PublicKey) error {
 	// is just a key for the IP address, but not for the
 	// hostname?
 
-	// Algorithm => key.
-	knownKeys := map[string]KnownKey{}
+	keyErr := &KeyError{}
+
 	for _, l := range db.lines {
-		if l.match(a) {
-			typ := l.knownKey.Key.Type()
-			if _, ok := knownKeys[typ]; !ok {
-				knownKeys[typ] = l.knownKey
-			}
+		if !l.match(a) {
+			continue
+		}
+
+		keyErr.Want = append(keyErr.Want, l.knownKey)
+		if keyEq(l.knownKey.Key, remoteKey) {
+			return nil
 		}
 	}
 
-	keyErr := &KeyError{}
-	for _, v := range knownKeys {
-		keyErr.Want = append(keyErr.Want, v)
-	}
-
-	// Unknown remote host.
-	if len(knownKeys) == 0 {
-		return keyErr
-	}
-
-	// If the remote host starts using a different, unknown key type, we
-	// also interpret that as a mismatch.
-	if known, ok := knownKeys[remoteKey.Type()]; !ok || !keyEq(known.Key, remoteKey) {
-		return keyErr
-	}
-
-	return nil
+	return keyErr
 }
 
 // The Read function parses file contents.

ssh/knownhosts/knownhosts_test.go

@@ -201,17 +201,6 @@ func TestHostNamePrecedence(t *testing.T) {
 	}
 }
 
-func TestDBOrderingPrecedenceKeyType(t *testing.T) {
-	str := fmt.Sprintf("server.org,%s %s\nserver.org,%s %s", testAddr, edKeyStr, testAddr, alternateEdKeyStr)
-	db := testDB(t, str)
-
-	if err := db.check("server.org:22", testAddr, alternateEdKey); err == nil {
-		t.Errorf("check succeeded")
-	} else if _, ok := err.(*KeyError); !ok {
-		t.Errorf("got %T, want *KeyError", err)
-	}
-}
-
 func TestNegate(t *testing.T) {
 	str := fmt.Sprintf("%s,!server.org %s", testAddr, edKeyStr)
 	db := testDB(t, str)
@@ -354,3 +343,16 @@ func TestHashedHostkeyCheck(t *testing.T) {
 		t.Errorf("got error %v, want %v", got, want)
 	}
 }
+
+func TestIssue36126(t *testing.T) {
+	str := fmt.Sprintf("server.org,%s %s\nserver.org,%s %s", testAddr, edKeyStr, testAddr, alternateEdKeyStr)
+	db := testDB(t, str)
+
+	if err := db.check("server.org:22", testAddr, edKey); err != nil {
+		t.Errorf("should have passed the check, got %v", err)
+	}
+
+	if err := db.check("server.org:22", testAddr, alternateEdKey); err != nil {
+		t.Errorf("should have passed the check, got %v", err)
+	}
+}

ssh/messages.go

@@ -818,6 +818,8 @@ func decode(packet []byte) (interface{}, error) {
 		return new(userAuthSuccessMsg), nil
 	case msgUserAuthFailure:
 		msg = new(userAuthFailureMsg)
+	case msgUserAuthBanner:
+		msg = new(userAuthBannerMsg)
 	case msgUserAuthPubKeyOk:
 		msg = new(userAuthPubKeyOkMsg)
 	case msgGlobalRequest:

ssh/messages_test.go

@@ -206,6 +206,62 @@ func TestMarshalMultiTag(t *testing.T) {
 	}
 }
 
+func TestDecode(t *testing.T) {
+	rnd := rand.New(rand.NewSource(0))
+	kexInit := new(kexInitMsg).Generate(rnd, 10).Interface()
+	kexDHInit := new(kexDHInitMsg).Generate(rnd, 10).Interface()
+	kexDHReply := new(kexDHReplyMsg)
+	kexDHReply.Y = randomInt(rnd)
+	// Note: userAuthSuccessMsg can't be tested directly since it
+	// doesn't have a field for sshtype. So it's tested separately
+	// at the end.
+	decodeMessageTypes := []interface{}{
+		new(disconnectMsg),
+		new(serviceRequestMsg),
+		new(serviceAcceptMsg),
+		new(extInfoMsg),
+		kexInit,
+		kexDHInit,
+		kexDHReply,
+		new(userAuthRequestMsg),
+		new(userAuthFailureMsg),
+		new(userAuthBannerMsg),
+		new(userAuthPubKeyOkMsg),
+		new(globalRequestMsg),
+		new(globalRequestSuccessMsg),
+		new(globalRequestFailureMsg),
+		new(channelOpenMsg),
+		new(channelDataMsg),
+		new(channelOpenConfirmMsg),
+		new(channelOpenFailureMsg),
+		new(windowAdjustMsg),
+		new(channelEOFMsg),
+		new(channelCloseMsg),
+		new(channelRequestMsg),
+		new(channelRequestSuccessMsg),
+		new(channelRequestFailureMsg),
+		new(userAuthGSSAPIToken),
+		new(userAuthGSSAPIMIC),
+		new(userAuthGSSAPIErrTok),
+		new(userAuthGSSAPIError),
+	}
+	for _, msg := range decodeMessageTypes {
+		decoded, err := decode(Marshal(msg))
+		if err != nil {
+			t.Errorf("error decoding %T", msg)
+		} else if reflect.TypeOf(msg) != reflect.TypeOf(decoded) {
+			t.Errorf("error decoding %T, unexpected %T", msg, decoded)
+		}
+	}
+
+	userAuthSuccess, err := decode([]byte{msgUserAuthSuccess})
+	if err != nil {
+		t.Errorf("error decoding userAuthSuccessMsg")
+	} else if reflect.TypeOf(userAuthSuccess) != reflect.TypeOf((*userAuthSuccessMsg)(nil)) {
+		t.Errorf("error decoding userAuthSuccessMsg, unexpected %T", userAuthSuccess)
+	}
+}
+
 func randomBytes(out []byte, rand *rand.Rand) {
 	for i := 0; i < len(out); i++ {
 		out[i] = byte(rand.Int31())

ssh/server.go

@@ -59,6 +59,27 @@ type GSSAPIWithMICConfig struct {
 	Server GSSAPIServer
 }
 
+// SendAuthBanner implements [ServerPreAuthConn].
+func (s *connection) SendAuthBanner(msg string) error {
+	return s.transport.writePacket(Marshal(&userAuthBannerMsg{
+		Message: msg,
+	}))
+}
+
+func (*connection) unexportedMethodForFutureProofing() {}
+
+// ServerPreAuthConn is the interface available on an incoming server
+// connection before authentication has completed.
+type ServerPreAuthConn interface {
+	unexportedMethodForFutureProofing() // permits growing ServerPreAuthConn safely later, ala testing.TB
+
+	ConnMetadata
+
+	// SendAuthBanner sends a banner message to the client.
+	// It returns an error once the authentication phase has ended.
+	SendAuthBanner(string) error
+}
+
 // ServerConfig holds server specific configuration data.
 type ServerConfig struct {
 	// Config contains configuration shared between client and server.
@@ -118,6 +139,12 @@ type ServerConfig struct {
 	// attempts.
 	AuthLogCallback func(conn ConnMetadata, method string, err error)
 
+	// PreAuthConnCallback, if non-nil, is called upon receiving a new connection
+	// before any authentication has started. The provided ServerPreAuthConn
+	// can be used at any time before authentication is complete, including
+	// after this callback has returned.
+	PreAuthConnCallback func(ServerPreAuthConn)
+
 	// ServerVersion is the version identification string to announce in
 	// the public handshake.
 	// If empty, a reasonable default is used.
@@ -149,7 +176,7 @@ func (s *ServerConfig) AddHostKey(key Signer) {
 }
 
 // cachedPubKey contains the results of querying whether a public key is
-// acceptable for a user.
+// acceptable for a user. This is a FIFO cache.
 type cachedPubKey struct {
 	user       string
 	pubKeyData []byte
@@ -157,7 +184,13 @@ type cachedPubKey struct {
 	perms      *Permissions
 }
 
-const maxCachedPubKeys = 16
+// maxCachedPubKeys is the number of cache entries we store.
+//
+// Due to consistent misuse of the PublicKeyCallback API, we have reduced this
+// to 1, such that the only key in the cache is the most recently seen one. This
+// forces the behavior that the last call to PublicKeyCallback will always be
+// with the key that is used for authentication.
+const maxCachedPubKeys = 1
 
 // pubKeyCache caches tests for public keys.  Since SSH clients
 // will query whether a public key is acceptable before attempting to
@@ -179,9 +212,10 @@ func (c *pubKeyCache) get(user string, pubKeyData []byte) (cachedPubKey, bool) {
 
 // add adds the given tuple to the cache.
 func (c *pubKeyCache) add(candidate cachedPubKey) {
-	if len(c.keys) < maxCachedPubKeys {
-		c.keys = append(c.keys, candidate)
+	if len(c.keys) >= maxCachedPubKeys {
+		c.keys = c.keys[1:]
 	}
+	c.keys = append(c.keys, candidate)
 }
 
 // ServerConn is an authenticated SSH connection, as seen from the
@@ -481,6 +515,10 @@ func (b *BannerError) Error() string {
 }
 
 func (s *connection) serverAuthenticate(config *ServerConfig) (*Permissions, error) {
+	if config.PreAuthConnCallback != nil {
+		config.PreAuthConnCallback(s)
+	}
+
 	sessionID := s.transport.getSessionID()
 	var cache pubKeyCache
 	var perms *Permissions
@@ -488,7 +526,7 @@ func (s *connection) serverAuthenticate(config *ServerConfig) (*Permissions, err
 	authFailures := 0
 	noneAuthCount := 0
 	var authErrs []error
-	var displayedBanner bool
+	var calledBannerCallback bool
 	partialSuccessReturned := false
 	// Set the initial authentication callbacks from the config. They can be
 	// changed if a PartialSuccessError is returned.
@@ -510,8 +548,8 @@ userAuthLoop:
 			if err := s.transport.writePacket(Marshal(discMsg)); err != nil {
 				return nil, err
 			}
-
-			return nil, discMsg
+			authErrs = append(authErrs, discMsg)
+			return nil, &ServerAuthError{Errors: authErrs}
 		}
 
 		var userAuthReq userAuthRequestMsg
@@ -535,14 +573,10 @@ userAuthLoop:
 
 		s.user = userAuthReq.User
 
-		if !displayedBanner && config.BannerCallback != nil {
-			displayedBanner = true
-			msg := config.BannerCallback(s)
-			if msg != "" {
-				bannerMsg := &userAuthBannerMsg{
-					Message: msg,
-				}
-				if err := s.transport.writePacket(Marshal(bannerMsg)); err != nil {
+		if !calledBannerCallback && config.BannerCallback != nil {
+			calledBannerCallback = true
+			if msg := config.BannerCallback(s); msg != "" {
+				if err := s.SendAuthBanner(msg); err != nil {
 					return nil, err
 				}
 			}
@@ -755,10 +789,7 @@ userAuthLoop:
 		var bannerErr *BannerError
 		if errors.As(authErr, &bannerErr) {
 			if bannerErr.Message != "" {
-				bannerMsg := &userAuthBannerMsg{
-					Message: bannerErr.Message,
-				}
-				if err := s.transport.writePacket(Marshal(bannerMsg)); err != nil {
+				if err := s.SendAuthBanner(bannerErr.Message); err != nil {
 					return nil, err
 				}
 			}

ssh/server_test.go

@@ -5,6 +5,7 @@
 package ssh
 
 import (
+	"bytes"
 	"errors"
 	"fmt"
 	"io"
@@ -299,6 +300,140 @@ func TestBannerError(t *testing.T) {
 	}
 }
 
+func TestPublicKeyCallbackLastSeen(t *testing.T) {
+	var lastSeenKey PublicKey
+
+	c1, c2, err := netPipe()
+	if err != nil {
+		t.Fatalf("netPipe: %v", err)
+	}
+	defer c1.Close()
+	defer c2.Close()
+	serverConf := &ServerConfig{
+		PublicKeyCallback: func(conn ConnMetadata, key PublicKey) (*Permissions, error) {
+			lastSeenKey = key
+			fmt.Printf("seen %#v\n", key)
+			if _, ok := key.(*dsaPublicKey); !ok {
+				return nil, errors.New("nope")
+			}
+			return nil, nil
+		},
+	}
+	serverConf.AddHostKey(testSigners["ecdsap256"])
+
+	done := make(chan struct{})
+	go func() {
+		defer close(done)
+		NewServerConn(c1, serverConf)
+	}()
+
+	clientConf := ClientConfig{
+		User: "user",
+		Auth: []AuthMethod{
+			PublicKeys(testSigners["rsa"], testSigners["dsa"], testSigners["ed25519"]),
+		},
+		HostKeyCallback: InsecureIgnoreHostKey(),
+	}
+
+	_, _, _, err = NewClientConn(c2, "", &clientConf)
+	if err != nil {
+		t.Fatal(err)
+	}
+	<-done
+
+	expectedPublicKey := testSigners["dsa"].PublicKey().Marshal()
+	lastSeenMarshalled := lastSeenKey.Marshal()
+	if !bytes.Equal(lastSeenMarshalled, expectedPublicKey) {
+		t.Errorf("unexpected key: got %#v, want %#v", lastSeenKey, testSigners["dsa"].PublicKey())
+	}
+}
+
+func TestPreAuthConnAndBanners(t *testing.T) {
+	testDone := make(chan struct{})
+	defer close(testDone)
+
+	authConnc := make(chan ServerPreAuthConn, 1)
+	serverConfig := &ServerConfig{
+		PreAuthConnCallback: func(c ServerPreAuthConn) {
+			t.Logf("got ServerPreAuthConn: %v", c)
+			authConnc <- c // for use later in the test
+			for _, s := range []string{"hello1", "hello2"} {
+				if err := c.SendAuthBanner(s); err != nil {
+					t.Errorf("failed to send banner %q: %v", s, err)
+				}
+			}
+			// Now start a goroutine to spam SendAuthBanner in hopes
+			// of hitting a race.
+			go func() {
+				for {
+					select {
+					case <-testDone:
+						return
+					default:
+						if err := c.SendAuthBanner("attempted-race"); err != nil && err != errSendBannerPhase {
+							t.Errorf("unexpected error from SendAuthBanner: %v", err)
+						}
+						time.Sleep(5 * time.Millisecond)
+					}
+				}
+			}()
+		},
+		NoClientAuth: true,
+		NoClientAuthCallback: func(ConnMetadata) (*Permissions, error) {
+			t.Logf("got NoClientAuthCallback")
+			return &Permissions{}, nil
+		},
+	}
+	serverConfig.AddHostKey(testSigners["rsa"])
+
+	var banners []string
+	clientConfig := &ClientConfig{
+		User:            "test",
+		HostKeyCallback: InsecureIgnoreHostKey(),
+		BannerCallback: func(msg string) error {
+			if msg != "attempted-race" {
+				banners = append(banners, msg)
+			}
+			return nil
+		},
+	}
+
+	c1, c2, err := netPipe()
+	if err != nil {
+		t.Fatalf("netPipe: %v", err)
+	}
+	defer c1.Close()
+	defer c2.Close()
+	go newServer(c1, serverConfig)
+	c, _, _, err := NewClientConn(c2, "", clientConfig)
+	if err != nil {
+		t.Fatalf("client connection failed: %v", err)
+	}
+	defer c.Close()
+
+	wantBanners := []string{
+		"hello1",
+		"hello2",
+	}
+	if !reflect.DeepEqual(banners, wantBanners) {
+		t.Errorf("got banners:\n%q\nwant banners:\n%q", banners, wantBanners)
+	}
+
+	// Now that we're authenticated, verify that use of SendBanner
+	// is an error.
+	var bc ServerPreAuthConn
+	select {
+	case bc = <-authConnc:
+	default:
+		t.Fatal("expected ServerPreAuthConn")
+	}
+	if err := bc.SendAuthBanner("wrong-phase"); err == nil {
+		t.Error("unexpected success of SendAuthBanner after authentication")
+	} else if err != errSendBannerPhase {
+		t.Errorf("unexpected error: %v; want %v", err, errSendBannerPhase)
+	}
+}
+
 type markerConn struct {
 	closed uint32
 	used   uint32

ssh/tcpip.go

@@ -459,7 +459,7 @@ func (c *Client) dial(laddr string, lport int, raddr string, rport int) (Channel
 		return nil, err
 	}
 	go DiscardRequests(in)
-	return ch, err
+	return ch, nil
 }
 
 type tcpChan struct {

x509roots/fallback/fallback.go

@@ -2,8 +2,6 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
-//go:build go1.20
-
 // Package fallback embeds a set of fallback X.509 trusted roots in the
 // application by automatically invoking [x509.SetFallbackRoots]. This allows
 // the application to work correctly even if the operating system does not
@@ -24,8 +22,19 @@ import "crypto/x509"
 
 func init() {
 	p := x509.NewCertPool()
-	for _, c := range bundle {
-		p.AddCert(c)
+	for _, c := range parsedCertificates {
+		if len(c.constraints) == 0 {
+			p.AddCert(c.cert)
+		} else {
+			p.AddCertWithConstraint(c.cert, func(chain []*x509.Certificate) error {
+				for _, constraint := range c.constraints {
+					if err := constraint(chain); err != nil {
+						return err
+					}
+				}
+				return nil
+			})
+		}
 	}
 	x509.SetFallbackRoots(p)
 }

x509roots/fallback/go.mod

@@ -1,3 +1,3 @@
 module golang.org/x/crypto/x509roots/fallback
 
-go 1.20
+go 1.23.0

x509roots/gen_fallback_bundle.go

@@ -21,46 +21,77 @@ import (
 	"net/http"
 	"os"
 	"sort"
+	"time"
 
 	"golang.org/x/crypto/x509roots/nss"
 )
 
 const tmpl = `// Code generated by gen_fallback_bundle.go; DO NOT EDIT.
 
-//go:build go1.20
-
 package fallback
 
-import "crypto/x509"
-import "encoding/pem"
+import (
+	"crypto/x509"
+	"encoding/pem"
+	"fmt"
+	"time"
+)
 
-func mustParse(b []byte) []*x509.Certificate {
-	var roots []*x509.Certificate
-	for len(b) > 0 {
-		var block *pem.Block
-		block, b = pem.Decode(b)
+type unparsedCertificate struct {
+	cn         string
+	sha256Hash string
+	pem        string
+
+	// possible constraints
+	distrustAfter string
+}
+
+type parsedCertificate struct {
+	cert        *x509.Certificate
+	constraints []func([]*x509.Certificate) error
+}
+
+func mustParse(unparsedCerts []unparsedCertificate) []parsedCertificate {
+	var b []parsedCertificate
+	for _, unparsed := range unparsedCerts {
+		block, rest := pem.Decode([]byte(unparsed.pem))
 		if block == nil {
-			break
+			panic(fmt.Sprintf("unexpected nil PEM block for %q", unparsed.cn))
+		}
+		if len(rest) != 0 {
+			panic(fmt.Sprintf("unexpected trailing data in PEM for %q", unparsed.cn))
 		}
 		if block.Type != "CERTIFICATE" {
-			panic("unexpected PEM block type: " + block.Type)
+			panic(fmt.Sprintf("unexpected PEM block type for %q: %s", unparsed.cn, block.Type))
 		}
 		cert, err := x509.ParseCertificate(block.Bytes)
 		if err != nil {
 			panic(err)
 		}
-		roots = append(roots, cert)
+		parsed := parsedCertificate{cert: cert}
+		// parse possible constraints, this should check all fields of unparsedCertificate.
+		if unparsed.distrustAfter != "" {
+			distrustAfter, err := time.Parse(time.RFC3339, unparsed.distrustAfter)
+			if err != nil {
+				panic(fmt.Sprintf("failed to parse distrustAfter %q: %s", unparsed.distrustAfter, err))
+			}
+			parsed.constraints = append(parsed.constraints, func(chain []*x509.Certificate) error {
+				for _, c := range chain {
+					if c.NotBefore.After(distrustAfter) {
+						return fmt.Errorf("certificate issued after distrust-after date %q", distrustAfter)
+					}
+				}
+				return nil
+			})
+		}
+		b = append(b, parsed)
 	}
-	return roots
+	return b
 }
 
-var bundle = mustParse([]byte(pemRoots))
-
-// Format of the PEM list is:
-//   * Subject common name
-//   * SHA256 hash
-//   * PEM block
+var parsedCertificates = mustParse(unparsedCertificates)
 
+var unparsedCertificates = []unparsedCertificate{
 `
 
 var (
@@ -124,19 +155,33 @@ func main() {
 
 	b := new(bytes.Buffer)
 	b.WriteString(tmpl)
-	fmt.Fprintln(b, "const pemRoots = `")
 	for _, c := range certs {
-		if len(c.Constraints) > 0 {
-			// Until the constrained roots API lands, skip anything that has any
-			// additional constraints. Once that API is available, we can add
-			// build constraints that support both the current version and the
-			// new version.
+		var constraints []string
+		var skip bool
+		for _, constraint := range c.Constraints {
+			switch t := constraint.(type) {
+			case nss.DistrustAfter:
+				constraints = append(constraints, fmt.Sprintf("distrustAfter: \"%s\",", time.Time(t).Format(time.RFC3339)))
+			default:
+				// If we encounter any constraints we don't support, skip the certificate.
+				skip = true
+				break
+			}
+		}
+		if skip {
 			continue
 		}
-		fmt.Fprintf(b, "# %s\n# %x\n", c.X509.Subject.String(), sha256.Sum256(c.X509.Raw))
-		pem.Encode(b, &pem.Block{Type: "CERTIFICATE", Bytes: c.X509.Raw})
+		fmt.Fprintf(b, "{\ncn: %q,\nsha256Hash: \"%x\",\npem: `%s`,\n",
+			c.X509.Subject.String(),
+			sha256.Sum256(c.X509.Raw),
+			string(pem.EncodeToMemory(&pem.Block{Type: "CERTIFICATE", Bytes: c.X509.Raw})),
+		)
+		for _, constraint := range constraints {
+			fmt.Fprintln(b, constraint)
+		}
+		fmt.Fprintln(b, "},")
 	}
-	fmt.Fprintln(b, "`")
+	fmt.Fprintln(b, "}")
 
 	formatted, err := format.Source(b.Bytes())
 	if err != nil {