diff --git a/src/cmd/compile/internal/types2/stdlib_test.go b/src/cmd/compile/internal/types2/stdlib_test.go
index d91d60d1241..00f9fab6139 100644
--- a/src/cmd/compile/internal/types2/stdlib_test.go
+++ b/src/cmd/compile/internal/types2/stdlib_test.go
@@ -357,6 +357,7 @@ var excluded = map[string]bool{
 	// go.dev/issue/46027: some imports are missing for this submodule.
 	"crypto/internal/bigmod/_asm":             true,
 	"crypto/internal/edwards25519/field/_asm": true,
+	"crypto/md5/_asm":                         true,
 	"crypto/sha1/_asm":                        true,
 	"crypto/sha256/_asm":                      true,
 	"crypto/sha512/_asm":                      true,
diff --git a/src/crypto/md5/_asm/go.mod b/src/crypto/md5/_asm/go.mod
new file mode 100644
index 00000000000..04e6e30a84f
--- /dev/null
+++ b/src/crypto/md5/_asm/go.mod
@@ -0,0 +1,11 @@
+module std/crypto/md5/_asm
+
+go 1.24
+
+require github.com/mmcloughlin/avo v0.6.0
+
+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
+)
diff --git a/src/crypto/md5/_asm/go.sum b/src/crypto/md5/_asm/go.sum
new file mode 100644
index 00000000000..76af484b2eb
--- /dev/null
+++ b/src/crypto/md5/_asm/go.sum
@@ -0,0 +1,8 @@
+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/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=
diff --git a/src/crypto/md5/_asm/md5block_amd64_asm.go b/src/crypto/md5/_asm/md5block_amd64_asm.go
new file mode 100644
index 00000000000..45133be83b7
--- /dev/null
+++ b/src/crypto/md5/_asm/md5block_amd64_asm.go
@@ -0,0 +1,228 @@
+// 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.
+
+// Original source:
+//	http://www.zorinaq.com/papers/md5-amd64.html
+//	http://www.zorinaq.com/papers/md5-amd64.tar.bz2
+//
+// Translated from Perl generating GNU assembly into
+// #defines generating 6a assembly by the Go Authors.
+
+package main
+
+import (
+	. "github.com/mmcloughlin/avo/build"
+	. "github.com/mmcloughlin/avo/operand"
+	. "github.com/mmcloughlin/avo/reg"
+)
+
+//go:generate go run . -out ../md5block_amd64.s -pkg md5
+
+func main() {
+	Package("crypto/md5")
+	ConstraintExpr("!purego")
+	block()
+	Generate()
+}
+
+// MD5 optimized for AMD64.
+//
+// Author: Marc Bevand <bevand_m (at) epita.fr>
+// Licence: I hereby disclaim the copyright on this code and place it
+// in the public domain.
+func block() {
+	Implement("block")
+	Attributes(NOSPLIT)
+	AllocLocal(8)
+
+	Load(Param("dig"), RBP)
+	Load(Param("p").Base(), RSI)
+	Load(Param("p").Len(), RDX)
+	SHRQ(Imm(6), RDX)
+	SHLQ(Imm(6), RDX)
+
+	LEAQ(Mem{Base: SI, Index: DX, Scale: 1}, RDI)
+	MOVL(Mem{Base: BP}.Offset(0*4), EAX)
+	MOVL(Mem{Base: BP}.Offset(1*4), EBX)
+	MOVL(Mem{Base: BP}.Offset(2*4), ECX)
+	MOVL(Mem{Base: BP}.Offset(3*4), EDX)
+	MOVL(Imm(0xffffffff), R11L)
+
+	CMPQ(RSI, RDI)
+	JEQ(LabelRef("end"))
+
+	loop()
+	end()
+}
+
+func loop() {
+	Label("loop")
+	MOVL(EAX, R12L)
+	MOVL(EBX, R13L)
+	MOVL(ECX, R14L)
+	MOVL(EDX, R15L)
+
+	MOVL(Mem{Base: SI}.Offset(0*4), R8L)
+	MOVL(EDX, R9L)
+
+	ROUND1(EAX, EBX, ECX, EDX, 1, 0xd76aa478, 7)
+	ROUND1(EDX, EAX, EBX, ECX, 2, 0xe8c7b756, 12)
+	ROUND1(ECX, EDX, EAX, EBX, 3, 0x242070db, 17)
+	ROUND1(EBX, ECX, EDX, EAX, 4, 0xc1bdceee, 22)
+	ROUND1(EAX, EBX, ECX, EDX, 5, 0xf57c0faf, 7)
+	ROUND1(EDX, EAX, EBX, ECX, 6, 0x4787c62a, 12)
+	ROUND1(ECX, EDX, EAX, EBX, 7, 0xa8304613, 17)
+	ROUND1(EBX, ECX, EDX, EAX, 8, 0xfd469501, 22)
+	ROUND1(EAX, EBX, ECX, EDX, 9, 0x698098d8, 7)
+	ROUND1(EDX, EAX, EBX, ECX, 10, 0x8b44f7af, 12)
+	ROUND1(ECX, EDX, EAX, EBX, 11, 0xffff5bb1, 17)
+	ROUND1(EBX, ECX, EDX, EAX, 12, 0x895cd7be, 22)
+	ROUND1(EAX, EBX, ECX, EDX, 13, 0x6b901122, 7)
+	ROUND1(EDX, EAX, EBX, ECX, 14, 0xfd987193, 12)
+	ROUND1(ECX, EDX, EAX, EBX, 15, 0xa679438e, 17)
+	ROUND1(EBX, ECX, EDX, EAX, 1, 0x49b40821, 22)
+
+	MOVL(EDX, R9L)
+	MOVL(EDX, R10L)
+
+	ROUND2(EAX, EBX, ECX, EDX, 6, 0xf61e2562, 5)
+	ROUND2(EDX, EAX, EBX, ECX, 11, 0xc040b340, 9)
+	ROUND2(ECX, EDX, EAX, EBX, 0, 0x265e5a51, 14)
+	ROUND2(EBX, ECX, EDX, EAX, 5, 0xe9b6c7aa, 20)
+	ROUND2(EAX, EBX, ECX, EDX, 10, 0xd62f105d, 5)
+	ROUND2(EDX, EAX, EBX, ECX, 15, 0x2441453, 9)
+	ROUND2(ECX, EDX, EAX, EBX, 4, 0xd8a1e681, 14)
+	ROUND2(EBX, ECX, EDX, EAX, 9, 0xe7d3fbc8, 20)
+	ROUND2(EAX, EBX, ECX, EDX, 14, 0x21e1cde6, 5)
+	ROUND2(EDX, EAX, EBX, ECX, 3, 0xc33707d6, 9)
+	ROUND2(ECX, EDX, EAX, EBX, 8, 0xf4d50d87, 14)
+	ROUND2(EBX, ECX, EDX, EAX, 13, 0x455a14ed, 20)
+	ROUND2(EAX, EBX, ECX, EDX, 2, 0xa9e3e905, 5)
+	ROUND2(EDX, EAX, EBX, ECX, 7, 0xfcefa3f8, 9)
+	ROUND2(ECX, EDX, EAX, EBX, 12, 0x676f02d9, 14)
+	ROUND2(EBX, ECX, EDX, EAX, 5, 0x8d2a4c8a, 20)
+
+	MOVL(ECX, R9L)
+
+	ROUND3FIRST(EAX, EBX, ECX, EDX, 8, 0xfffa3942, 4)
+	ROUND3(EDX, EAX, EBX, ECX, 11, 0x8771f681, 11)
+	ROUND3(ECX, EDX, EAX, EBX, 14, 0x6d9d6122, 16)
+	ROUND3(EBX, ECX, EDX, EAX, 1, 0xfde5380c, 23)
+	ROUND3(EAX, EBX, ECX, EDX, 4, 0xa4beea44, 4)
+	ROUND3(EDX, EAX, EBX, ECX, 7, 0x4bdecfa9, 11)
+	ROUND3(ECX, EDX, EAX, EBX, 10, 0xf6bb4b60, 16)
+	ROUND3(EBX, ECX, EDX, EAX, 13, 0xbebfbc70, 23)
+	ROUND3(EAX, EBX, ECX, EDX, 0, 0x289b7ec6, 4)
+	ROUND3(EDX, EAX, EBX, ECX, 3, 0xeaa127fa, 11)
+	ROUND3(ECX, EDX, EAX, EBX, 6, 0xd4ef3085, 16)
+	ROUND3(EBX, ECX, EDX, EAX, 9, 0x4881d05, 23)
+	ROUND3(EAX, EBX, ECX, EDX, 12, 0xd9d4d039, 4)
+	ROUND3(EDX, EAX, EBX, ECX, 15, 0xe6db99e5, 11)
+	ROUND3(ECX, EDX, EAX, EBX, 2, 0x1fa27cf8, 16)
+	ROUND3(EBX, ECX, EDX, EAX, 0, 0xc4ac5665, 23)
+
+	MOVL(R11L, R9L)
+	XORL(EDX, R9L)
+
+	ROUND4(EAX, EBX, ECX, EDX, 7, 0xf4292244, 6)
+	ROUND4(EDX, EAX, EBX, ECX, 14, 0x432aff97, 10)
+	ROUND4(ECX, EDX, EAX, EBX, 5, 0xab9423a7, 15)
+	ROUND4(EBX, ECX, EDX, EAX, 12, 0xfc93a039, 21)
+	ROUND4(EAX, EBX, ECX, EDX, 3, 0x655b59c3, 6)
+	ROUND4(EDX, EAX, EBX, ECX, 10, 0x8f0ccc92, 10)
+	ROUND4(ECX, EDX, EAX, EBX, 1, 0xffeff47d, 15)
+	ROUND4(EBX, ECX, EDX, EAX, 8, 0x85845dd1, 21)
+	ROUND4(EAX, EBX, ECX, EDX, 15, 0x6fa87e4f, 6)
+	ROUND4(EDX, EAX, EBX, ECX, 6, 0xfe2ce6e0, 10)
+	ROUND4(ECX, EDX, EAX, EBX, 13, 0xa3014314, 15)
+	ROUND4(EBX, ECX, EDX, EAX, 4, 0x4e0811a1, 21)
+	ROUND4(EAX, EBX, ECX, EDX, 11, 0xf7537e82, 6)
+	ROUND4(EDX, EAX, EBX, ECX, 2, 0xbd3af235, 10)
+	ROUND4(ECX, EDX, EAX, EBX, 9, 0x2ad7d2bb, 15)
+	ROUND4(EBX, ECX, EDX, EAX, 0, 0xeb86d391, 21)
+
+	ADDL(R12L, EAX)
+	ADDL(R13L, EBX)
+	ADDL(R14L, ECX)
+	ADDL(R15L, EDX)
+
+	ADDQ(Imm(64), RSI)
+	CMPQ(RSI, RDI)
+	JB(LabelRef("loop"))
+}
+
+func end() {
+	Label("end")
+	MOVL(EAX, Mem{Base: BP}.Offset(0*4))
+	MOVL(EBX, Mem{Base: BP}.Offset(1*4))
+	MOVL(ECX, Mem{Base: BP}.Offset(2*4))
+	MOVL(EDX, Mem{Base: BP}.Offset(3*4))
+	RET()
+}
+
+func ROUND1(a, b, c, d GPPhysical, index int, konst, shift uint64) {
+	XORL(c, R9L)
+	ADDL(Imm(konst), a)
+	ADDL(R8L, a)
+	ANDL(b, R9L)
+	XORL(d, R9L)
+	MOVL(Mem{Base: SI}.Offset(index*4), R8L)
+	ADDL(R9L, a)
+	ROLL(Imm(shift), a)
+	MOVL(c, R9L)
+	ADDL(b, a)
+}
+
+// Uses https://github.com/animetosho/md5-optimisation#dependency-shortcut-in-g-function
+func ROUND2(a, b, c, d GPPhysical, index int, konst, shift uint64) {
+	XORL(R11L, R9L)
+	ADDL(Imm(konst), a)
+	ADDL(R8L, a)
+	ANDL(b, R10L)
+	ANDL(c, R9L)
+	MOVL(Mem{Base: SI}.Offset(index*4), R8L)
+	ADDL(R9L, a)
+	ADDL(R10L, a)
+	MOVL(c, R9L)
+	MOVL(c, R10L)
+	ROLL(Imm(shift), a)
+	ADDL(b, a)
+}
+
+// Uses https://github.com/animetosho/md5-optimisation#h-function-re-use
+func ROUND3FIRST(a, b, c, d GPPhysical, index int, konst, shift uint64) {
+	MOVL(d, R9L)
+	XORL(c, R9L)
+	XORL(b, R9L)
+	ADDL(Imm(konst), a)
+	ADDL(R8L, a)
+	MOVL(Mem{Base: SI}.Offset(index*4), R8L)
+	ADDL(R9L, a)
+	ROLL(Imm(shift), a)
+	ADDL(b, a)
+}
+
+func ROUND3(a, b, c, d GPPhysical, index int, konst, shift uint64) {
+	XORL(a, R9L)
+	XORL(b, R9L)
+	ADDL(Imm(konst), a)
+	ADDL(R8L, a)
+	MOVL(Mem{Base: SI}.Offset(index*4), R8L)
+	ADDL(R9L, a)
+	ROLL(Imm(shift), a)
+	ADDL(b, a)
+}
+
+func ROUND4(a, b, c, d GPPhysical, index int, konst, shift uint64) {
+	ADDL(Imm(konst), a)
+	ADDL(R8L, a)
+	ORL(b, R9L)
+	XORL(c, R9L)
+	ADDL(R9L, a)
+	MOVL(Mem{Base: SI}.Offset(index*4), R8L)
+	MOVL(Imm(0xffffffff), R9L)
+	ROLL(Imm(shift), a)
+	XORL(c, R9L)
+	ADDL(b, a)
+}
diff --git a/src/crypto/md5/md5block_amd64.s b/src/crypto/md5/md5block_amd64.s
index 652506ae27f..979eb22dbf7 100644
--- a/src/crypto/md5/md5block_amd64.s
+++ b/src/crypto/md5/md5block_amd64.s
@@ -1,197 +1,689 @@
-// Original source:
-//	http://www.zorinaq.com/papers/md5-amd64.html
-//	http://www.zorinaq.com/papers/md5-amd64.tar.bz2
-//
-// Translated from Perl generating GNU assembly into
-// #defines generating 6a assembly by the Go Authors.
+// Code generated by command: go run md5block_amd64_asm.go -out ../md5block_amd64.s -pkg md5. DO NOT EDIT.
 
 //go:build !purego
 
 #include "textflag.h"
 
-// MD5 optimized for AMD64.
-//
-// Author: Marc Bevand <bevand_m (at) epita.fr>
-// Licence: I hereby disclaim the copyright on this code and place it
-// in the public domain.
-
-TEXT	·block(SB),NOSPLIT,$8-32
-	MOVQ	dig+0(FP),	BP
-	MOVQ	p+8(FP),	SI
-	MOVQ	p_len+16(FP), DX
-	SHRQ	$6,		DX
-	SHLQ	$6,		DX
-
-	LEAQ	(SI)(DX*1),	DI
-	MOVL	(0*4)(BP),	AX
-	MOVL	(1*4)(BP),	BX
-	MOVL	(2*4)(BP),	CX
-	MOVL	(3*4)(BP),	DX
-	MOVL	$0xffffffff,	R11
-
-	CMPQ	SI,		DI
-	JEQ	end
+// func block(dig *digest, p []byte)
+TEXT ·block(SB), NOSPLIT, $8-32
+	MOVQ dig+0(FP), BP
+	MOVQ p_base+8(FP), SI
+	MOVQ p_len+16(FP), DX
+	SHRQ $0x06, DX
+	SHLQ $0x06, DX
+	LEAQ (SI)(DX*1), DI
+	MOVL (BP), AX
+	MOVL 4(BP), BX
+	MOVL 8(BP), CX
+	MOVL 12(BP), DX
+	MOVL $0xffffffff, R11
+	CMPQ SI, DI
+	JEQ  end
 
 loop:
-	MOVL	AX,		R12
-	MOVL	BX,		R13
-	MOVL	CX,		R14
-	MOVL	DX,		R15
-
-	MOVL	(0*4)(SI),	R8
-	MOVL	DX,		R9
-
-#define ROUND1(a, b, c, d, index, const, shift) \
-	XORL	c, R9; \
-	ADDL	$const, a; \
-	ADDL	R8, a; \
-	ANDL	b, R9; \
-	XORL	d, R9; \
-	MOVL	(index*4)(SI), R8; \
-	ADDL	R9, a; \
-	ROLL	$shift, a; \
-	MOVL	c, R9; \
-	ADDL	b, a
-
-	ROUND1(AX,BX,CX,DX, 1,0xd76aa478, 7);
-	ROUND1(DX,AX,BX,CX, 2,0xe8c7b756,12);
-	ROUND1(CX,DX,AX,BX, 3,0x242070db,17);
-	ROUND1(BX,CX,DX,AX, 4,0xc1bdceee,22);
-	ROUND1(AX,BX,CX,DX, 5,0xf57c0faf, 7);
-	ROUND1(DX,AX,BX,CX, 6,0x4787c62a,12);
-	ROUND1(CX,DX,AX,BX, 7,0xa8304613,17);
-	ROUND1(BX,CX,DX,AX, 8,0xfd469501,22);
-	ROUND1(AX,BX,CX,DX, 9,0x698098d8, 7);
-	ROUND1(DX,AX,BX,CX,10,0x8b44f7af,12);
-	ROUND1(CX,DX,AX,BX,11,0xffff5bb1,17);
-	ROUND1(BX,CX,DX,AX,12,0x895cd7be,22);
-	ROUND1(AX,BX,CX,DX,13,0x6b901122, 7);
-	ROUND1(DX,AX,BX,CX,14,0xfd987193,12);
-	ROUND1(CX,DX,AX,BX,15,0xa679438e,17);
-	ROUND1(BX,CX,DX,AX, 1,0x49b40821,22);
-
-	MOVL	DX,		R9
-	MOVL	DX,		R10
-
-// Uses https://github.com/animetosho/md5-optimisation#dependency-shortcut-in-g-function
-
-#define ROUND2(a, b, c, d, index, const, shift) \
-	XORL	R11, R9; \
-	ADDL	$const,	a; \
-	ADDL	R8,	a; \
-	ANDL	b,		R10; \
-	ANDL	c,		R9; \
-	MOVL	(index*4)(SI),R8; \
-	ADDL	R9,	a; \
-	ADDL	R10,	a; \
-	MOVL	c,		R9; \
-	MOVL	c,		R10; \
-	ROLL	$shift,	a; \
-	ADDL	b,		a
-
-	ROUND2(AX,BX,CX,DX, 6,0xf61e2562, 5);
-	ROUND2(DX,AX,BX,CX,11,0xc040b340, 9);
-	ROUND2(CX,DX,AX,BX, 0,0x265e5a51,14);
-	ROUND2(BX,CX,DX,AX, 5,0xe9b6c7aa,20);
-	ROUND2(AX,BX,CX,DX,10,0xd62f105d, 5);
-	ROUND2(DX,AX,BX,CX,15, 0x2441453, 9);
-	ROUND2(CX,DX,AX,BX, 4,0xd8a1e681,14);
-	ROUND2(BX,CX,DX,AX, 9,0xe7d3fbc8,20);
-	ROUND2(AX,BX,CX,DX,14,0x21e1cde6, 5);
-	ROUND2(DX,AX,BX,CX, 3,0xc33707d6, 9);
-	ROUND2(CX,DX,AX,BX, 8,0xf4d50d87,14);
-	ROUND2(BX,CX,DX,AX,13,0x455a14ed,20);
-	ROUND2(AX,BX,CX,DX, 2,0xa9e3e905, 5);
-	ROUND2(DX,AX,BX,CX, 7,0xfcefa3f8, 9);
-	ROUND2(CX,DX,AX,BX,12,0x676f02d9,14);
-	ROUND2(BX,CX,DX,AX, 5,0x8d2a4c8a,20);
-
-	MOVL	CX,		R9
-
-// Uses https://github.com/animetosho/md5-optimisation#h-function-re-use
-
-#define ROUND3FIRST(a, b, c, d, index, const, shift) \
-	MOVL	d,		R9; \
-	XORL	c,		R9; \
-	XORL	b,		R9; \
-	ADDL	$const,	a; \
-	ADDL	R8,		a; \
-	MOVL	(index*4)(SI),R8; \
-	ADDL	R9,		a; \
-	ROLL	$shift,		a; \
-	ADDL	b,		a
-
-#define ROUND3(a, b, c, d, index, const, shift) \
-	XORL	a,		R9; \
-	XORL	b,		R9; \
-	ADDL	$const,	a; \
-	ADDL	R8,		a; \
-	MOVL	(index*4)(SI),R8; \
-	ADDL	R9,		a; \
-	ROLL	$shift,		a; \
-	ADDL	b,		a
-
-	ROUND3FIRST(AX,BX,CX,DX, 8,0xfffa3942, 4);
-	ROUND3(DX,AX,BX,CX,11,0x8771f681,11);
-	ROUND3(CX,DX,AX,BX,14,0x6d9d6122,16);
-	ROUND3(BX,CX,DX,AX, 1,0xfde5380c,23);
-	ROUND3(AX,BX,CX,DX, 4,0xa4beea44, 4);
-	ROUND3(DX,AX,BX,CX, 7,0x4bdecfa9,11);
-	ROUND3(CX,DX,AX,BX,10,0xf6bb4b60,16);
-	ROUND3(BX,CX,DX,AX,13,0xbebfbc70,23);
-	ROUND3(AX,BX,CX,DX, 0,0x289b7ec6, 4);
-	ROUND3(DX,AX,BX,CX, 3,0xeaa127fa,11);
-	ROUND3(CX,DX,AX,BX, 6,0xd4ef3085,16);
-	ROUND3(BX,CX,DX,AX, 9, 0x4881d05,23);
-	ROUND3(AX,BX,CX,DX,12,0xd9d4d039, 4);
-	ROUND3(DX,AX,BX,CX,15,0xe6db99e5,11);
-	ROUND3(CX,DX,AX,BX, 2,0x1fa27cf8,16);
-	ROUND3(BX,CX,DX,AX, 0,0xc4ac5665,23);
-
-	MOVL	R11,	R9
-	XORL	DX,		R9
-
-#define ROUND4(a, b, c, d, index, const, shift) \
-	ADDL	$const,	a; \
-	ADDL	R8,		a; \
-	ORL		b,		R9; \
-	XORL	c,		R9; \
-	ADDL	R9,		a; \
-	MOVL	(index*4)(SI),R8; \
-	MOVL	$0xffffffff,	R9; \
-	ROLL	$shift,		a; \
-	XORL	c,		R9; \
-	ADDL	b,		a
-
-	ROUND4(AX,BX,CX,DX, 7,0xf4292244, 6);
-	ROUND4(DX,AX,BX,CX,14,0x432aff97,10);
-	ROUND4(CX,DX,AX,BX, 5,0xab9423a7,15);
-	ROUND4(BX,CX,DX,AX,12,0xfc93a039,21);
-	ROUND4(AX,BX,CX,DX, 3,0x655b59c3, 6);
-	ROUND4(DX,AX,BX,CX,10,0x8f0ccc92,10);
-	ROUND4(CX,DX,AX,BX, 1,0xffeff47d,15);
-	ROUND4(BX,CX,DX,AX, 8,0x85845dd1,21);
-	ROUND4(AX,BX,CX,DX,15,0x6fa87e4f, 6);
-	ROUND4(DX,AX,BX,CX, 6,0xfe2ce6e0,10);
-	ROUND4(CX,DX,AX,BX,13,0xa3014314,15);
-	ROUND4(BX,CX,DX,AX, 4,0x4e0811a1,21);
-	ROUND4(AX,BX,CX,DX,11,0xf7537e82, 6);
-	ROUND4(DX,AX,BX,CX, 2,0xbd3af235,10);
-	ROUND4(CX,DX,AX,BX, 9,0x2ad7d2bb,15);
-	ROUND4(BX,CX,DX,AX, 0,0xeb86d391,21);
-
-	ADDL	R12,	AX
-	ADDL	R13,	BX
-	ADDL	R14,	CX
-	ADDL	R15,	DX
-
-	ADDQ	$64,		SI
-	CMPQ	SI,		DI
-	JB	loop
+	MOVL AX, R12
+	MOVL BX, R13
+	MOVL CX, R14
+	MOVL DX, R15
+	MOVL (SI), R8
+	MOVL DX, R9
+	XORL CX, R9
+	ADDL $0xd76aa478, AX
+	ADDL R8, AX
+	ANDL BX, R9
+	XORL DX, R9
+	MOVL 4(SI), R8
+	ADDL R9, AX
+	ROLL $0x07, AX
+	MOVL CX, R9
+	ADDL BX, AX
+	XORL BX, R9
+	ADDL $0xe8c7b756, DX
+	ADDL R8, DX
+	ANDL AX, R9
+	XORL CX, R9
+	MOVL 8(SI), R8
+	ADDL R9, DX
+	ROLL $0x0c, DX
+	MOVL BX, R9
+	ADDL AX, DX
+	XORL AX, R9
+	ADDL $0x242070db, CX
+	ADDL R8, CX
+	ANDL DX, R9
+	XORL BX, R9
+	MOVL 12(SI), R8
+	ADDL R9, CX
+	ROLL $0x11, CX
+	MOVL AX, R9
+	ADDL DX, CX
+	XORL DX, R9
+	ADDL $0xc1bdceee, BX
+	ADDL R8, BX
+	ANDL CX, R9
+	XORL AX, R9
+	MOVL 16(SI), R8
+	ADDL R9, BX
+	ROLL $0x16, BX
+	MOVL DX, R9
+	ADDL CX, BX
+	XORL CX, R9
+	ADDL $0xf57c0faf, AX
+	ADDL R8, AX
+	ANDL BX, R9
+	XORL DX, R9
+	MOVL 20(SI), R8
+	ADDL R9, AX
+	ROLL $0x07, AX
+	MOVL CX, R9
+	ADDL BX, AX
+	XORL BX, R9
+	ADDL $0x4787c62a, DX
+	ADDL R8, DX
+	ANDL AX, R9
+	XORL CX, R9
+	MOVL 24(SI), R8
+	ADDL R9, DX
+	ROLL $0x0c, DX
+	MOVL BX, R9
+	ADDL AX, DX
+	XORL AX, R9
+	ADDL $0xa8304613, CX
+	ADDL R8, CX
+	ANDL DX, R9
+	XORL BX, R9
+	MOVL 28(SI), R8
+	ADDL R9, CX
+	ROLL $0x11, CX
+	MOVL AX, R9
+	ADDL DX, CX
+	XORL DX, R9
+	ADDL $0xfd469501, BX
+	ADDL R8, BX
+	ANDL CX, R9
+	XORL AX, R9
+	MOVL 32(SI), R8
+	ADDL R9, BX
+	ROLL $0x16, BX
+	MOVL DX, R9
+	ADDL CX, BX
+	XORL CX, R9
+	ADDL $0x698098d8, AX
+	ADDL R8, AX
+	ANDL BX, R9
+	XORL DX, R9
+	MOVL 36(SI), R8
+	ADDL R9, AX
+	ROLL $0x07, AX
+	MOVL CX, R9
+	ADDL BX, AX
+	XORL BX, R9
+	ADDL $0x8b44f7af, DX
+	ADDL R8, DX
+	ANDL AX, R9
+	XORL CX, R9
+	MOVL 40(SI), R8
+	ADDL R9, DX
+	ROLL $0x0c, DX
+	MOVL BX, R9
+	ADDL AX, DX
+	XORL AX, R9
+	ADDL $0xffff5bb1, CX
+	ADDL R8, CX
+	ANDL DX, R9
+	XORL BX, R9
+	MOVL 44(SI), R8
+	ADDL R9, CX
+	ROLL $0x11, CX
+	MOVL AX, R9
+	ADDL DX, CX
+	XORL DX, R9
+	ADDL $0x895cd7be, BX
+	ADDL R8, BX
+	ANDL CX, R9
+	XORL AX, R9
+	MOVL 48(SI), R8
+	ADDL R9, BX
+	ROLL $0x16, BX
+	MOVL DX, R9
+	ADDL CX, BX
+	XORL CX, R9
+	ADDL $0x6b901122, AX
+	ADDL R8, AX
+	ANDL BX, R9
+	XORL DX, R9
+	MOVL 52(SI), R8
+	ADDL R9, AX
+	ROLL $0x07, AX
+	MOVL CX, R9
+	ADDL BX, AX
+	XORL BX, R9
+	ADDL $0xfd987193, DX
+	ADDL R8, DX
+	ANDL AX, R9
+	XORL CX, R9
+	MOVL 56(SI), R8
+	ADDL R9, DX
+	ROLL $0x0c, DX
+	MOVL BX, R9
+	ADDL AX, DX
+	XORL AX, R9
+	ADDL $0xa679438e, CX
+	ADDL R8, CX
+	ANDL DX, R9
+	XORL BX, R9
+	MOVL 60(SI), R8
+	ADDL R9, CX
+	ROLL $0x11, CX
+	MOVL AX, R9
+	ADDL DX, CX
+	XORL DX, R9
+	ADDL $0x49b40821, BX
+	ADDL R8, BX
+	ANDL CX, R9
+	XORL AX, R9
+	MOVL 4(SI), R8
+	ADDL R9, BX
+	ROLL $0x16, BX
+	MOVL DX, R9
+	ADDL CX, BX
+	MOVL DX, R9
+	MOVL DX, R10
+	XORL R11, R9
+	ADDL $0xf61e2562, AX
+	ADDL R8, AX
+	ANDL BX, R10
+	ANDL CX, R9
+	MOVL 24(SI), R8
+	ADDL R9, AX
+	ADDL R10, AX
+	MOVL CX, R9
+	MOVL CX, R10
+	ROLL $0x05, AX
+	ADDL BX, AX
+	XORL R11, R9
+	ADDL $0xc040b340, DX
+	ADDL R8, DX
+	ANDL AX, R10
+	ANDL BX, R9
+	MOVL 44(SI), R8
+	ADDL R9, DX
+	ADDL R10, DX
+	MOVL BX, R9
+	MOVL BX, R10
+	ROLL $0x09, DX
+	ADDL AX, DX
+	XORL R11, R9
+	ADDL $0x265e5a51, CX
+	ADDL R8, CX
+	ANDL DX, R10
+	ANDL AX, R9
+	MOVL (SI), R8
+	ADDL R9, CX
+	ADDL R10, CX
+	MOVL AX, R9
+	MOVL AX, R10
+	ROLL $0x0e, CX
+	ADDL DX, CX
+	XORL R11, R9
+	ADDL $0xe9b6c7aa, BX
+	ADDL R8, BX
+	ANDL CX, R10
+	ANDL DX, R9
+	MOVL 20(SI), R8
+	ADDL R9, BX
+	ADDL R10, BX
+	MOVL DX, R9
+	MOVL DX, R10
+	ROLL $0x14, BX
+	ADDL CX, BX
+	XORL R11, R9
+	ADDL $0xd62f105d, AX
+	ADDL R8, AX
+	ANDL BX, R10
+	ANDL CX, R9
+	MOVL 40(SI), R8
+	ADDL R9, AX
+	ADDL R10, AX
+	MOVL CX, R9
+	MOVL CX, R10
+	ROLL $0x05, AX
+	ADDL BX, AX
+	XORL R11, R9
+	ADDL $0x02441453, DX
+	ADDL R8, DX
+	ANDL AX, R10
+	ANDL BX, R9
+	MOVL 60(SI), R8
+	ADDL R9, DX
+	ADDL R10, DX
+	MOVL BX, R9
+	MOVL BX, R10
+	ROLL $0x09, DX
+	ADDL AX, DX
+	XORL R11, R9
+	ADDL $0xd8a1e681, CX
+	ADDL R8, CX
+	ANDL DX, R10
+	ANDL AX, R9
+	MOVL 16(SI), R8
+	ADDL R9, CX
+	ADDL R10, CX
+	MOVL AX, R9
+	MOVL AX, R10
+	ROLL $0x0e, CX
+	ADDL DX, CX
+	XORL R11, R9
+	ADDL $0xe7d3fbc8, BX
+	ADDL R8, BX
+	ANDL CX, R10
+	ANDL DX, R9
+	MOVL 36(SI), R8
+	ADDL R9, BX
+	ADDL R10, BX
+	MOVL DX, R9
+	MOVL DX, R10
+	ROLL $0x14, BX
+	ADDL CX, BX
+	XORL R11, R9
+	ADDL $0x21e1cde6, AX
+	ADDL R8, AX
+	ANDL BX, R10
+	ANDL CX, R9
+	MOVL 56(SI), R8
+	ADDL R9, AX
+	ADDL R10, AX
+	MOVL CX, R9
+	MOVL CX, R10
+	ROLL $0x05, AX
+	ADDL BX, AX
+	XORL R11, R9
+	ADDL $0xc33707d6, DX
+	ADDL R8, DX
+	ANDL AX, R10
+	ANDL BX, R9
+	MOVL 12(SI), R8
+	ADDL R9, DX
+	ADDL R10, DX
+	MOVL BX, R9
+	MOVL BX, R10
+	ROLL $0x09, DX
+	ADDL AX, DX
+	XORL R11, R9
+	ADDL $0xf4d50d87, CX
+	ADDL R8, CX
+	ANDL DX, R10
+	ANDL AX, R9
+	MOVL 32(SI), R8
+	ADDL R9, CX
+	ADDL R10, CX
+	MOVL AX, R9
+	MOVL AX, R10
+	ROLL $0x0e, CX
+	ADDL DX, CX
+	XORL R11, R9
+	ADDL $0x455a14ed, BX
+	ADDL R8, BX
+	ANDL CX, R10
+	ANDL DX, R9
+	MOVL 52(SI), R8
+	ADDL R9, BX
+	ADDL R10, BX
+	MOVL DX, R9
+	MOVL DX, R10
+	ROLL $0x14, BX
+	ADDL CX, BX
+	XORL R11, R9
+	ADDL $0xa9e3e905, AX
+	ADDL R8, AX
+	ANDL BX, R10
+	ANDL CX, R9
+	MOVL 8(SI), R8
+	ADDL R9, AX
+	ADDL R10, AX
+	MOVL CX, R9
+	MOVL CX, R10
+	ROLL $0x05, AX
+	ADDL BX, AX
+	XORL R11, R9
+	ADDL $0xfcefa3f8, DX
+	ADDL R8, DX
+	ANDL AX, R10
+	ANDL BX, R9
+	MOVL 28(SI), R8
+	ADDL R9, DX
+	ADDL R10, DX
+	MOVL BX, R9
+	MOVL BX, R10
+	ROLL $0x09, DX
+	ADDL AX, DX
+	XORL R11, R9
+	ADDL $0x676f02d9, CX
+	ADDL R8, CX
+	ANDL DX, R10
+	ANDL AX, R9
+	MOVL 48(SI), R8
+	ADDL R9, CX
+	ADDL R10, CX
+	MOVL AX, R9
+	MOVL AX, R10
+	ROLL $0x0e, CX
+	ADDL DX, CX
+	XORL R11, R9
+	ADDL $0x8d2a4c8a, BX
+	ADDL R8, BX
+	ANDL CX, R10
+	ANDL DX, R9
+	MOVL 20(SI), R8
+	ADDL R9, BX
+	ADDL R10, BX
+	MOVL DX, R9
+	MOVL DX, R10
+	ROLL $0x14, BX
+	ADDL CX, BX
+	MOVL CX, R9
+	MOVL DX, R9
+	XORL CX, R9
+	XORL BX, R9
+	ADDL $0xfffa3942, AX
+	ADDL R8, AX
+	MOVL 32(SI), R8
+	ADDL R9, AX
+	ROLL $0x04, AX
+	ADDL BX, AX
+	XORL DX, R9
+	XORL AX, R9
+	ADDL $0x8771f681, DX
+	ADDL R8, DX
+	MOVL 44(SI), R8
+	ADDL R9, DX
+	ROLL $0x0b, DX
+	ADDL AX, DX
+	XORL CX, R9
+	XORL DX, R9
+	ADDL $0x6d9d6122, CX
+	ADDL R8, CX
+	MOVL 56(SI), R8
+	ADDL R9, CX
+	ROLL $0x10, CX
+	ADDL DX, CX
+	XORL BX, R9
+	XORL CX, R9
+	ADDL $0xfde5380c, BX
+	ADDL R8, BX
+	MOVL 4(SI), R8
+	ADDL R9, BX
+	ROLL $0x17, BX
+	ADDL CX, BX
+	XORL AX, R9
+	XORL BX, R9
+	ADDL $0xa4beea44, AX
+	ADDL R8, AX
+	MOVL 16(SI), R8
+	ADDL R9, AX
+	ROLL $0x04, AX
+	ADDL BX, AX
+	XORL DX, R9
+	XORL AX, R9
+	ADDL $0x4bdecfa9, DX
+	ADDL R8, DX
+	MOVL 28(SI), R8
+	ADDL R9, DX
+	ROLL $0x0b, DX
+	ADDL AX, DX
+	XORL CX, R9
+	XORL DX, R9
+	ADDL $0xf6bb4b60, CX
+	ADDL R8, CX
+	MOVL 40(SI), R8
+	ADDL R9, CX
+	ROLL $0x10, CX
+	ADDL DX, CX
+	XORL BX, R9
+	XORL CX, R9
+	ADDL $0xbebfbc70, BX
+	ADDL R8, BX
+	MOVL 52(SI), R8
+	ADDL R9, BX
+	ROLL $0x17, BX
+	ADDL CX, BX
+	XORL AX, R9
+	XORL BX, R9
+	ADDL $0x289b7ec6, AX
+	ADDL R8, AX
+	MOVL (SI), R8
+	ADDL R9, AX
+	ROLL $0x04, AX
+	ADDL BX, AX
+	XORL DX, R9
+	XORL AX, R9
+	ADDL $0xeaa127fa, DX
+	ADDL R8, DX
+	MOVL 12(SI), R8
+	ADDL R9, DX
+	ROLL $0x0b, DX
+	ADDL AX, DX
+	XORL CX, R9
+	XORL DX, R9
+	ADDL $0xd4ef3085, CX
+	ADDL R8, CX
+	MOVL 24(SI), R8
+	ADDL R9, CX
+	ROLL $0x10, CX
+	ADDL DX, CX
+	XORL BX, R9
+	XORL CX, R9
+	ADDL $0x04881d05, BX
+	ADDL R8, BX
+	MOVL 36(SI), R8
+	ADDL R9, BX
+	ROLL $0x17, BX
+	ADDL CX, BX
+	XORL AX, R9
+	XORL BX, R9
+	ADDL $0xd9d4d039, AX
+	ADDL R8, AX
+	MOVL 48(SI), R8
+	ADDL R9, AX
+	ROLL $0x04, AX
+	ADDL BX, AX
+	XORL DX, R9
+	XORL AX, R9
+	ADDL $0xe6db99e5, DX
+	ADDL R8, DX
+	MOVL 60(SI), R8
+	ADDL R9, DX
+	ROLL $0x0b, DX
+	ADDL AX, DX
+	XORL CX, R9
+	XORL DX, R9
+	ADDL $0x1fa27cf8, CX
+	ADDL R8, CX
+	MOVL 8(SI), R8
+	ADDL R9, CX
+	ROLL $0x10, CX
+	ADDL DX, CX
+	XORL BX, R9
+	XORL CX, R9
+	ADDL $0xc4ac5665, BX
+	ADDL R8, BX
+	MOVL (SI), R8
+	ADDL R9, BX
+	ROLL $0x17, BX
+	ADDL CX, BX
+	MOVL R11, R9
+	XORL DX, R9
+	ADDL $0xf4292244, AX
+	ADDL R8, AX
+	ORL  BX, R9
+	XORL CX, R9
+	ADDL R9, AX
+	MOVL 28(SI), R8
+	MOVL $0xffffffff, R9
+	ROLL $0x06, AX
+	XORL CX, R9
+	ADDL BX, AX
+	ADDL $0x432aff97, DX
+	ADDL R8, DX
+	ORL  AX, R9
+	XORL BX, R9
+	ADDL R9, DX
+	MOVL 56(SI), R8
+	MOVL $0xffffffff, R9
+	ROLL $0x0a, DX
+	XORL BX, R9
+	ADDL AX, DX
+	ADDL $0xab9423a7, CX
+	ADDL R8, CX
+	ORL  DX, R9
+	XORL AX, R9
+	ADDL R9, CX
+	MOVL 20(SI), R8
+	MOVL $0xffffffff, R9
+	ROLL $0x0f, CX
+	XORL AX, R9
+	ADDL DX, CX
+	ADDL $0xfc93a039, BX
+	ADDL R8, BX
+	ORL  CX, R9
+	XORL DX, R9
+	ADDL R9, BX
+	MOVL 48(SI), R8
+	MOVL $0xffffffff, R9
+	ROLL $0x15, BX
+	XORL DX, R9
+	ADDL CX, BX
+	ADDL $0x655b59c3, AX
+	ADDL R8, AX
+	ORL  BX, R9
+	XORL CX, R9
+	ADDL R9, AX
+	MOVL 12(SI), R8
+	MOVL $0xffffffff, R9
+	ROLL $0x06, AX
+	XORL CX, R9
+	ADDL BX, AX
+	ADDL $0x8f0ccc92, DX
+	ADDL R8, DX
+	ORL  AX, R9
+	XORL BX, R9
+	ADDL R9, DX
+	MOVL 40(SI), R8
+	MOVL $0xffffffff, R9
+	ROLL $0x0a, DX
+	XORL BX, R9
+	ADDL AX, DX
+	ADDL $0xffeff47d, CX
+	ADDL R8, CX
+	ORL  DX, R9
+	XORL AX, R9
+	ADDL R9, CX
+	MOVL 4(SI), R8
+	MOVL $0xffffffff, R9
+	ROLL $0x0f, CX
+	XORL AX, R9
+	ADDL DX, CX
+	ADDL $0x85845dd1, BX
+	ADDL R8, BX
+	ORL  CX, R9
+	XORL DX, R9
+	ADDL R9, BX
+	MOVL 32(SI), R8
+	MOVL $0xffffffff, R9
+	ROLL $0x15, BX
+	XORL DX, R9
+	ADDL CX, BX
+	ADDL $0x6fa87e4f, AX
+	ADDL R8, AX
+	ORL  BX, R9
+	XORL CX, R9
+	ADDL R9, AX
+	MOVL 60(SI), R8
+	MOVL $0xffffffff, R9
+	ROLL $0x06, AX
+	XORL CX, R9
+	ADDL BX, AX
+	ADDL $0xfe2ce6e0, DX
+	ADDL R8, DX
+	ORL  AX, R9
+	XORL BX, R9
+	ADDL R9, DX
+	MOVL 24(SI), R8
+	MOVL $0xffffffff, R9
+	ROLL $0x0a, DX
+	XORL BX, R9
+	ADDL AX, DX
+	ADDL $0xa3014314, CX
+	ADDL R8, CX
+	ORL  DX, R9
+	XORL AX, R9
+	ADDL R9, CX
+	MOVL 52(SI), R8
+	MOVL $0xffffffff, R9
+	ROLL $0x0f, CX
+	XORL AX, R9
+	ADDL DX, CX
+	ADDL $0x4e0811a1, BX
+	ADDL R8, BX
+	ORL  CX, R9
+	XORL DX, R9
+	ADDL R9, BX
+	MOVL 16(SI), R8
+	MOVL $0xffffffff, R9
+	ROLL $0x15, BX
+	XORL DX, R9
+	ADDL CX, BX
+	ADDL $0xf7537e82, AX
+	ADDL R8, AX
+	ORL  BX, R9
+	XORL CX, R9
+	ADDL R9, AX
+	MOVL 44(SI), R8
+	MOVL $0xffffffff, R9
+	ROLL $0x06, AX
+	XORL CX, R9
+	ADDL BX, AX
+	ADDL $0xbd3af235, DX
+	ADDL R8, DX
+	ORL  AX, R9
+	XORL BX, R9
+	ADDL R9, DX
+	MOVL 8(SI), R8
+	MOVL $0xffffffff, R9
+	ROLL $0x0a, DX
+	XORL BX, R9
+	ADDL AX, DX
+	ADDL $0x2ad7d2bb, CX
+	ADDL R8, CX
+	ORL  DX, R9
+	XORL AX, R9
+	ADDL R9, CX
+	MOVL 36(SI), R8
+	MOVL $0xffffffff, R9
+	ROLL $0x0f, CX
+	XORL AX, R9
+	ADDL DX, CX
+	ADDL $0xeb86d391, BX
+	ADDL R8, BX
+	ORL  CX, R9
+	XORL DX, R9
+	ADDL R9, BX
+	MOVL (SI), R8
+	MOVL $0xffffffff, R9
+	ROLL $0x15, BX
+	XORL DX, R9
+	ADDL CX, BX
+	ADDL R12, AX
+	ADDL R13, BX
+	ADDL R14, CX
+	ADDL R15, DX
+	ADDQ $0x40, SI
+	CMPQ SI, DI
+	JB   loop
 
 end:
-	MOVL	AX,		(0*4)(BP)
-	MOVL	BX,		(1*4)(BP)
-	MOVL	CX,		(2*4)(BP)
-	MOVL	DX,		(3*4)(BP)
+	MOVL AX, (BP)
+	MOVL BX, 4(BP)
+	MOVL CX, 8(BP)
+	MOVL DX, 12(BP)
 	RET
diff --git a/src/go/types/stdlib_test.go b/src/go/types/stdlib_test.go
index b82fd0ddc1a..d41a3d10df6 100644
--- a/src/go/types/stdlib_test.go
+++ b/src/go/types/stdlib_test.go
@@ -359,6 +359,7 @@ var excluded = map[string]bool{
 	// See go.dev/issue/46027: some imports are missing for this submodule.
 	"crypto/internal/bigmod/_asm":             true,
 	"crypto/internal/edwards25519/field/_asm": true,
+	"crypto/md5/_asm":                         true,
 	"crypto/sha1/_asm":                        true,
 	"crypto/sha256/_asm":                      true,
 	"crypto/sha512/_asm":                      true,