diff --git a/src/cmd/asm/internal/asm/asm.go b/src/cmd/asm/internal/asm/asm.go
index bd9ba91b22..5cbf2232d1 100644
--- a/src/cmd/asm/internal/asm/asm.go
+++ b/src/cmd/asm/internal/asm/asm.go
@@ -460,17 +460,6 @@ func (p *Parser) asmInstruction(op int, cond string, a []obj.Addr) {
 		prog.From = a[0]
 		prog.To = a[1]
 		switch p.arch.Thechar {
-		case '6', '8':
-			// DX:AX as a register pair can only appear on the RHS.
-			// Bizarrely, to obj it's specified by setting index on the LHS.
-			// TODO: can we fix this?
-			if a[1].Reg2 != 0 {
-				if a[0].Reg2 != 0 {
-					p.errorf("register pair must be on LHS")
-				}
-				prog.From.Index = int16(a[1].Reg2)
-				prog.To.Reg2 = 0
-			}
 		case '9':
 			var reg0, reg1 int16
 			// Handle (R1+R2)
@@ -488,7 +477,7 @@ func (p *Parser) asmInstruction(op int, cond string, a []obj.Addr) {
 	case 3:
 		switch p.arch.Thechar {
 		case '5':
-			// Strange special case.
+			// Special cases.
 			if arch.IsARMSTREX(op) {
 				/*
 					STREX x, (y), z
@@ -504,20 +493,9 @@ func (p *Parser) asmInstruction(op int, cond string, a []obj.Addr) {
 			prog.Reg = p.getRegister(prog, op, &a[1])
 			prog.To = a[2]
 		case '6', '8':
-			// CMPSD etc.; third operand is imm8, stored in offset, or a register.
 			prog.From = a[0]
-			prog.To = a[1]
-			switch a[2].Type {
-			case obj.TYPE_MEM:
-				prog.To.Offset = p.getConstant(prog, op, &a[2])
-			case obj.TYPE_REG:
-				// Strange reordering.
-				prog.To = a[2]
-				prog.From = a[1]
-				prog.To.Offset = p.getImmediate(prog, op, &a[0])
-			default:
-				p.errorf("expected offset or register for 3rd operand")
-			}
+			prog.From3 = a[1]
+			prog.To = a[2]
 		case '9':
 			if arch.IsPPC64CMP(op) {
 				// CMPW etc.; third argument is a CR register that goes into prog.Reg.
diff --git a/src/cmd/asm/internal/asm/endtoend_test.go b/src/cmd/asm/internal/asm/endtoend_test.go
index 011a86cc36..7b4bdfccd0 100644
--- a/src/cmd/asm/internal/asm/endtoend_test.go
+++ b/src/cmd/asm/internal/asm/endtoend_test.go
@@ -79,11 +79,9 @@ func TestARMEndToEnd(t *testing.T) {
 }
 
 func TestAMD64EndToEnd(t *testing.T) {
-	t.Skip("broken")
 	testEndToEnd(t, "amd64")
 }
 
 func Test386EndToEnd(t *testing.T) {
-	t.Skip("broken")
 	testEndToEnd(t, "386")
 }
diff --git a/src/cmd/asm/internal/asm/operand_test.go b/src/cmd/asm/internal/asm/operand_test.go
index 5190ed45eb..179d63db26 100644
--- a/src/cmd/asm/internal/asm/operand_test.go
+++ b/src/cmd/asm/internal/asm/operand_test.go
@@ -41,47 +41,14 @@ func testOperandParser(t *testing.T, parser *Parser, tests []operandTest) {
 	}
 }
 
-func testX86RegisterPair(t *testing.T, parser *Parser) {
-	// Special case for AX:DX, which is really two operands so isn't printed correcctly
-	// by Aconv, but is OK by the -S output.
-	parser.start(lex.Tokenize("AX:BX)"))
-	addr := obj.Addr{}
-	parser.operand(&addr)
-	want := obj.Addr{
-		Type: obj.TYPE_REG,
-		Reg:  parser.arch.Register["AX"],
-		Reg2: parser.arch.Register["BX"], // TODO: clean up how this is encoded in parse.go
-	}
-	if want != addr {
-		t.Errorf("AX:DX: expected %+v got %+v", want, addr)
-	}
-	// Special case for foo(SB):DX, which is really two operands so isn't printed correctly
-	// by Aconv, but is OK by the -S output.
-	parser.start(lex.Tokenize("foo+4(SB):AX"))
-	addr = obj.Addr{}
-	parser.operand(&addr)
-	want = obj.Addr{
-		Type:   obj.TYPE_MEM,
-		Name:   obj.NAME_EXTERN,
-		Offset: 4,
-		Sym:    obj.Linklookup(parser.ctxt, "foo", 0),
-		Reg2:   parser.arch.Register["AX"], // TODO: clean up how this is encoded in parse.go
-	}
-	if want != addr {
-		t.Errorf("foo+4(SB):AX: expected %+v got %+v", want, addr)
-	}
-}
-
 func TestAMD64OperandParser(t *testing.T) {
 	parser := newParser("amd64")
 	testOperandParser(t, parser, amd64OperandTests)
-	testX86RegisterPair(t, parser)
 }
 
 func Test386OperandParser(t *testing.T) {
 	parser := newParser("386")
 	testOperandParser(t, parser, x86OperandTests)
-	testX86RegisterPair(t, parser)
 }
 
 func TestARMOperandParser(t *testing.T) {
@@ -113,7 +80,6 @@ type operandTest struct {
 // Examples collected by scanning all the assembly in the standard repo.
 
 var amd64OperandTests = []operandTest{
-	// {"AX:DX", "AX:DX"}, Handled in TestAMD64OperandParser directly.
 	{"$(-1.0)", "$(-1.0)"},
 	{"$(0.0)", "$(0.0)"},
 	{"$(0x2000000+116)", "$33554548"},
diff --git a/src/cmd/asm/internal/asm/parse.go b/src/cmd/asm/internal/asm/parse.go
index 27d311293f..f37c5a0168 100644
--- a/src/cmd/asm/internal/asm/parse.go
+++ b/src/cmd/asm/internal/asm/parse.go
@@ -109,6 +109,7 @@ func (p *Parser) line() bool {
 	operands := make([][]lex.Token, 0, 3)
 	// Zero or more comma-separated operands, one per loop.
 	nesting := 0
+	colon := -1
 	for tok != '\n' && tok != ';' {
 		// Process one operand.
 		items := make([]lex.Token, 0, 3)
@@ -135,7 +136,16 @@ func (p *Parser) line() bool {
 				p.errorf("unexpected EOF")
 				return false
 			}
-			if tok == '\n' || tok == ';' || (nesting == 0 && tok == ',') {
+			// Split operands on comma. Also, the old syntax on x86 for a "register pair"
+			// was AX:DX, for which the new syntax is DX, AX. Note the reordering.
+			if tok == '\n' || tok == ';' || (nesting == 0 && (tok == ',' || tok == ':')) {
+				if tok == ':' {
+					// Remember this location so we can swap the operands below.
+					if colon >= 0 {
+						p.errorf("invalid ':' in operand")
+					}
+					colon = len(operands)
+				}
 				break
 			}
 			if tok == '(' || tok == '[' {
@@ -148,8 +158,13 @@ func (p *Parser) line() bool {
 		}
 		if len(items) > 0 {
 			operands = append(operands, items)
-		} else if len(operands) > 0 || tok == ',' {
-			// Had a comma with nothing after.
+			if colon >= 0 && len(operands) == colon+2 {
+				// AX:DX becomes DX, AX.
+				operands[colon], operands[colon+1] = operands[colon+1], operands[colon]
+				colon = -1
+			}
+		} else if len(operands) > 0 || tok == ',' || colon >= 0 {
+			// Had a separator with nothing after.
 			p.errorf("missing operand")
 		}
 	}
@@ -226,7 +241,7 @@ func (p *Parser) parseScale(s string) int8 {
 
 // operand parses a general operand and stores the result in *a.
 func (p *Parser) operand(a *obj.Addr) bool {
-	// fmt.Printf("Operand: %v\n", p.input)
+	//fmt.Printf("Operand: %v\n", p.input)
 	if len(p.input) == 0 {
 		p.errorf("empty operand: cannot happen")
 		return false
@@ -297,6 +312,8 @@ func (p *Parser) operand(a *obj.Addr) bool {
 			if r2 != 0 {
 				// Form is R1:R2. It is on RHS and the second register
 				// needs to go into the LHS. This is a horrible hack. TODO.
+				// TODO: If we never see this again, can delete Addr.Reg2.
+				panic("cannot happen")
 				a.Reg2 = r2
 			}
 		}
@@ -362,18 +379,8 @@ func (p *Parser) operand(a *obj.Addr) bool {
 		// fmt.Printf("offset %d \n", a.Offset)
 	}
 
-	// Odd x86 case: sym+4(SB):AX. Have name, colon, register.
-	if p.peek() == ':' && a.Name != obj.NAME_NONE && a.Reg2 == 0 && (p.arch.Thechar == '6' || p.arch.Thechar == '8') {
-		p.get(':')
-		r2, ok := p.registerReference(p.next().String())
-		if !ok {
-			return false
-		}
-		a.Reg2 = r2 // TODO: See comment about Reg3 above.
-	} else {
-		// Register indirection: (reg) or (index*scale). We are on the opening paren.
-		p.registerIndirect(a, prefix)
-	}
+	// Register indirection: (reg) or (index*scale). We are on the opening paren.
+	p.registerIndirect(a, prefix)
 	// fmt.Printf("DONE %s\n", p.arch.Dconv(&emptyProg, 0, a))
 
 	p.expect(scanner.EOF)
@@ -449,15 +456,10 @@ func (p *Parser) register(name string, prefix rune) (r1, r2 int16, scale int8, o
 	}
 	c := p.peek()
 	if c == ':' || c == ',' || c == '+' {
-		// 2nd register; syntax (R1:R2) etc. No two architectures agree.
+		// 2nd register; syntax (R1+R2) etc. No two architectures agree.
 		// Check the architectures match the syntax.
 		char := p.arch.Thechar
 		switch p.next().ScanToken {
-		case ':':
-			if char != '6' && char != '8' {
-				p.errorf("illegal register pair syntax")
-				return
-			}
 		case ',':
 			if char != '5' {
 				p.errorf("illegal register pair syntax")
diff --git a/src/cmd/asm/internal/asm/testdata/386.out b/src/cmd/asm/internal/asm/testdata/386.out
index 8ac94db82b..d88277d1ae 100644
--- a/src/cmd/asm/internal/asm/testdata/386.out
+++ b/src/cmd/asm/internal/asm/testdata/386.out
@@ -1,49 +1,49 @@
-5 00001 (testdata/386.s:5)	TEXT	foo(SB),$0
-8 00002 (testdata/386.s:8)	SETCC	,AX
-9 00003 (testdata/386.s:9)	SETCC	,foo+4(SB)
-12 00004 (testdata/386.s:12)	DIVB	AX,
-13 00005 (testdata/386.s:13)	DIVB	foo+4(SB),
-14 00006 (testdata/386.s:14)	PUSHL	$foo+4(SB),
-15 00007 (testdata/386.s:15)	POPL	,AX
-18 00008 (testdata/386.s:18)	SUBB	$1,AX
-19 00009 (testdata/386.s:19)	SUBB	$1,foo+4(SB)
-20 00010 (testdata/386.s:20)	SUBB	BX,AX
-21 00011 (testdata/386.s:21)	SUBB	BX,foo+4(SB)
-24 00012 (testdata/386.s:24)	CMPB	AX,$1
-25 00013 (testdata/386.s:25)	CMPB	foo+4(SB),$4
-26 00014 (testdata/386.s:26)	CMPB	BX,AX
-27 00015 (testdata/386.s:27)	CMPB	foo+4(SB),BX
-31 00016 (testdata/386.s:31)	JCS	,
-32 00017 (testdata/386.s:32)	JCS	,16(PC)
-35 00018 (testdata/386.s:35)	CALL	,AX
-36 00019 (testdata/386.s:36)	JMP	,AX
-37 00020 (testdata/386.s:37)	CALL	,*foo(SB)
-38 00021 (testdata/386.s:38)	JMP	,$4
-39 00022 (testdata/386.s:39)	JMP	,16
-40 00023 (testdata/386.s:40)	CALL	,foo(SB)
-42 00024 (testdata/386.s:42)	CALL	,foo+4(SB)(AX*4)
-43 00025 (testdata/386.s:43)	CALL	,4(SP)
-44 00026 (testdata/386.s:44)	CALL	,(AX)
-45 00027 (testdata/386.s:45)	CALL	,(SP)
-47 00028 (testdata/386.s:47)	CALL	,(AX)(AX*4)
-48 00029 (testdata/386.s:48)	CALL	,4(SP)
-49 00030 (testdata/386.s:49)	CALL	,(AX)
-50 00031 (testdata/386.s:50)	CALL	,(SP)
-52 00032 (testdata/386.s:52)	JMP	,(AX)(AX*4)
-55 00033 (testdata/386.s:55)	NOP	,
-56 00034 (testdata/386.s:56)	NOP	AX,
-57 00035 (testdata/386.s:57)	NOP	foo+4(SB),
-60 00036 (testdata/386.s:60)	SHLL	$4,BX
-61 00037 (testdata/386.s:61)	SHLL	$4,foo+4(SB)
-62 00038 (testdata/386.s:62)	SHLL	$4,foo+4(SB):AX
-65 00039 (testdata/386.s:65)	MOVL	AX,BX
-66 00040 (testdata/386.s:66)	MOVL	$4,BX
-69 00041 (testdata/386.s:69)	IMULL	AX,
-70 00042 (testdata/386.s:70)	IMULL	$4,CX
-71 00043 (testdata/386.s:71)	IMULL	AX,BX
-74 00044 (testdata/386.s:74)	CMPPD	X0,$4,X1
-75 00045 (testdata/386.s:75)	CMPPD	X0,foo+4(SB)
-78 00046 (testdata/386.s:78)	PINSRD	(AX),$1,X0
-79 00047 (testdata/386.s:79)	PINSRD	foo+4(FP),$2,X0
-83 00048 (testdata/386.s:83)	LOOP	,
-86 00049 (testdata/386.s:86)	RET	,
+5 00001 (testdata/386.s:5)	TEXT	foo(SB), $0
+8 00002 (testdata/386.s:8)	SETCC	AX
+9 00003 (testdata/386.s:9)	SETCC	foo+4(SB)
+12 00004 (testdata/386.s:12)	DIVB	AX
+13 00005 (testdata/386.s:13)	DIVB	foo+4(SB)
+14 00006 (testdata/386.s:14)	PUSHL	$foo+4(SB)
+15 00007 (testdata/386.s:15)	POPL	AX
+18 00008 (testdata/386.s:18)	SUBB	$1, AX
+19 00009 (testdata/386.s:19)	SUBB	$1, foo+4(SB)
+20 00010 (testdata/386.s:20)	SUBB	BX, AX
+21 00011 (testdata/386.s:21)	SUBB	BX, foo+4(SB)
+24 00012 (testdata/386.s:24)	CMPB	AX, $1
+25 00013 (testdata/386.s:25)	CMPB	foo+4(SB), $4
+26 00014 (testdata/386.s:26)	CMPB	BX, AX
+27 00015 (testdata/386.s:27)	CMPB	foo+4(SB), BX
+31 00016 (testdata/386.s:31)	JCS
+32 00017 (testdata/386.s:32)	JCS	16(PC)
+35 00018 (testdata/386.s:35)	CALL	AX
+36 00019 (testdata/386.s:36)	JMP	AX
+37 00020 (testdata/386.s:37)	CALL	*foo(SB)
+38 00021 (testdata/386.s:38)	JMP	$4
+39 00022 (testdata/386.s:39)	JMP	16
+40 00023 (testdata/386.s:40)	CALL	foo(SB)
+42 00024 (testdata/386.s:42)	CALL	foo+4(SB)(AX*4)
+43 00025 (testdata/386.s:43)	CALL	4(SP)
+44 00026 (testdata/386.s:44)	CALL	(AX)
+45 00027 (testdata/386.s:45)	CALL	(SP)
+47 00028 (testdata/386.s:47)	CALL	(AX)(AX*4)
+48 00029 (testdata/386.s:48)	CALL	4(SP)
+49 00030 (testdata/386.s:49)	CALL	(AX)
+50 00031 (testdata/386.s:50)	CALL	(SP)
+52 00032 (testdata/386.s:52)	JMP	(AX)(AX*4)
+55 00033 (testdata/386.s:55)	NOP
+56 00034 (testdata/386.s:56)	NOP	AX
+57 00035 (testdata/386.s:57)	NOP	foo+4(SB)
+60 00036 (testdata/386.s:60)	SHLL	$4, BX
+61 00037 (testdata/386.s:61)	SHLL	$4, foo+4(SB)
+62 00038 (testdata/386.s:62)	SHLL	$4, AX, foo+4(SB)
+65 00039 (testdata/386.s:65)	MOVL	AX, BX
+66 00040 (testdata/386.s:66)	MOVL	$4, BX
+69 00041 (testdata/386.s:69)	IMULL	AX
+70 00042 (testdata/386.s:70)	IMULL	$4, CX
+71 00043 (testdata/386.s:71)	IMULL	AX, BX
+74 00044 (testdata/386.s:74)	CMPPD	X0, X1, 4
+75 00045 (testdata/386.s:75)	CMPPD	X0, foo+4(SB), 4
+78 00046 (testdata/386.s:78)	PINSRD	$1, (AX), X0
+79 00047 (testdata/386.s:79)	PINSRD	$2, foo+4(FP), X0
+83 00048 (testdata/386.s:83)	LOOP
+86 00049 (testdata/386.s:86)	RET
diff --git a/src/cmd/asm/internal/asm/testdata/amd64.out b/src/cmd/asm/internal/asm/testdata/amd64.out
index e647a5c6bd..020a422976 100644
--- a/src/cmd/asm/internal/asm/testdata/amd64.out
+++ b/src/cmd/asm/internal/asm/testdata/amd64.out
@@ -1,57 +1,57 @@
-5 00001 (testdata/amd64.s:5)	TEXT	foo(SB),$0
-8 00002 (testdata/amd64.s:8)	NEGQ	,R11
-9 00003 (testdata/amd64.s:9)	NEGQ	,4(R11)
-10 00004 (testdata/amd64.s:10)	NEGQ	,foo+4(SB)
-13 00005 (testdata/amd64.s:13)	INT	$4,
-14 00006 (testdata/amd64.s:14)	DIVB	R11,
-15 00007 (testdata/amd64.s:15)	DIVB	4(R11),
-16 00008 (testdata/amd64.s:16)	DIVB	foo+4(SB),
-19 00009 (testdata/amd64.s:19)	SUBQ	$4,DI
-20 00010 (testdata/amd64.s:20)	SUBQ	R11,DI
-21 00011 (testdata/amd64.s:21)	SUBQ	4(R11),DI
-22 00012 (testdata/amd64.s:22)	SUBQ	foo+4(SB),DI
-23 00013 (testdata/amd64.s:23)	SUBQ	$4,8(R12)
-24 00014 (testdata/amd64.s:24)	SUBQ	R11,8(R12)
-25 00015 (testdata/amd64.s:25)	SUBQ	R11,foo+4(SB)
-28 00016 (testdata/amd64.s:28)	CMPB	CX,$4
-32 00017 (testdata/amd64.s:32)	JCS	,13(PC)
-33 00018 (testdata/amd64.s:33)	JCS	,17
-36 00019 (testdata/amd64.s:36)	JMP	,15(PC)
-37 00020 (testdata/amd64.s:37)	JMP	,17
-38 00021 (testdata/amd64.s:38)	JMP	,foo+4(SB)
-39 00022 (testdata/amd64.s:39)	JMP	,bar<>+4(SB)
-40 00023 (testdata/amd64.s:40)	JMP	,bar<>+4(SB)(R11*4)
-41 00024 (testdata/amd64.s:41)	JMP	,4(SP)
-42 00025 (testdata/amd64.s:42)	JMP	,(R12)
-44 00026 (testdata/amd64.s:44)	JMP	,(R12)(R13*4)
-45 00027 (testdata/amd64.s:45)	JMP	,(AX)
-46 00028 (testdata/amd64.s:46)	JMP	,(SP)
-48 00029 (testdata/amd64.s:48)	JMP	,(AX)(AX*4)
-49 00030 (testdata/amd64.s:49)	JMP	,4(SP)
-50 00031 (testdata/amd64.s:50)	JMP	,(R12)
-52 00032 (testdata/amd64.s:52)	JMP	,(R12)(R13*4)
-53 00033 (testdata/amd64.s:53)	JMP	,(AX)
-54 00034 (testdata/amd64.s:54)	JMP	,(SP)
-56 00035 (testdata/amd64.s:56)	JMP	,(AX)(AX*4)
-57 00036 (testdata/amd64.s:57)	JMP	,R13
-60 00037 (testdata/amd64.s:60)	NOP	,
-61 00038 (testdata/amd64.s:61)	NOP	AX,
-62 00039 (testdata/amd64.s:62)	NOP	foo+4(SB),
-65 00040 (testdata/amd64.s:65)	SHLL	R11,R12
-66 00041 (testdata/amd64.s:66)	SHLL	R11,foo+4(SB)
-67 00042 (testdata/amd64.s:67)	SHLL	R11,R11:AX
-70 00043 (testdata/amd64.s:70)	MOVL	AX,R11
-71 00044 (testdata/amd64.s:71)	MOVL	$4,R11
-72 00045 (testdata/amd64.s:72)	MOVL	AX,AX:CS
-75 00046 (testdata/amd64.s:75)	IMULB	$4,
-76 00047 (testdata/amd64.s:76)	IMULB	R11,
-77 00048 (testdata/amd64.s:77)	IMULB	$4,R11
-78 00049 (testdata/amd64.s:78)	IMULB	R11,R12
-79 00050 (testdata/amd64.s:79)	IMULB	R11,foo+4(SB)
-82 00051 (testdata/amd64.s:82)	CMPPD	R11,$4,R12
-83 00052 (testdata/amd64.s:83)	CMPPD	R11,foo+4(SB)
-86 00053 (testdata/amd64.s:86)	PINSRW	R11,$4,AX
-87 00054 (testdata/amd64.s:87)	PINSRW	foo+4(SB),$4,AX
-90 00055 (testdata/amd64.s:90)	RETFL	$4,
-94 00056 (testdata/amd64.s:94)	LOOP	,
-97 00057 (testdata/amd64.s:97)	RET	,
+5 00001 (testdata/amd64.s:5)	TEXT	foo(SB), $0
+8 00002 (testdata/amd64.s:8)	NEGQ	R11
+9 00003 (testdata/amd64.s:9)	NEGQ	4(R11)
+10 00004 (testdata/amd64.s:10)	NEGQ	foo+4(SB)
+13 00005 (testdata/amd64.s:13)	INT	$4
+14 00006 (testdata/amd64.s:14)	DIVB	R11
+15 00007 (testdata/amd64.s:15)	DIVB	4(R11)
+16 00008 (testdata/amd64.s:16)	DIVB	foo+4(SB)
+19 00009 (testdata/amd64.s:19)	SUBQ	$4, DI
+20 00010 (testdata/amd64.s:20)	SUBQ	R11, DI
+21 00011 (testdata/amd64.s:21)	SUBQ	4(R11), DI
+22 00012 (testdata/amd64.s:22)	SUBQ	foo+4(SB), DI
+23 00013 (testdata/amd64.s:23)	SUBQ	$4, 8(R12)
+24 00014 (testdata/amd64.s:24)	SUBQ	R11, 8(R12)
+25 00015 (testdata/amd64.s:25)	SUBQ	R11, foo+4(SB)
+28 00016 (testdata/amd64.s:28)	CMPB	CX, $4
+32 00017 (testdata/amd64.s:32)	JCS	13(PC)
+33 00018 (testdata/amd64.s:33)	JCS	17
+36 00019 (testdata/amd64.s:36)	JMP	15(PC)
+37 00020 (testdata/amd64.s:37)	JMP	17
+38 00021 (testdata/amd64.s:38)	JMP	foo+4(SB)
+39 00022 (testdata/amd64.s:39)	JMP	bar<>+4(SB)
+40 00023 (testdata/amd64.s:40)	JMP	bar<>+4(SB)(R11*4)
+41 00024 (testdata/amd64.s:41)	JMP	4(SP)
+42 00025 (testdata/amd64.s:42)	JMP	(R12)
+44 00026 (testdata/amd64.s:44)	JMP	(R12)(R13*4)
+45 00027 (testdata/amd64.s:45)	JMP	(AX)
+46 00028 (testdata/amd64.s:46)	JMP	(SP)
+48 00029 (testdata/amd64.s:48)	JMP	(AX)(AX*4)
+49 00030 (testdata/amd64.s:49)	JMP	4(SP)
+50 00031 (testdata/amd64.s:50)	JMP	(R12)
+52 00032 (testdata/amd64.s:52)	JMP	(R12)(R13*4)
+53 00033 (testdata/amd64.s:53)	JMP	(AX)
+54 00034 (testdata/amd64.s:54)	JMP	(SP)
+56 00035 (testdata/amd64.s:56)	JMP	(AX)(AX*4)
+57 00036 (testdata/amd64.s:57)	JMP	R13
+60 00037 (testdata/amd64.s:60)	NOP
+61 00038 (testdata/amd64.s:61)	NOP	AX
+62 00039 (testdata/amd64.s:62)	NOP	foo+4(SB)
+65 00040 (testdata/amd64.s:65)	SHLL	R11, R12
+66 00041 (testdata/amd64.s:66)	SHLL	R11, foo+4(SB)
+67 00042 (testdata/amd64.s:67)	SHLL	R11, AX, R11
+70 00043 (testdata/amd64.s:70)	MOVL	AX, R11
+71 00044 (testdata/amd64.s:71)	MOVL	$4, R11
+72 00045 (testdata/amd64.s:72)	MOVL	AX, CS, AX
+75 00046 (testdata/amd64.s:75)	IMULB	$4
+76 00047 (testdata/amd64.s:76)	IMULB	R11
+77 00048 (testdata/amd64.s:77)	IMULB	$4, R11
+78 00049 (testdata/amd64.s:78)	IMULB	R11, R12
+79 00050 (testdata/amd64.s:79)	IMULB	R11, foo+4(SB)
+82 00051 (testdata/amd64.s:82)	CMPPD	R11, R12, 4
+83 00052 (testdata/amd64.s:83)	CMPPD	R11, foo+4(SB), 4
+86 00053 (testdata/amd64.s:86)	PINSRW	$4, R11, AX
+87 00054 (testdata/amd64.s:87)	PINSRW	$4, foo+4(SB), AX
+90 00055 (testdata/amd64.s:90)	RETFL	$4
+94 00056 (testdata/amd64.s:94)	LOOP
+97 00057 (testdata/amd64.s:97)	RET
diff --git a/src/cmd/asm/internal/asm/testdata/amd64.s b/src/cmd/asm/internal/asm/testdata/amd64.s
index eb13a1f96e..410057612b 100644
--- a/src/cmd/asm/internal/asm/testdata/amd64.s
+++ b/src/cmd/asm/internal/asm/testdata/amd64.s
@@ -64,7 +64,7 @@ label:
 // LTYPES spec5	{ outcode($1, &$2); }
 	SHLL	R11, R12
 	SHLL	R11, foo+4(SB)
-	SHLL	R11, R11:AX
+	SHLL	R11, R11:AX // Old syntax, still accepted.
 
 // LTYPEM spec6	{ outcode($1, &$2); }
 	MOVL	AX, R11