diff --git a/src/cmd/asm/internal/asm/testdata/mips64.s b/src/cmd/asm/internal/asm/testdata/mips64.s
index 2d1bc18cec..e80f4d83d1 100644
--- a/src/cmd/asm/internal/asm/testdata/mips64.s
+++ b/src/cmd/asm/internal/asm/testdata/mips64.s
@@ -130,27 +130,27 @@ TEXT foo(SB),DUPOK|NOSPLIT,$0
 //	{
 //		outcode(int($1), &$2, 0, &$4);
 //	}
-	MOVW	FCR0, R1
+	MOVW	FCR31, R1 // 4441f800
 
 //	LMOVW freg ','  fpscr
 //	{
 //		outcode(int($1), &$2, 0, &$4);
 //	}
-	MOVW	R1, FCR0
+	MOVW	R1, FCR31 // 44c1f800
 
 //	LMOVW rreg ',' mreg
 //	{
 //		outcode(int($1), &$2, 0, &$4);
 //	}
-	MOVW	R1, M1
-	MOVV	R1, M1
+	MOVW	R1, M1 // 40810800
+	MOVV	R1, M1 // 40a10800
 
 //	LMOVW mreg ',' rreg
 //	{
 //		outcode(int($1), &$2, 0, &$4);
 //	}
-	MOVW	M1, R1
-	MOVV	M1, R1
+	MOVW	M1, R1 // 40010800
+	MOVV	M1, R1 // 40210800
 
 
 //
@@ -406,6 +406,7 @@ label4:
 
 	NEGW	R1, R2 // 00011023
 	NEGV	R1, R2 // 0001102f
+	RET
 
 // END
 //
diff --git a/src/cmd/internal/obj/mips/a.out.go b/src/cmd/internal/obj/mips/a.out.go
index 026e8db76a..b0205ec11a 100644
--- a/src/cmd/internal/obj/mips/a.out.go
+++ b/src/cmd/internal/obj/mips/a.out.go
@@ -46,7 +46,7 @@ const (
 )
 
 const (
-	REG_R0 = obj.RBaseMIPS + iota
+	REG_R0 = obj.RBaseMIPS + iota // must be a multiple of 32
 	REG_R1
 	REG_R2
 	REG_R3
@@ -79,7 +79,7 @@ const (
 	REG_R30
 	REG_R31
 
-	REG_F0
+	REG_F0 // must be a multiple of 32
 	REG_F1
 	REG_F2
 	REG_F3
@@ -112,11 +112,8 @@ const (
 	REG_F30
 	REG_F31
 
-	REG_HI
-	REG_LO
-
 	// co-processor 0 control registers
-	REG_M0
+	REG_M0 // must be a multiple of 32
 	REG_M1
 	REG_M2
 	REG_M3
@@ -150,7 +147,7 @@ const (
 	REG_M31
 
 	// FPU control registers
-	REG_FCR0
+	REG_FCR0 // must be a multiple of 32
 	REG_FCR1
 	REG_FCR2
 	REG_FCR3
@@ -183,7 +180,10 @@ const (
 	REG_FCR30
 	REG_FCR31
 
-	REG_LAST = REG_FCR31 // the last defined register
+	REG_HI
+	REG_LO
+
+	REG_LAST = REG_LO // the last defined register
 
 	REG_SPECIAL = REG_M0
 
@@ -412,3 +412,22 @@ const (
 	AJAL = obj.ACALL
 	ARET = obj.ARET
 )
+
+func init() {
+	// The asm encoder generally assumes that the lowest 5 bits of the
+	// REG_XX constants match the machine instruction encoding, i.e.
+	// the lowest 5 bits is the register number.
+	// Check this here.
+	if REG_R0%32 != 0 {
+		panic("REG_R0 is not a multiple of 32")
+	}
+	if REG_F0%32 != 0 {
+		panic("REG_F0 is not a multiple of 32")
+	}
+	if REG_M0%32 != 0 {
+		panic("REG_M0 is not a multiple of 32")
+	}
+	if REG_FCR0%32 != 0 {
+		panic("REG_FCR0 is not a multiple of 32")
+	}
+}
diff --git a/src/cmd/internal/obj/mips/asm0.go b/src/cmd/internal/obj/mips/asm0.go
index 77aa24a4f6..76a3eec6bf 100644
--- a/src/cmd/internal/obj/mips/asm0.go
+++ b/src/cmd/internal/obj/mips/asm0.go
@@ -362,8 +362,8 @@ var optab = []Optab{
 
 	{AWORD, C_LCON, C_NONE, C_NONE, 40, 4, 0, 0},
 
-	{AMOVW, C_REG, C_NONE, C_FCREG, 41, 8, 0, 0},
-	{AMOVV, C_REG, C_NONE, C_FCREG, 41, 8, 0, sys.MIPS64},
+	{AMOVW, C_REG, C_NONE, C_FCREG, 41, 4, 0, 0},
+	{AMOVV, C_REG, C_NONE, C_FCREG, 41, 4, 0, sys.MIPS64},
 	{AMOVW, C_FCREG, C_NONE, C_REG, 42, 4, 0, 0},
 	{AMOVV, C_FCREG, C_NONE, C_REG, 42, 4, 0, sys.MIPS64},
 
@@ -1476,8 +1476,7 @@ func (c *ctxt0) asmout(p *obj.Prog, o *Optab, out []uint32) {
 		o1 = uint32(c.regoff(&p.From))
 
 	case 41: /* movw f,fcr */
-		o1 = OP_RRR(SP(2, 1)|(2<<21), uint32(REGZERO), uint32(0), uint32(p.To.Reg))    /* mfcc1 */
-		o2 = OP_RRR(SP(2, 1)|(6<<21), uint32(p.From.Reg), uint32(0), uint32(p.To.Reg)) /* mtcc1 */
+		o1 = OP_RRR(SP(2, 1)|(6<<21), uint32(p.From.Reg), uint32(0), uint32(p.To.Reg)) /* mtcc1 */
 
 	case 42: /* movw fcr,r */
 		o1 = OP_RRR(SP(2, 1)|(2<<21), uint32(p.To.Reg), uint32(0), uint32(p.From.Reg)) /* mfcc1 */