merge

sys/src/cmd/5c/peep.c

@@ -306,6 +306,7 @@ subprop(Reg *r0)
 		case AAND:
 		case AEOR:
 		case AMUL:
+		case AMULU:
 		case ADIV:
 		case ADIVU:
 
@@ -977,6 +978,7 @@ copyu(Prog *p, Adr *v, Adr *s)
 	case AAND:
 	case AEOR:
 	case AMUL:
+	case AMULU:
 	case ADIV:
 	case ADIVU:
 	case AADDF:
@@ -1112,6 +1114,7 @@ a2type(Prog *p)
 	case AAND:
 	case AEOR:
 	case AMUL:
+	case AMULU:
 	case ADIV:
 	case ADIVU:
 		return D_REG;

sys/src/libmp/arm/mkfile

@@ -2,12 +2,14 @@ objtype=arm
 </$objtype/mkfile
 
 LIB=/$objtype/lib/libmp.a
-OFILES=	\
+SFILES=mpvecdigmuladd.s mpvecdigmulsub.s
 
 HFILES=/$objtype/include/u.h /sys/include/mp.h ../port/dat.h
 
-UPDATE=\
-	mkfile\
+OFILES=${SFILES:%.s=%.$O}
+
+UPDATE=mkfile\
 	$HFILES\
+	$SFILES\
 
 </sys/src/cmd/mksyslib

sys/src/libmp/arm/mpvecdigmuladd.s

@@ -0,0 +1,19 @@
+TEXT mpvecdigmuladd(SB),$0
+	MOVW	n+4(FP),R4
+	MOVW	m+8(FP),R5
+	MOVW	p+12(FP),R6
+	MOVW	$0, R2
+_muladdloop:
+	MOVW	$0, R1
+	MOVW.W.P 4(R0), R3
+	MULALU	R3, R5, (R1, R2)
+ 	MOVW	(R6), R7
+	ADD.S	R2, R7
+	ADC	$0, R1, R2
+	MOVW.W.P R7, 4(R6)
+	SUB.S	$1, R4
+	B.NE	_muladdloop
+	MOVW	(R6), R7
+	ADD	R2, R7
+	MOVW	R7, (R6)
+	RET

sys/src/libmp/arm/mpvecdigmulsub.s

@@ -0,0 +1,22 @@
+TEXT mpvecdigmulsub(SB),$0
+	MOVW	n+4(FP),R4
+	MOVW	m+8(FP),R5
+	MOVW	p+12(FP),R6
+	MOVW	$0, R2
+_mulsubloop:
+	MOVW	$0, R1
+	MOVW.W.P 4(R0), R3
+	MULALU	R3, R5, (R1, R2)
+ 	MOVW	(R6), R7
+	SUB.S	R2, R7
+	ADD.CC	$1, R1
+	MOVW	R1, R2
+	MOVW.W.P R7, 4(R6)
+	SUB.S	$1, R4
+	B.NE	_mulsubloop
+	MOVW	(R6), R7
+	SUB.S	R2, R7
+	MOVW.CS	$1, R0
+	MOVW.CC	$-1, R0
+	MOVW	R7, (R6)
+	RET

sys/src/libmp/port/mpdigdiv.c

@@ -21,6 +21,19 @@ mpdigdiv(mpdigit *dividend, mpdigit divisor, mpdigit *quotient)
 		return;
 	}
 
+	// very common case
+	if(~divisor == 0){
+		lo += hi;
+		if(lo < hi){
+			hi++;
+			lo++;
+		}
+		if(lo+1 == 0)
+			hi++;
+		*quotient = hi;
+		return;
+	}
+
 	// at this point we know that hi < divisor
 	// just shift and subtract till we're done
 	q = 0;