diff -u nbsd/Makefile ./Makefile
--- nbsd/Makefile	Wed Apr 24 14:48:20 2002
+++ ./Makefile	Wed Apr 24 14:49:36 2002
@@ -39,7 +39,8 @@
 	@echo "make aix-tk3play    IBM AIX"
 	@echo "make os2            IBM OS/2"
 	@echo "make netbsd         NetBSD"
-	@echo "make netbsd-arm32   NetBSD for strongarm"
+	@echo "make netbsd-arm32   NetBSD for strongarm (32bit fixedpoint)"
+	@echo "make netbsd-shark   NetBSD for strongarm (64bit fixedpoint)"
 	@echo "make netbsd-i386    NetBSD optimized for i386"
 	@echo "make netbsd-i386-elf NetBSD optimized for i386, ELF format"
 	@echo "make netbsd-m68k    NetBSD optimized for m68k"
@@ -392,6 +393,16 @@
 			-march=armv4 -mtune=strongarm -mcpu=strongarm \
 			-DREAL_IS_FIXED -DUSE_MMAP -DREAD_MMAP -DTERM_CONTROL \
 			-DNETBSD' \
+		mpg123-make
+
+netbsd-shark:
+	$(MAKE) CC=cc LDFLAGS= \
+		OBJECTS='decode.o dct64.o audio_sun.o term.o' \
+		CFLAGS='$(CFLAGS) -Wall -O6 -fomit-frame-pointer \
+			-funroll-all-loops -finline-functions -ffast-math \
+			-march=armv4 -mtune=strongarm110 -mcpu=strongarm110 \
+			-DREAL_IS_FIXED_ARMV4 -DUSE_MMAP -DREAD_MMAP \
+			-DTERM_CONTROL -DNETBSD' \
 		mpg123-make
 
 solaris-gcc-esd:
Only in ./: mpg123-64.patch
diff -u nbsd/mpg123.h ./mpg123.h
--- nbsd/mpg123.h	Wed Apr 24 14:49:13 2002
+++ ./mpg123.h	Wed Apr 24 14:49:41 2002
@@ -67,7 +67,87 @@
 
 # define DOUBLE_TO_REAL(x)	((int)((x) * REAL_FACTOR))
 # define REAL_TO_SHORT(x)	((x) >> REAL_RADIX)
+
 # define REAL_MUL(x, y)		(((long long)(x) * (long long)(y)) >> REAL_RADIX)
+
+#elif defined(REAL_IS_FIXED_ARMV4)
+# define REAL_IS_FIXED
+# define real long long
+
+# define REAL_RADIX		32
+# define REAL_FACTOR		(4294967296.0)
+
+# define REAL_PLUS_32767	( 32767LL << REAL_RADIX )
+# define REAL_MINUS_32768	( -32768LL << REAL_RADIX )
+
+# define DOUBLE_TO_REAL(x)	((long long)((x) * REAL_FACTOR))
+# define REAL_TO_SHORT(x)	((x) >> REAL_RADIX)
+
+/*
+ * "Reals" are signed, 15 bits before the point, 32 bits after.
+ * Multiplication is done like so:
+ * 
+ *   if |x| < 2^8, shift x >> 7, else shift x >>15
+ *   if |y| < 2^8, shift y >> 7, else shift y >>15
+ *   multiply x * y (just avoids clipping MSB)
+ *   shift the result >> 2, 10 or 18 bits, to make a total shift of 32 bits
+ *
+ * I'd do it in C, but gcc optimises it rather poorly (2K stack
+ * frames, anyone?).  Hence the bit of (ARMv4) asm magic below.
+ * (Unfortunately, the CLZ opcode, which would make for a much neater 
+ * solution to this, doesn't appear until ARMv5)
+ *
+ * (tjd April 2002) 
+ */
+
+# define REAL_MUL(x, y)         real_mul((x),(y))
+
+static __inline__ long long real_mul(long long x, long long y) {
+    const register long long _x asm("r0") = (x);
+    const register long long _y asm("r2") = (y);
+    register long long rv asm("r0");
+    __asm__ __volatile__ ("
+@ we'll need to shift the result at least >> 2 
+	mov	r4, #2
+@ both args >> 7, 
+	mov	r0, r0, lsr #7
+	mov	r2, r2, lsr #7
+	orr	r0, r0, r1, lsl #25
+	orr	r2, r2, r3, lsl #25
+	mov	r1, r1, asr #7
+	mov	r3, r3, asr #7
+@ check if x needs further shiftage
+	cmp	r1, #0
+	cmnne	r1, #1
+	addeq	r4, r4, #8
+@ shift x if needed
+	movne	r0, r0, lsr #8
+	orrne	r0, r0, r1, lsl #24
+	movne	r1, r1, asr #8
+@ check if y needs further shiftage
+	cmp	r3, #0
+	cmnne	r3, #1
+	addeq	r4, r4, #8
+@ shift y if needed
+	movne	r2, r2, lsr #8
+	orrne	r2, r2, r3, lsl #24
+	movne	r3, r3, asr #8
+@ multiply (ignoring overflow)
+	mul	r3, r0, r3
+	mla	r3, r1, r2, r3
+	umull	r0, r1, r2, r0
+	add	r1, r1, r3
+@ shift the result the appropriate amount
+	rsb	r3, r4, #32
+	mov	r0, r0, lsr r4
+	orr	r0, r0, r1, lsl r3
+	mov	r1, r1, asr r4"
+        : "=r" (rv)
+        : "0" (_x), "r" (_y)
+	: "r1", "r3", "r4", "cc");
+    return rv;
+}
+
 
 #else
 #  define real double