libmad: fix mips using gcc attribute mode(DI)

Use gcc __attribute__ ((mode (DI))) for a 64 bit multiplication instead
of (broken) assembler code.

srcpkgs/libmad/patches/mips-mode_di.patch

@@ -0,0 +1,27 @@
+--- fixed.h	2004-02-17 03:02:03.000000000 +0100
++++ fixed.h	2017-12-03 21:02:42.248609296 +0100
+@@ -299,6 +299,16 @@
+ 
+ # elif defined(FPM_MIPS)
+ 
++#if 1
++
++typedef unsigned int u64_di_t __attribute__ ((mode (DI)));
++#define MAD_F_MLX(hi, lo, x, y) \
++  do { \
++    u64_di_t __ll = (u64_di_t) (x) * (y); \
++    hi = __ll >> 32; \
++    lo = __ll; \
++  } while (0)
++#else
+ /*
+  * This MIPS version is fast and accurate; the disposition of the least
+  * significant bit depends on OPT_ACCURACY via mad_f_scale64().
+@@ -328,6 +338,7 @@
+ 	 : "%r" ((x) >> 12), "r" ((y) >> 16))
+ #  define MAD_F_MLZ(hi, lo)  ((mad_fixed_t) (lo))
+ # endif
++#endif 
+ 
+ # if defined(OPT_SPEED)
+ #  define mad_f_scale64(hi, lo)  \