Shorten the coldfire fixmul32() by 2 instructions and use one register less. Smaller code, but practically no speed change.

git-svn-id: svn://svn.rockbox.org/rockbox/trunk@13783 a1c6a512-1295-4272-9138-f99709370657
This commit is contained in:
Jens Arnold 2007-07-04 00:31:17 +00:00
parent c7c21ae03d
commit 5717a0af87

View file

@ -97,27 +97,27 @@ static fixed64 Fixed32To64(fixed32 x)
__result; \ __result; \
}) })
#elif defined(CPU_COLDFIRE) #elif defined(CPU_COLDFIRE)
static inline int32_t fixmul32(int32_t x, int32_t y) static inline int32_t fixmul32(int32_t x, int32_t y)
{ {
int32_t t1, t2; #if PRECISION != 16
asm volatile ( #warning Coldfire fixmul32() only works for PRECISION == 16
"mac.l %[x],%[y],%%acc0\n" /* multiply */ #endif
"mulu.l %[y],%[x] \n" /* get lower half, avoid emac stall */ int32_t t1;
"movclr.l %%acc0,%[t1] \n" /* get higher half */ asm (
"moveq.l #15,%[t2] \n" "mac.l %[x], %[y], %%acc0 \n" /* multiply */
"asl.l %[t2],%[t1] \n" /* hi <<= 15, plus one free */ "mulu.l %[y], %[x] \n" /* get lower half, avoid emac stall */
"moveq.l #16,%[t2] \n" "movclr.l %%acc0, %[t1] \n" /* get higher half */
"lsr.l %[t2],%[x] \n" /* (unsigned)lo >>= 16 */ "lsr.l #1, %[t1] \n"
"or.l %[x],%[t1] \n" /* combine result */ "move.w %[t1], %[x] \n"
: /* outputs */ "swap %[x] \n"
[t1]"=&d"(t1), : /* outputs */
[t2]"=&d"(t2), [t1]"=&d"(t1),
[x] "+d" (x) [x] "+d" (x)
: /* inputs */ : /* inputs */
[y] "d" (y) [y] "d" (y)
); );
return t1; return x;
} }
#else #else
fixed32 fixmul32(fixed32 x, fixed32 y) fixed32 fixmul32(fixed32 x, fixed32 y)
{ {