len0rd/rockbox
1
0
Fork 1
mirror of https://github.com/Rockbox/rockbox.git synced 2025-10-14 02:27:39 -04:00
Code Issues Projects Releases Packages Wiki Activity Actions

Opus: Coldfire inline asm for comb_filter_const

Browse source 
Speeds up decoding a 64kbps test file by 2.6MHz

Change-Id: Ibeb30f37cc00a4a6f65b64851009753f40e06fc1
This commit is contained in:
Nils Wallménius 2013-09-01 17:37:58 +02:00
parent 516f7fbd6c
commit a35c1b3595
1 changed files with 33 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

33
lib/rbcodec/codecs/libopus/celt/cf/fixed_cf.h
View file

@ -53,4 +53,37 @@ static inline int32_t MULT32_32_Q31_cf(int32_t a, int32_t b)
} }
#define MULT32_32_Q31(a, b) (MULT32_32_Q31_cf(a, b)) #define MULT32_32_Q31(a, b) (MULT32_32_Q31_cf(a, b))
#define OVERRIDE_COMB_FILTER_CONST
static void comb_filter_const(opus_val32 *y, opus_val32 *x, int T, int N,
opus_val16 g10, opus_val16 g11, opus_val16 g12)
{
opus_val32 x0, x1, x2, x3, x4;
int i;
x4 = x[-T-2];
x3 = x[-T-1];
x2 = x[-T];
x1 = x[-T+1];
for (i=0;i<N;i++)
{
x0=x[i-T+2];
asm volatile("mac.l %[g10], %[x2], %%acc0;"
/* just doing straight MACs here is faster than pre-adding */
"mac.l %[g11], %[x1], %%acc0;"
"mac.l %[g11], %[x3], %%acc0;"
"mac.l %[g12], %[x0], %%acc0;"
"mac.l %[g12], %[x4], %%acc0;"
"move.l %[x3], %[x4];"
"move.l %[x2], %[x3];"
"move.l %[x1], %[x2];"
"move.l %[x0], %[x1];"
"movclr.l %%acc0, %[x0];"
: [x0] "+r" (x0), [x1] "+r" (x1), [x2] "+r" (x2),
[x3] "+r" (x3), [x4] "+r" (x4)
: [g10] "r" (g10 << 16), [g11] "r" (g11 << 16),
[g12] "r" (g12 << 16)
: "cc");
y[i] = x[i] + x0;
}
}
#endif #endif