summaryrefslogtreecommitdiffstats
path: root/lib/rbcodec/codecs/libopus/celt
diff options
context:
space:
mode:
authorNils Wallménius <nils@rockbox.org>2013-09-01 17:37:58 +0200
committerNils Wallménius <nils@rockbox.org>2013-09-01 17:39:15 +0200
commita35c1b35951ce11e4c6ff658794c3142ba5dd218 (patch)
tree305f0e87019018b29e4bed9e2dfd6ee4a3bdbdf9 /lib/rbcodec/codecs/libopus/celt
parent516f7fbd6c60eb153a0ba9d82e9b911e2f9732ab (diff)
downloadrockbox-a35c1b35951ce11e4c6ff658794c3142ba5dd218.tar.gz
rockbox-a35c1b35951ce11e4c6ff658794c3142ba5dd218.zip
Opus: Coldfire inline asm for comb_filter_const
Speeds up decoding a 64kbps test file by 2.6MHz Change-Id: Ibeb30f37cc00a4a6f65b64851009753f40e06fc1
Diffstat (limited to 'lib/rbcodec/codecs/libopus/celt')
-rw-r--r--lib/rbcodec/codecs/libopus/celt/cf/fixed_cf.h33
1 files changed, 33 insertions, 0 deletions
diff --git a/lib/rbcodec/codecs/libopus/celt/cf/fixed_cf.h b/lib/rbcodec/codecs/libopus/celt/cf/fixed_cf.h
index c442a55663..a2b5d5acef 100644
--- a/lib/rbcodec/codecs/libopus/celt/cf/fixed_cf.h
+++ b/lib/rbcodec/codecs/libopus/celt/cf/fixed_cf.h
@@ -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 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