tdspeed: Tweak source and autocorrelation

C99-ify variables to keep them localized
Use sum of absolute differences instead of sum of squared differences for
autocorrelation which should reach the same conclusion and avoid 64-bit
multiplication.

Change-Id: If4f3715a36225420db3b05e9814b81766d04f0f3
This commit is contained in:
Michael Sevakis 2012-01-13 17:14:47 -05:00
parent 420ae56cec
commit e4a5762147

View file

@ -202,8 +202,8 @@ static int tdspeed_apply(int32_t *buf_out[2], int32_t *buf_in[2],
/* data_len in samples */
{
struct tdspeed_state_s *st = &tdspeed_state;
int32_t *curr, *prev, *dest[2], *d;
int32_t i, j, next_frame, prev_frame, shift, src_frame_sz;
int32_t *dest[2];
int32_t next_frame, prev_frame, src_frame_sz;
bool stereo = buf_in[0] != buf_in[1];
assert(stereo == st->stereo);
@ -259,7 +259,7 @@ static int tdspeed_apply(int32_t *buf_out[2], int32_t *buf_in[2],
}
assert(have + copy <= FIXED_BUFSIZE);
i = tdspeed_apply(buf_out, st->ovl_buff, have+copy, -1, out_size);
int i = tdspeed_apply(buf_out, st->ovl_buff, have+copy, -1, out_size);
dest[0] = buf_out[0] + i;
dest[1] = buf_out[1] + i;
@ -287,11 +287,11 @@ static int tdspeed_apply(int32_t *buf_out[2], int32_t *buf_in[2],
while (data_len - next_frame >= src_frame_sz)
{
/* find frame overlap by autocorelation */
int32_t const INC1 = 8;
int32_t const INC2 = 32;
int const INC1 = 8;
int const INC2 = 32;
int64_t min_delta = ~(1ll << 63); /* most positive */
shift = 0;
int shift = 0;
/* Power of 2 of a 28bit number requires 56bits, can accumulate
256times in a 64bit variable. */
@ -299,17 +299,17 @@ static int tdspeed_apply(int32_t *buf_out[2], int32_t *buf_in[2],
assert(next_frame + st->shift_max - 1 + st->dst_step - 1 < data_len);
assert(prev_frame + st->dst_step - 1 < data_len);
for (i = 0; i < st->shift_max; i += INC1)
for (int i = 0; i < st->shift_max; i += INC1)
{
int64_t delta = 0;
curr = buf_in[0] + next_frame + i;
prev = buf_in[0] + prev_frame;
int32_t *curr = buf_in[0] + next_frame + i;
int32_t *prev = buf_in[0] + prev_frame;
for (j = 0; j < st->dst_step; j += INC2, curr += INC2, prev += INC2)
for (int j = 0; j < st->dst_step; j += INC2, curr += INC2, prev += INC2)
{
int32_t diff = *curr - *prev;
delta += (int64_t)diff * diff;
delta += abs(diff);
if (delta >= min_delta)
goto skip;
@ -320,10 +320,10 @@ static int tdspeed_apply(int32_t *buf_out[2], int32_t *buf_in[2],
curr = buf_in[1] + next_frame + i;
prev = buf_in[1] + prev_frame;
for (j = 0; j < st->dst_step; j += INC2, curr += INC2, prev += INC2)
for (int j = 0; j < st->dst_step; j += INC2, curr += INC2, prev += INC2)
{
int32_t diff = *curr - *prev;
delta += (int64_t)diff * diff;
delta += abs(diff);
if (delta >= min_delta)
goto skip;
@ -336,16 +336,16 @@ skip:;
}
/* overlap fading-out previous frame with fading-in current frame */
curr = buf_in[0] + next_frame + shift;
prev = buf_in[0] + prev_frame;
int32_t *curr = buf_in[0] + next_frame + shift;
int32_t *prev = buf_in[0] + prev_frame;
d = dest[0];
int32_t *d = dest[0];
assert(next_frame + shift + st->dst_step - 1 < data_len);
assert(prev_frame + st->dst_step - 1 < data_len);
assert(dest[0] - buf_out[0] + st->dst_step - 1 < out_size);
for (i = 0, j = st->dst_step; j; i++, j--)
for (int i = 0, j = st->dst_step; j; i++, j--)
{
*d++ = (*curr++ * (int64_t)i +
*prev++ * (int64_t)j) >> st->dst_order;
@ -360,7 +360,7 @@ skip:;
d = dest[1];
for (i = 0, j = st->dst_step; j; i++, j--)
for (int i = 0, j = st->dst_step; j; i++, j--)
{
assert(d < buf_out[1] + out_size);
@ -388,7 +388,7 @@ skip:;
else if (last != 0)
{
/* last call: purge all remaining data to output buffer */
i = data_len - prev_frame;
int i = data_len - prev_frame;
assert(dest[0] + i <= buf_out[0] + out_size);
memcpy(dest[0], buf_in[0] + prev_frame, i * sizeof(int32_t));
@ -406,7 +406,7 @@ skip:;
{
/* preserve remaining data + needed overlap data for next call */
st->ovl_shift = next_frame - prev_frame;
i = (st->ovl_shift < 0) ? next_frame : prev_frame;
int i = (st->ovl_shift < 0) ? next_frame : prev_frame;
st->ovl_size = data_len - i;
assert(st->ovl_size <= FIXED_BUFSIZE);