Don't access the band splitter fields in the processing loops

perf shows a 5% drop in relative execution time on the alffplay example with an
audio-only file (20% to 15%). Kinda figured the optimizer would handle it
better, but I guess not.
This commit is contained in:
Chris Robinson 2016-05-31 07:50:23 -07:00
parent d1c4fb6364
commit 72d2febccb

View File

@ -34,35 +34,42 @@ static void bandsplit_process(BandSplitter *splitter, ALfloat *restrict hpout, A
const ALfloat *input, ALuint count)
{
ALfloat coeff, d, x;
ALfloat z1, z2;
ALuint i;
coeff = splitter->coeff*0.5f + 0.5f;
z1 = splitter->lp_z1;
z2 = splitter->lp_z2;
for(i = 0;i < count;i++)
{
x = input[i];
d = (x - splitter->lp_z1) * coeff;
x = splitter->lp_z1 + d;
splitter->lp_z1 = x + d;
d = (x - z1) * coeff;
x = z1 + d;
z1 = x + d;
d = (x - splitter->lp_z2) * coeff;
x = splitter->lp_z2 + d;
splitter->lp_z2 = x + d;
d = (x - z2) * coeff;
x = z2 + d;
z2 = x + d;
lpout[i] = x;
}
splitter->lp_z1 = z1;
splitter->lp_z2 = z2;
coeff = splitter->coeff;
z1 = splitter->hp_z1;
for(i = 0;i < count;i++)
{
x = input[i];
d = x - coeff*splitter->hp_z1;
x = splitter->hp_z1 + coeff*d;
splitter->hp_z1 = d;
d = x - coeff*z1;
x = z1 + coeff*d;
z1 = d;
hpout[i] = x - lpout[i];
}
splitter->hp_z1 = z1;
}