eden-emu/opus
20
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Split cwrsi() by pulses vs. dimensions.

Browse source 
This lets us cut out a bunch of work in the large _n, small _k case
 where most of the dimensions won't have any pulses.
It also gets rid of all remaining usage of CELT_PVQ_U() in cwrsi(),
 leaving just a single test instead of lots of mins and maxes, and
 makes a bunch of the jump threading more obvious.

This is a 1.6% decoder speedup on a 96 kbps comp48-stereo encode on
 a Cortex A8.
This commit is contained in:
Timothy B. Terriberry 2013-06-14 23:57:19 -07:00 • committed by Jean-Marc Valin
parent 63f744d583
commit ce15e65319
1 changed files with 35 additions and 20 deletions
Download patch file Download diff file Expand all files Collapse all files

47
celt/cwrs.c
View file

@ -467,34 +467,49 @@ static void cwrsi(int _n,int _k,opus_uint32 _i,int *_y){
celt_assert(_k>0); celt_assert(_k>0);
celt_assert(_n>1); celt_assert(_n>1);
while(_n>2){ while(_n>2){
opus_uint32 q;
/*Lots of pulses case:*/
if(_k>=_n){
const opus_uint32 *row;
row=CELT_PVQ_U_ROW[_n];
/*Are the pulses in this dimension negative?*/ /*Are the pulses in this dimension negative?*/
p=CELT_PVQ_U(_n,_k+1); p=row[_k+1];
s=-(_i>=p); s=-(_i>=p);
_i-=p&s; _i-=p&s;
/*Count how many pulses were placed in this dimension.*/ /*Count how many pulses were placed in this dimension.*/
k0=_k; k0=_k;
p=CELT_PVQ_U(_n,_k);
if(_k>_n){
const opus_uint32 *row;
opus_uint32 q;
row=CELT_PVQ_U_ROW[_n];
q=row[_n]; q=row[_n];
if(q>_i){ if(q>_i){
celt_assert(p>q); celt_assert(p>q);
/*Setting p=q is unnecessary, but it helps the optimizer prove p>_i,
allowing it to jump straight past the initial test in the second
loop below.
Once it's removed that first comparison, a smart compiler should be
able to figure out that the result of this assignment isn't used and
optimize it away anyway.*/
p=q;
_k=_n; _k=_n;
do p=CELT_PVQ_U_ROW[--_k][_n];
while(p>_i);
} }
else for(;p>_i;p=row[_k])_k--; else for(p=row[_k];p>_i;p=row[_k])_k--;
}
for(;p>_i;p=CELT_PVQ_U_ROW[_k][_n])_k--;
_i-=p; _i-=p;
*_y++=(k0-_k+s)^s; *_y++=(k0-_k+s)^s;
}
/*Lots of dimensions case:*/
else{
/*Are there any pulses in this dimension at all?*/
p=CELT_PVQ_U_ROW[_k][_n];
q=CELT_PVQ_U_ROW[_k+1][_n];
if(p<=_i&&_i<q){
_i-=p;
*_y++=0;
}
else{
/*Are the pulses in this dimension negative?*/
s=-(_i>=q);
_i-=q&s;
/*Count how many pulses were placed in this dimension.*/
k0=_k;
do p=CELT_PVQ_U_ROW[--_k][_n];
while(p>_i);
_i-=p;
*_y++=(k0-_k+s)^s;
}
}
_n--; _n--;
} }
/*_n==2*/ /*_n==2*/