portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-13 20:57:06 +01:00
Code Releases Activity

Vectorise the XYZT face gathering better.

Browse Source 
Hard coded for simd_layout <= 2 in any given spread out direction; full generality is inconsistent
with efficiency.
This commit is contained in:
paboyle
2017-02-15 11:11:04 +00:00
parent aca7a3ef0a
commit bd600702cf
10 changed files with 510 additions and 34 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
lib/tensors/Tensor_class.h
View File

@ -105,6 +105,11 @@ class iScalar {
friend strong_inline void rotate(iScalar<vtype> &out,const iScalar<vtype> &in,int rot){
rotate(out._internal,in._internal,rot);
}
friend strong_inline void exchange(iScalar<vtype> &out1,iScalar<vtype> &out2,
const iScalar<vtype> &in1,const iScalar<vtype> &in2,int type){
exchange(out1._internal,out2._internal,
in1._internal, in2._internal,type);
}
// Unary negation
friend strong_inline iScalar<vtype> operator-(const iScalar<vtype> &r) {
@ -248,6 +253,13 @@ class iVector {
rotate(out._internal[i],in._internal[i],rot);
}
}
friend strong_inline void exchange(iVector<vtype,N> &out1,iVector<vtype,N> &out2,
const iVector<vtype,N> &in1,const iVector<vtype,N> &in2,int type){
for(int i=0;i<N;i++){
exchange(out1._internal[i],out2._internal[i],
in1._internal[i], in2._internal[i],type);
}
}
// Unary negation
friend strong_inline iVector<vtype, N> operator-(const iVector<vtype, N> &r) {
@ -374,6 +386,14 @@ class iMatrix {
rotate(out._internal[i][j],in._internal[i][j],rot);
}}
}
friend strong_inline void exchange(iMatrix<vtype,N> &out1,iMatrix<vtype,N> &out2,
const iMatrix<vtype,N> &in1,const iMatrix<vtype,N> &in2,int type){
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
exchange(out1._internal[i][j],out2._internal[i][j],
in1._internal[i][j], in2._internal[i][j],type);
}}
}
// Unary negation
friend strong_inline iMatrix<vtype, N> operator-(const iMatrix<vtype, N> &r) {