portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-25 13:15:55 +01:00
Code Releases Activity

QPX exchange support

Browse Source
This commit is contained in:
Lanny91 2017-06-14 10:53:39 +01:00
parent 4a8c4ccfba
commit 2ad54c5a02
1 changed files with 43 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

43
lib/simd/Grid_qpx.h
View File

@ -375,6 +375,49 @@ namespace Optimization {
FLOAT_WRAP_2(operator(), inline) FLOAT_WRAP_2(operator(), inline)
}; };
//////////////////////////////////////////////
// Exchange support
#define FLOAT_WRAP_EXCHANGE(fn) \
static inline void fn(vector4float &out1, vector4float &out2, \
vector4float in1, vector4float in2) \
{ \
vector4double out1d, out2d, in1d, in2d; \
in1d = Vset()(in1); \
in2d = Vset()(in2); \
fn(out1d, out2d, in1d, in2d); \
Vstore()(out1d, out1); \
Vstore()(out2d, out2); \
}
struct Exchange{
// double precision
static inline void Exchange0(vector4double &out1, vector4double &out2,
vector4double in1, vector4double in2) {
out1 = vec_perm(in1, in2, vec_gpci(0145));
out2 = vec_perm(in1, in2, vec_gpci(02367));
}
static inline void Exchange1(vector4double &out1, vector4double &out2,
vector4double in1, vector4double in2) {
out1 = vec_perm(in1, in2, vec_gpci(0426));
out2 = vec_perm(in1, in2, vec_gpci(01537));
}
static inline void Exchange2(vector4double &out1, vector4double &out2,
vector4double in1, vector4double in2) {
assert(0);
}
static inline void Exchange3(vector4double &out1, vector4double &out2,
vector4double in1, vector4double in2) {
assert(0);
}
// single precision
FLOAT_WRAP_EXCHANGE(Exchange0);
FLOAT_WRAP_EXCHANGE(Exchange1);
FLOAT_WRAP_EXCHANGE(Exchange2);
FLOAT_WRAP_EXCHANGE(Exchange3);
};
struct Permute{ struct Permute{
//Complex double //Complex double
static inline vector4double Permute0(vector4double v){ //0123 -> 2301 static inline vector4double Permute0(vector4double v){ //0123 -> 2301