portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-17 23:37:06 +01:00
Code Releases Activity

Merge branch 'develop' into bugfix/LatTransfer

Browse Source 
* develop:
  Better SIMD usage/coalescence
This commit is contained in:
Michael Marshall
2021-02-27 00:21:36 +00:00
parent 69f41469dd 2e61556389
commit 1059a81a3c
5 changed files with 143 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files

23
Grid/simd/Grid_gpu_vec.h
View File

@ -60,11 +60,25 @@ template<class pair>
class GpuComplex {
public:
pair z;
typedef decltype(z.x) real;
typedef decltype(z.x) Real;
public:
accelerator_inline GpuComplex() = default;
accelerator_inline GpuComplex(real re,real im) { z.x=re; z.y=im; };
accelerator_inline GpuComplex(Real re,Real im) { z.x=re; z.y=im; };
accelerator_inline GpuComplex(const GpuComplex &zz) { z = zz.z;};
accelerator_inline Real real(void) const { return z.x; };
accelerator_inline Real imag(void) const { return z.y; };
accelerator_inline GpuComplex &operator*=(const GpuComplex &r) {
*this = (*this) * r;
return *this;
}
accelerator_inline GpuComplex &operator+=(const GpuComplex &r) {
*this = (*this) + r;
return *this;
}
accelerator_inline GpuComplex &operator-=(const GpuComplex &r) {
*this = (*this) - r;
return *this;
}
friend accelerator_inline GpuComplex operator+(const GpuComplex &lhs,const GpuComplex &rhs) {
GpuComplex r ;
r.z.x = lhs.z.x + rhs.z.x;
@ -157,6 +171,11 @@ typedef GpuVector<NSIMD_RealD, double > GpuVectorRD;
typedef GpuVector<NSIMD_ComplexD, GpuComplexD > GpuVectorCD;
typedef GpuVector<NSIMD_Integer, Integer > GpuVectorI;
accelerator_inline GpuComplexF timesI(const GpuComplexF &r) { return(GpuComplexF(-r.imag(),r.real()));}
accelerator_inline GpuComplexD timesI(const GpuComplexD &r) { return(GpuComplexD(-r.imag(),r.real()));}
accelerator_inline GpuComplexF timesMinusI(const GpuComplexF &r){ return(GpuComplexF(r.imag(),-r.real()));}
accelerator_inline GpuComplexD timesMinusI(const GpuComplexD &r){ return(GpuComplexD(r.imag(),-r.real()));}
accelerator_inline float half2float(half h)
{
float f;

12
Grid/simd/Simd.h
View File

@ -148,10 +148,14 @@ accelerator_inline void sub (ComplexF * __restrict__ y,const ComplexF * __restri
accelerator_inline void add (ComplexF * __restrict__ y,const ComplexF * __restrict__ l,const ComplexF *__restrict__ r){ *y = (*l) + (*r); }
//conjugate already supported for complex
accelerator_inline ComplexF timesI(const ComplexF &r) { return(r*ComplexF(0.0,1.0));}
accelerator_inline ComplexD timesI(const ComplexD &r) { return(r*ComplexD(0.0,1.0));}
accelerator_inline ComplexF timesMinusI(const ComplexF &r){ return(r*ComplexF(0.0,-1.0));}
accelerator_inline ComplexD timesMinusI(const ComplexD &r){ return(r*ComplexD(0.0,-1.0));}
accelerator_inline ComplexF timesI(const ComplexF &r) { return(ComplexF(-r.imag(),r.real()));}
accelerator_inline ComplexD timesI(const ComplexD &r) { return(ComplexD(-r.imag(),r.real()));}
accelerator_inline ComplexF timesMinusI(const ComplexF &r){ return(ComplexF(r.imag(),-r.real()));}
accelerator_inline ComplexD timesMinusI(const ComplexD &r){ return(ComplexD(r.imag(),-r.real()));}
//accelerator_inline ComplexF timesI(const ComplexF &r) { return(r*ComplexF(0.0,1.0));}
//accelerator_inline ComplexD timesI(const ComplexD &r) { return(r*ComplexD(0.0,1.0));}
//accelerator_inline ComplexF timesMinusI(const ComplexF &r){ return(r*ComplexF(0.0,-1.0));}
//accelerator_inline ComplexD timesMinusI(const ComplexD &r){ return(r*ComplexD(0.0,-1.0));}
// define projections to real and imaginay parts
accelerator_inline ComplexF projReal(const ComplexF &r){return( ComplexF(r.real(), 0.0));}