sse engine: changed the vector order to reduce copy operations
This commit is contained in:
parent
d26986fe0d
commit
63ef1b3ebf
@ -33,6 +33,7 @@ Engine_sse::Engine_sse(const Operator_sse* op) : Engine(op)
|
||||
{
|
||||
numLines[n] = Op->GetNumberOfLines(n);
|
||||
}
|
||||
numVectors = ceil((double)numLines[2]/4.0);
|
||||
}
|
||||
|
||||
Engine_sse::~Engine_sse()
|
||||
@ -63,34 +64,49 @@ void Engine_sse::UpdateVoltages()
|
||||
bool shift[2];
|
||||
f4vector temp;
|
||||
|
||||
unsigned int maxZ = numVectors;
|
||||
|
||||
for (pos[0]=0;pos[0]<numLines[0];++pos[0])
|
||||
{
|
||||
shift[0]=pos[0];
|
||||
for (pos[1]=0;pos[1]<numLines[1];++pos[1])
|
||||
{
|
||||
shift[1]=pos[1];
|
||||
for (pos[2]=0;pos[2]<ceil(numLines[2]/4);++pos[2])
|
||||
for (pos[2]=1;pos[2]<maxZ;++pos[2])
|
||||
{
|
||||
// x-polarization
|
||||
temp.f[0] = f4_curr[1][pos[0]][pos[1]][pos[2]-(bool)pos[2]].f[3];
|
||||
temp.f[1] = f4_curr[1][pos[0]][pos[1]][pos[2]].f[0];
|
||||
temp.f[2] = f4_curr[1][pos[0]][pos[1]][pos[2]].f[1];
|
||||
temp.f[3] = f4_curr[1][pos[0]][pos[1]][pos[2]].f[2];
|
||||
f4_volt[0][pos[0]][pos[1]][pos[2]].v *= Op->f4_vv[0][pos[0]][pos[1]][pos[2]].v;
|
||||
f4_volt[0][pos[0]][pos[1]][pos[2]].v += Op->f4_vi[0][pos[0]][pos[1]][pos[2]].v * ( f4_curr[2][pos[0]][pos[1]][pos[2]].v - f4_curr[2][pos[0]][pos[1]-shift[1]][pos[2]].v - f4_curr[1][pos[0]][pos[1]][pos[2]].v + temp.v );
|
||||
f4_volt[0][pos[0]][pos[1]][pos[2]].v += Op->f4_vi[0][pos[0]][pos[1]][pos[2]].v * ( f4_curr[2][pos[0]][pos[1]][pos[2]].v - f4_curr[2][pos[0]][pos[1]-shift[1]][pos[2]].v - f4_curr[1][pos[0]][pos[1]][pos[2]].v + f4_curr[1][pos[0]][pos[1]][pos[2]-1].v );
|
||||
|
||||
// y-polarization
|
||||
temp.f[0] = f4_curr[0][pos[0]][pos[1]][pos[2]-(bool)pos[2]].f[3];
|
||||
temp.f[1] = f4_curr[0][pos[0]][pos[1]][pos[2]].f[0];
|
||||
temp.f[2] = f4_curr[0][pos[0]][pos[1]][pos[2]].f[1];
|
||||
temp.f[3] = f4_curr[0][pos[0]][pos[1]][pos[2]].f[2];
|
||||
f4_volt[1][pos[0]][pos[1]][pos[2]].v *= Op->f4_vv[1][pos[0]][pos[1]][pos[2]].v;
|
||||
f4_volt[1][pos[0]][pos[1]][pos[2]].v += Op->f4_vi[1][pos[0]][pos[1]][pos[2]].v * ( f4_curr[0][pos[0]][pos[1]][pos[2]].v - temp.v - f4_curr[2][pos[0]][pos[1]][pos[2]].v + f4_curr[2][pos[0]-shift[0]][pos[1]][pos[2]].v);
|
||||
f4_volt[1][pos[0]][pos[1]][pos[2]].v += Op->f4_vi[1][pos[0]][pos[1]][pos[2]].v * ( f4_curr[0][pos[0]][pos[1]][pos[2]].v - f4_curr[0][pos[0]][pos[1]][pos[2]-1].v - f4_curr[2][pos[0]][pos[1]][pos[2]].v + f4_curr[2][pos[0]-shift[0]][pos[1]][pos[2]].v);
|
||||
|
||||
// z-polarization
|
||||
f4_volt[2][pos[0]][pos[1]][pos[2]].v *= Op->f4_vv[2][pos[0]][pos[1]][pos[2]].v;
|
||||
f4_volt[2][pos[0]][pos[1]][pos[2]].v += Op->f4_vi[2][pos[0]][pos[1]][pos[2]].v * ( f4_curr[1][pos[0]][pos[1]][pos[2]].v - f4_curr[1][pos[0]-shift[0]][pos[1]][pos[2]].v - f4_curr[0][pos[0]][pos[1]][pos[2]].v + f4_curr[0][pos[0]][pos[1]-shift[1]][pos[2]].v);
|
||||
}
|
||||
|
||||
// for pos[2] = 0
|
||||
// x-polarization
|
||||
temp.f[0] = 0;
|
||||
temp.f[1] = f4_curr[1][pos[0]][pos[1]][maxZ-1].f[0];
|
||||
temp.f[2] = f4_curr[1][pos[0]][pos[1]][maxZ-1].f[1];
|
||||
temp.f[3] = f4_curr[1][pos[0]][pos[1]][maxZ-1].f[2];
|
||||
f4_volt[0][pos[0]][pos[1]][0].v *= Op->f4_vv[0][pos[0]][pos[1]][0].v;
|
||||
f4_volt[0][pos[0]][pos[1]][0].v += Op->f4_vi[0][pos[0]][pos[1]][0].v * ( f4_curr[2][pos[0]][pos[1]][0].v - f4_curr[2][pos[0]][pos[1]-shift[1]][0].v - f4_curr[1][pos[0]][pos[1]][0].v + temp.v );
|
||||
|
||||
// y-polarization
|
||||
temp.f[0] = 0;
|
||||
temp.f[1] = f4_curr[0][pos[0]][pos[1]][maxZ-1].f[0];
|
||||
temp.f[2] = f4_curr[0][pos[0]][pos[1]][maxZ-1].f[1];
|
||||
temp.f[3] = f4_curr[0][pos[0]][pos[1]][maxZ-1].f[2];
|
||||
f4_volt[1][pos[0]][pos[1]][0].v *= Op->f4_vv[1][pos[0]][pos[1]][0].v;
|
||||
f4_volt[1][pos[0]][pos[1]][0].v += Op->f4_vi[1][pos[0]][pos[1]][0].v * ( f4_curr[0][pos[0]][pos[1]][0].v - temp.v - f4_curr[2][pos[0]][pos[1]][0].v + f4_curr[2][pos[0]-shift[0]][pos[1]][0].v);
|
||||
|
||||
// z-polarization
|
||||
f4_volt[2][pos[0]][pos[1]][0].v *= Op->f4_vv[2][pos[0]][pos[1]][0].v;
|
||||
f4_volt[2][pos[0]][pos[1]][0].v += Op->f4_vi[2][pos[0]][pos[1]][0].v * ( f4_curr[1][pos[0]][pos[1]][0].v - f4_curr[1][pos[0]-shift[0]][pos[1]][0].v - f4_curr[0][pos[0]][pos[1]][0].v + f4_curr[0][pos[0]][pos[1]-shift[1]][0].v);
|
||||
}
|
||||
}
|
||||
}
|
||||
@ -100,32 +116,47 @@ void Engine_sse::UpdateCurrents()
|
||||
unsigned int pos[5];
|
||||
f4vector temp;
|
||||
|
||||
unsigned int maxZ = numVectors;
|
||||
|
||||
for (pos[0]=0;pos[0]<numLines[0]-1;++pos[0])
|
||||
{
|
||||
for (pos[1]=0;pos[1]<numLines[1]-1;++pos[1])
|
||||
{
|
||||
for (pos[2]=0;pos[2]<ceil(numLines[2]/4);++pos[2]) // FIXME is this correct?
|
||||
for (pos[2]=0;pos[2]<maxZ-1;++pos[2])
|
||||
{
|
||||
// x-pol
|
||||
temp.f[0] = f4_volt[1][pos[0]][pos[1]][pos[2]].f[1];
|
||||
temp.f[1] = f4_volt[1][pos[0]][pos[1]][pos[2]].f[2];
|
||||
temp.f[2] = f4_volt[1][pos[0]][pos[1]][pos[2]].f[3];
|
||||
temp.f[3] = f4_volt[1][pos[0]][pos[1]][pos[2]+1].f[0]; // FIXME outside sim area
|
||||
f4_curr[0][pos[0]][pos[1]][pos[2]].v *= Op->f4_ii[0][pos[0]][pos[1]][pos[2]].v;
|
||||
f4_curr[0][pos[0]][pos[1]][pos[2]].v += Op->f4_iv[0][pos[0]][pos[1]][pos[2]].v * ( f4_volt[2][pos[0]][pos[1]][pos[2]].v - f4_volt[2][pos[0]][pos[1]+1][pos[2]].v - f4_volt[1][pos[0]][pos[1]][pos[2]].v + temp.v);
|
||||
f4_curr[0][pos[0]][pos[1]][pos[2]].v += Op->f4_iv[0][pos[0]][pos[1]][pos[2]].v * ( f4_volt[2][pos[0]][pos[1]][pos[2]].v - f4_volt[2][pos[0]][pos[1]+1][pos[2]].v - f4_volt[1][pos[0]][pos[1]][pos[2]].v + f4_volt[1][pos[0]][pos[1]][pos[2]+1].v);
|
||||
|
||||
// y-pol
|
||||
temp.f[0] = f4_volt[0][pos[0]][pos[1]][pos[2]].f[1];
|
||||
temp.f[1] = f4_volt[0][pos[0]][pos[1]][pos[2]].f[2];
|
||||
temp.f[2] = f4_volt[0][pos[0]][pos[1]][pos[2]].f[3];
|
||||
temp.f[3] = f4_volt[0][pos[0]][pos[1]][pos[2]+1].f[0]; // FIXME outside sim area
|
||||
f4_curr[1][pos[0]][pos[1]][pos[2]].v *= Op->f4_ii[1][pos[0]][pos[1]][pos[2]].v;
|
||||
f4_curr[1][pos[0]][pos[1]][pos[2]].v += Op->f4_iv[1][pos[0]][pos[1]][pos[2]].v * ( f4_volt[0][pos[0]][pos[1]][pos[2]].v - temp.v - f4_volt[2][pos[0]][pos[1]][pos[2]].v + f4_volt[2][pos[0]+1][pos[1]][pos[2]].v);
|
||||
f4_curr[1][pos[0]][pos[1]][pos[2]].v += Op->f4_iv[1][pos[0]][pos[1]][pos[2]].v * ( f4_volt[0][pos[0]][pos[1]][pos[2]].v - f4_volt[0][pos[0]][pos[1]][pos[2]+1].v - f4_volt[2][pos[0]][pos[1]][pos[2]].v + f4_volt[2][pos[0]+1][pos[1]][pos[2]].v);
|
||||
|
||||
// z-pol
|
||||
f4_curr[2][pos[0]][pos[1]][pos[2]].v *= Op->f4_ii[2][pos[0]][pos[1]][pos[2]].v;
|
||||
f4_curr[2][pos[0]][pos[1]][pos[2]].v += Op->f4_iv[2][pos[0]][pos[1]][pos[2]].v * ( f4_volt[1][pos[0]][pos[1]][pos[2]].v - f4_volt[1][pos[0]+1][pos[1]][pos[2]].v - f4_volt[0][pos[0]][pos[1]][pos[2]].v + f4_volt[0][pos[0]][pos[1]+1][pos[2]].v);
|
||||
}
|
||||
|
||||
// for pos[2] = maxZ-1
|
||||
// x-pol
|
||||
temp.f[0] = f4_volt[1][pos[0]][pos[1]][0].f[1];
|
||||
temp.f[1] = f4_volt[1][pos[0]][pos[1]][0].f[2];
|
||||
temp.f[2] = f4_volt[1][pos[0]][pos[1]][0].f[3];
|
||||
temp.f[3] = 0;
|
||||
f4_curr[0][pos[0]][pos[1]][maxZ-1].v *= Op->f4_ii[0][pos[0]][pos[1]][maxZ-1].v;
|
||||
f4_curr[0][pos[0]][pos[1]][maxZ-1].v += Op->f4_iv[0][pos[0]][pos[1]][maxZ-1].v * ( f4_volt[2][pos[0]][pos[1]][maxZ-1].v - f4_volt[2][pos[0]][pos[1]+1][maxZ-1].v - f4_volt[1][pos[0]][pos[1]][maxZ-1].v + temp.v);
|
||||
|
||||
// y-pol
|
||||
temp.f[0] = f4_volt[0][pos[0]][pos[1]][0].f[1];
|
||||
temp.f[1] = f4_volt[0][pos[0]][pos[1]][0].f[2];
|
||||
temp.f[2] = f4_volt[0][pos[0]][pos[1]][0].f[3];
|
||||
temp.f[3] = 0;
|
||||
f4_curr[1][pos[0]][pos[1]][maxZ-1].v *= Op->f4_ii[1][pos[0]][pos[1]][maxZ-1].v;
|
||||
f4_curr[1][pos[0]][pos[1]][maxZ-1].v += Op->f4_iv[1][pos[0]][pos[1]][maxZ-1].v * ( f4_volt[0][pos[0]][pos[1]][maxZ-1].v - temp.v - f4_volt[2][pos[0]][pos[1]][maxZ-1].v + f4_volt[2][pos[0]+1][pos[1]][maxZ-1].v);
|
||||
|
||||
// z-pol
|
||||
f4_curr[2][pos[0]][pos[1]][maxZ-1].v *= Op->f4_ii[2][pos[0]][pos[1]][maxZ-1].v;
|
||||
f4_curr[2][pos[0]][pos[1]][maxZ-1].v += Op->f4_iv[2][pos[0]][pos[1]][maxZ-1].v * ( f4_volt[1][pos[0]][pos[1]][maxZ-1].v - f4_volt[1][pos[0]+1][pos[1]][maxZ-1].v - f4_volt[0][pos[0]][pos[1]][maxZ-1].v + f4_volt[0][pos[0]][pos[1]+1][maxZ-1].v);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
@ -32,8 +32,8 @@ public:
|
||||
|
||||
virtual unsigned int GetNumberOfTimesteps() {return numTS;};
|
||||
|
||||
inline virtual FDTD_FLOAT& GetVolt( unsigned int n, unsigned int x, unsigned int y, unsigned int z ) const { return f4_volt[n][x][y][z/4].f[z%4]; }
|
||||
inline virtual FDTD_FLOAT& GetCurr( unsigned int n, unsigned int x, unsigned int y, unsigned int z ) const { return f4_curr[n][x][y][z/4].f[z%4]; }
|
||||
inline virtual FDTD_FLOAT& GetVolt( unsigned int n, unsigned int x, unsigned int y, unsigned int z ) const { return f4_volt[n][x][y][z%numVectors].f[z/numVectors]; }
|
||||
inline virtual FDTD_FLOAT& GetCurr( unsigned int n, unsigned int x, unsigned int y, unsigned int z ) const { return f4_curr[n][x][y][z%numVectors].f[z/numVectors]; }
|
||||
|
||||
protected:
|
||||
Engine_sse(const Operator_sse* op);
|
||||
@ -42,6 +42,8 @@ protected:
|
||||
virtual void UpdateVoltages();
|
||||
virtual void UpdateCurrents();
|
||||
|
||||
unsigned int numVectors;
|
||||
|
||||
f4vector**** f4_volt;
|
||||
f4vector**** f4_curr;
|
||||
};
|
||||
|
@ -63,5 +63,7 @@ void Operator_sse::InitOperator()
|
||||
f4_vi = Create_N_3DArray_v4sf(numLines);
|
||||
f4_iv = Create_N_3DArray_v4sf(numLines);
|
||||
f4_ii = Create_N_3DArray_v4sf(numLines);
|
||||
|
||||
numVectors = ceil((double)numLines[2]/4.0);
|
||||
}
|
||||
|
||||
|
@ -28,11 +28,11 @@ public:
|
||||
static Operator_sse* New();
|
||||
virtual ~Operator_sse();
|
||||
|
||||
inline virtual FDTD_FLOAT& GetVV( unsigned int n, unsigned int x, unsigned int y, unsigned int z ) const { return f4_vv[n][x][y][z/4].f[z%4]; }
|
||||
inline virtual FDTD_FLOAT& GetVI( unsigned int n, unsigned int x, unsigned int y, unsigned int z ) const { return f4_vi[n][x][y][z/4].f[z%4]; }
|
||||
inline virtual FDTD_FLOAT& GetVV( unsigned int n, unsigned int x, unsigned int y, unsigned int z ) const { return f4_vv[n][x][y][z%numVectors].f[z/numVectors]; }
|
||||
inline virtual FDTD_FLOAT& GetVI( unsigned int n, unsigned int x, unsigned int y, unsigned int z ) const { return f4_vi[n][x][y][z%numVectors].f[z/numVectors]; }
|
||||
|
||||
inline virtual FDTD_FLOAT& GetII( unsigned int n, unsigned int x, unsigned int y, unsigned int z ) const { return f4_ii[n][x][y][z/4].f[z%4]; }
|
||||
inline virtual FDTD_FLOAT& GetIV( unsigned int n, unsigned int x, unsigned int y, unsigned int z ) const { return f4_iv[n][x][y][z/4].f[z%4]; }
|
||||
inline virtual FDTD_FLOAT& GetII( unsigned int n, unsigned int x, unsigned int y, unsigned int z ) const { return f4_ii[n][x][y][z%numVectors].f[z/numVectors]; }
|
||||
inline virtual FDTD_FLOAT& GetIV( unsigned int n, unsigned int x, unsigned int y, unsigned int z ) const { return f4_iv[n][x][y][z%numVectors].f[z/numVectors]; }
|
||||
|
||||
protected:
|
||||
//! use New() for creating a new Operator
|
||||
@ -42,6 +42,8 @@ protected:
|
||||
virtual void Reset();
|
||||
virtual void InitOperator();
|
||||
|
||||
unsigned int numVectors;
|
||||
|
||||
// engine/post-proc needs access
|
||||
public:
|
||||
f4vector**** f4_vv; //calc new voltage from old voltage
|
||||
|
@ -149,7 +149,7 @@ void Delete_N_3DArray_v4sf(f4vector**** array, const unsigned int* numLines)
|
||||
|
||||
f4vector*** Create3DArray_v4sf(const unsigned int* numLines)
|
||||
{
|
||||
unsigned int numZ = ceil(numLines[2]/4)+1;
|
||||
unsigned int numZ = ceil((double)numLines[2]/4.0);
|
||||
|
||||
f4vector*** array=NULL;
|
||||
unsigned int pos[3];
|
||||
|
Loading…
Reference in New Issue
Block a user