zcy
/
openEMS
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

engine_sse_compressed: speed up

pull/1/head
Sebastian Held 2010-10-07 14:39:54 +02:00
parent 4a42d707cf
commit 5cc6438ab7
1 changed files with 19 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

20
FDTD/engine_sse_compressed.cpp
View File

@ -16,7 +16,9 @@
*/ */
#include "engine_sse_compressed.h" #include "engine_sse_compressed.h"
#ifdef __SSE2__
#include <emmintrin.h>
#endif
Engine_SSE_Compressed* Engine_SSE_Compressed::New(const Operator_SSE_Compressed* op) Engine_SSE_Compressed* Engine_SSE_Compressed::New(const Operator_SSE_Compressed* op)
{ {
@ -68,18 +70,26 @@ void Engine_SSE_Compressed::UpdateVoltages(unsigned int startX, unsigned int num
// for pos[2] = 0 // for pos[2] = 0
// x-polarization // x-polarization
index = Op->m_Op_index[pos[0]][pos[1]][0]; index = Op->m_Op_index[pos[0]][pos[1]][0];
#ifdef __SSE2__
temp.v = (__m128)_mm_slli_si128( (__m128i)f4_curr[1][pos[0]][pos[1]][numVectors-1].v, 4 );
#else
temp.f[0] = 0; temp.f[0] = 0;
temp.f[1] = f4_curr[1][pos[0]][pos[1]][numVectors-1].f[0]; temp.f[1] = f4_curr[1][pos[0]][pos[1]][numVectors-1].f[0];
temp.f[2] = f4_curr[1][pos[0]][pos[1]][numVectors-1].f[1]; temp.f[2] = f4_curr[1][pos[0]][pos[1]][numVectors-1].f[1];
temp.f[3] = f4_curr[1][pos[0]][pos[1]][numVectors-1].f[2]; temp.f[3] = f4_curr[1][pos[0]][pos[1]][numVectors-1].f[2];
#endif
f4_volt[0][pos[0]][pos[1]][0].v *= Op->f4_vv_Compressed[0][index].v; f4_volt[0][pos[0]][pos[1]][0].v *= Op->f4_vv_Compressed[0][index].v;
f4_volt[0][pos[0]][pos[1]][0].v += Op->f4_vi_Compressed[0][index].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 ); f4_volt[0][pos[0]][pos[1]][0].v += Op->f4_vi_Compressed[0][index].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 // y-polarization
#ifdef __SSE2__
temp.v = (__m128)_mm_slli_si128( (__m128i)f4_curr[0][pos[0]][pos[1]][numVectors-1].v, 4 );
#else
temp.f[0] = 0; temp.f[0] = 0;
temp.f[1] = f4_curr[0][pos[0]][pos[1]][numVectors-1].f[0]; temp.f[1] = f4_curr[0][pos[0]][pos[1]][numVectors-1].f[0];
temp.f[2] = f4_curr[0][pos[0]][pos[1]][numVectors-1].f[1]; temp.f[2] = f4_curr[0][pos[0]][pos[1]][numVectors-1].f[1];
temp.f[3] = f4_curr[0][pos[0]][pos[1]][numVectors-1].f[2]; temp.f[3] = f4_curr[0][pos[0]][pos[1]][numVectors-1].f[2];
#endif
f4_volt[1][pos[0]][pos[1]][0].v *= Op->f4_vv_Compressed[1][index].v; f4_volt[1][pos[0]][pos[1]][0].v *= Op->f4_vv_Compressed[1][index].v;
f4_volt[1][pos[0]][pos[1]][0].v += Op->f4_vi_Compressed[1][index].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); f4_volt[1][pos[0]][pos[1]][0].v += Op->f4_vi_Compressed[1][index].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);
@ -121,18 +131,26 @@ void Engine_SSE_Compressed::UpdateCurrents(unsigned int startX, unsigned int num
index = Op->m_Op_index[pos[0]][pos[1]][numVectors-1]; index = Op->m_Op_index[pos[0]][pos[1]][numVectors-1];
// for pos[2] = numVectors-1 // for pos[2] = numVectors-1
// x-pol // x-pol
#ifdef __SSE2__
temp.v = (__m128)_mm_srli_si128( (__m128i)f4_volt[1][pos[0]][pos[1]][0].v, 4 );
#else
temp.f[0] = f4_volt[1][pos[0]][pos[1]][0].f[1]; 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[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[2] = f4_volt[1][pos[0]][pos[1]][0].f[3];
temp.f[3] = 0; temp.f[3] = 0;
#endif
f4_curr[0][pos[0]][pos[1]][numVectors-1].v *= Op->f4_ii_Compressed[0][index].v; f4_curr[0][pos[0]][pos[1]][numVectors-1].v *= Op->f4_ii_Compressed[0][index].v;
f4_curr[0][pos[0]][pos[1]][numVectors-1].v += Op->f4_iv_Compressed[0][index].v * ( f4_volt[2][pos[0]][pos[1]][numVectors-1].v - f4_volt[2][pos[0]][pos[1]+1][numVectors-1].v - f4_volt[1][pos[0]][pos[1]][numVectors-1].v + temp.v); f4_curr[0][pos[0]][pos[1]][numVectors-1].v += Op->f4_iv_Compressed[0][index].v * ( f4_volt[2][pos[0]][pos[1]][numVectors-1].v - f4_volt[2][pos[0]][pos[1]+1][numVectors-1].v - f4_volt[1][pos[0]][pos[1]][numVectors-1].v + temp.v);
// y-pol // y-pol
#ifdef __SSE2__
temp.v = (__m128)_mm_srli_si128( (__m128i)f4_volt[0][pos[0]][pos[1]][0].v, 4 );
#else
temp.f[0] = f4_volt[0][pos[0]][pos[1]][0].f[1]; 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[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[2] = f4_volt[0][pos[0]][pos[1]][0].f[3];
temp.f[3] = 0; temp.f[3] = 0;
#endif
f4_curr[1][pos[0]][pos[1]][numVectors-1].v *= Op->f4_ii_Compressed[1][index].v; f4_curr[1][pos[0]][pos[1]][numVectors-1].v *= Op->f4_ii_Compressed[1][index].v;
f4_curr[1][pos[0]][pos[1]][numVectors-1].v += Op->f4_iv_Compressed[1][index].v * ( f4_volt[0][pos[0]][pos[1]][numVectors-1].v - temp.v - f4_volt[2][pos[0]][pos[1]][numVectors-1].v + f4_volt[2][pos[0]+1][pos[1]][numVectors-1].v); f4_curr[1][pos[0]][pos[1]][numVectors-1].v += Op->f4_iv_Compressed[1][index].v * ( f4_volt[0][pos[0]][pos[1]][numVectors-1].v - temp.v - f4_volt[2][pos[0]][pos[1]][numVectors-1].v + f4_volt[2][pos[0]+1][pos[1]][numVectors-1].v);