LIS_INT lis_matvec_ilu(LIS_MATRIX A, LIS_MATRIX_ILU LU, LIS_VECTOR X, LIS_VECTOR Y)
{
LIS_INT i,j,jj,n,np;
LIS_SCALAR *x,*y;
#ifdef _OPENMP
LIS_INT nprocs,k;
LIS_SCALAR t,*w;
#endif
#ifdef USE_QUAD_PRECISION
LIS_INT j0,j1;
#ifdef _OPENMP
LIS_SCALAR *ww,*wwl;
#endif
#endif
LIS_QUAD_DECLAR;
LIS_DEBUG_FUNC_IN;
np = A->np;
n = LU->n;
x = X->value;
y = Y->value;
#ifdef USE_QUAD_PRECISION
if( X->precision==LIS_PRECISION_DEFAULT )
#endif
{
#ifdef USE_MPI
LIS_MATVEC_SENDRECV;
#endif
#ifdef _OPENMP
nprocs = omp_get_max_threads();
w = (LIS_SCALAR *)lis_malloc( nprocs*np*sizeof(LIS_SCALAR),"lis_matvect_crs::w" );
#pragma omp parallel private(i,j,k,jj,t)
{
k = omp_get_thread_num();
#pragma omp for
for(j=0;j<nprocs;j++)
{
memset( &w[j*np], 0, np*sizeof(LIS_SCALAR) );
}
#pragma omp for
for(i=0;i<n;i++)
{
for(j=0;j<LU->nnz[i];j++)
{
jj = k*np + LU->index[i][j];
w[jj] += LU->value[i][j] * X->value[i];
}
}
#pragma omp for
for(i=0;i<np;i++)
{
t = 0.0;
for(j=0;j<nprocs;j++)
{
t += w[j*np+i];
}
Y->value[i] = t;
}
}
lis_free(w);
#else
for(i=0;i<np;i++)
{
Y->value[i] = 0.0;
}
for(i=0;i<n;i++)
{
for(j=0;j<LU->nnz[i];j++)
{
jj = LU->index[i][j];
Y->value[jj] += LU->value[i][j] * X->value[i];
}
}
#endif
}
#ifdef USE_QUAD_PRECISION
else
{
#ifdef USE_MPI
lis_send_recv_mp(A->commtable,X);
#endif
#ifdef _OPENMP
#ifndef USE_FMA2_SSE2
nprocs = omp_get_max_threads();
ww = (LIS_SCALAR *)lis_malloc( 2*nprocs*np*sizeof(LIS_SCALAR),"lis_matvect_crs_mp::ww" );
wwl = &ww[nprocs*np];
#ifndef USE_SSE2
#pragma omp parallel private(i,j,jj,k,p1,p2,tq,bhi,blo,chi,clo,sh,sl,th,tl,eh,el)
#else
#pragma omp parallel private(i,j,jj,k,bh,ch,sh,wh,th,bl,cl,sl,wl,tl,p1,p2,t0,t1,t2,eh)
#endif
{
k = omp_get_thread_num();
#pragma omp for
for(j=0;j<nprocs;j++)
{
memset( &ww[j*np], 0, np*sizeof(LIS_SCALAR) );
memset( &wwl[j*np], 0, np*sizeof(LIS_SCALAR) );
}
#pragma omp for
for(i=0;i<n;i++)
{
for(j=0;j<LU->nnz[i];j++)
{
jj = k*np + LU->index[i][j];
#ifndef USE_SSE2
LIS_QUAD_FMAD(ww[jj],wwl[jj],ww[jj],wwl[jj],X->value[i],X->value_lo[i],LU->value[i][j]);
#else
LIS_QUAD_FMAD_SSE2(ww[jj],wwl[jj],ww[jj],wwl[jj],X->value[i],X->value_lo[i],LU->value[i][j]);
#endif
}
}
#pragma omp for
for(i=0;i<np;i++)
{
Y->value[i] = Y->value_lo[i] = 0.0;
for(j=0;j<nprocs;j++)
{
#ifndef USE_SSE2
LIS_QUAD_ADD(Y->value[i],Y->value_lo[i],Y->value[i],Y->value_lo[i],ww[j*np+i],wwl[j*np+i]);
#else
LIS_QUAD_ADD_SSE2(Y->value[i],Y->value_lo[i],Y->value[i],Y->value_lo[i],ww[j*np+i],wwl[j*np+i]);
#endif
}
}
}
lis_free(ww);
#else
nprocs = omp_get_max_threads();
ww = (LIS_SCALAR *)lis_malloc( 2*nprocs*np*sizeof(LIS_SCALAR), "lis_matvect_crs_mp2::ww" );
wwl = &ww[nprocs*np];
#pragma omp parallel private(i,j,j0,j1,k,bh,ch,sh,wh,th,bl,cl,sl,wl,tl,p1,p2,t0,t1,t2,eh)
{
k = omp_get_thread_num();
#pragma omp for
for(j=0;j<nprocs;j++)
{
memset( &ww[j*np], 0, np*sizeof(LIS_SCALAR) );
memset( &wwl[j*np], 0, np*sizeof(LIS_SCALAR) );
}
#pragma omp for
for(i=0; i<n; i++)
{
for(j=0;j<LU->nnz[i]-1;j+=2)
{
j0 = k*np + LU->index[i][j];
j1 = k*np + LU->index[i][j+1];
#ifdef USE_SSE2
LIS_QUAD_FMAD2_SSE2_STSD(ww[j0],wwl[j0],ww[j1],wwl[j1],ww[j0],wwl[j0],ww[j1],wwl[j1],X->value[i],X->value_lo[i],X->value[i],X->value_lo[i],LU->value[i][j]);
#endif
}
for(;j<LU->nnz[i];j++)
{
j0 = LU->index[i][j];
#ifdef USE_SSE2
LIS_QUAD_FMAD_SSE2(ww[j0],wwl[j0],ww[j0],wwl[j0],X->value[i],X->value_lo[i],LU->value[i][j]);
#endif
}
}
#pragma omp for
for(i=0;i<np;i++)
{
Y->value[i] = Y->value_lo[i] = 0.0;
for(j=0;j<nprocs;j++)
{
#ifdef USE_SSE2
LIS_QUAD_ADD_SSE2(Y->value[i],Y->value_lo[i],Y->value[i],Y->value_lo[i],ww[j*np+i],wwl[j*np+i]);
#endif
}
}
}
lis_free(ww);
#endif
#else
#ifndef USE_FMA2_SSE2
for(i=0;i<np;i++)
{
Y->value[i] = 0.0;
Y->value_lo[i] = 0.0;
}
for(i=0;i<n;i++)
{
for(j=0;j<LU->nnz[i];j++)
{
jj = LU->index[i][j];
#ifndef USE_SSE2
LIS_QUAD_FMAD(Y->value[jj],Y->value_lo[jj],Y->value[jj],Y->value_lo[jj],X->value[i],X->value_lo[i],LU->value[i][j]);
#else
LIS_QUAD_FMAD_SSE2(Y->value[jj],Y->value_lo[jj],Y->value[jj],Y->value_lo[jj],X->value[i],X->value_lo[i],LU->value[i][j]);
#endif
}
}
#else
for(i=0; i<np; i++)
{
Y->value[i] = 0.0;
Y->value_lo[i] = 0.0;
}
for(i=0; i<n; i++)
{
for(j=0;j<LU->nnz[i]-1;j+=2)
{
j0 = LU->index[i][j];
j1 = LU->index[i][j+1];
#ifdef USE_SSE2
LIS_QUAD_FMAD2_SSE2_STSD(Y->value[j0],Y->value_lo[j0],Y->value[j1],Y->value_lo[j1],Y->value[j0],Y->value_lo[j0],Y->value[j1],Y->value_lo[j1],X->value[i],X->value_lo[i],X->value[i],X->value_lo[i],LU->value[i][j]);
#endif
}
for(;j<LU->nnz[i];j++)
{
j0 = LU->index[i][j];
#ifdef USE_SSE2
LIS_QUAD_FMAD_SSE2(Y->value[j0],Y->value_lo[j0],Y->value[j0],Y->value_lo[j0],X->value[i],X->value_lo[i],LU->value[i][j]);
#endif
}
}
#endif
#endif
}
#endif
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
}
LIS_INT lis_matrix_solvet_csr(LIS_MATRIX A, LIS_VECTOR B, LIS_VECTOR X, LIS_INT flag)
{
LIS_INT i,j,jj,n;
LIS_SCALAR t;
LIS_SCALAR *x;
#ifdef _OPENMP
LIS_INT is,ie,my_rank,nprocs;
#endif
#ifdef USE_QUAD_PRECISION
LIS_QUAD w1,w2;
LIS_SCALAR *xl;
#endif
LIS_QUAD_DECLAR;
LIS_DEBUG_FUNC_IN;
n = A->n;
x = X->value;
#ifdef USE_QUAD_PRECISION
xl = X->value_lo;
#endif
#ifdef USE_QUAD_PRECISION
if( B->precision==LIS_PRECISION_DEFAULT )
{
#endif
lis_vector_copy(B,X);
#ifdef USE_QUAD_PRECISION
}
else
{
lis_vector_copyex_mm(B,X);
}
#endif
switch(flag)
{
case LIS_MATRIX_LOWER:
for(i=0;i<n;i++)
{
x[i] = x[i] * A->WD->value[i];
for(j=A->U->ptr[i];j<A->U->ptr[i+1];j++)
{
x[A->U->index[j]] -= A->U->value[j] * x[i];
}
}
break;
case LIS_MATRIX_UPPER:
for(i=n-1;i>=0;i--)
{
x[i] = x[i] * A->WD->value[i];
for(j=A->L->ptr[i];j<A->L->ptr[i+1];j++)
{
x[A->L->index[j]] -= A->L->value[j] * x[i];
}
}
break;
case LIS_MATRIX_SSOR:
#ifdef USE_QUAD_PRECISION
if( B->precision==LIS_PRECISION_DEFAULT )
{
#endif
#ifdef _OPENMP
nprocs = omp_get_max_threads();
#pragma omp parallel private(i,j,jj,t,is,ie,my_rank)
{
my_rank = omp_get_thread_num();
LIS_GET_ISIE(my_rank,nprocs,n,is,ie);
for(i=is;i<ie;i++)
{
t = x[i] * A->WD->value[i];
for(j=A->U->ptr[i];j<A->U->ptr[i+1];j++)
{
jj = A->U->index[j];
if( jj<is || jj>=ie ) continue;
x[jj] -= A->U->value[j] * t;
}
}
for(i=ie-1;i>=is;i--)
{
t = x[i] * A->WD->value[i];
x[i] = t;
for(j=A->L->ptr[i];j<A->L->ptr[i+1];j++)
{
jj = A->L->index[j];
if( jj<is ) continue;
x[jj] -= A->L->value[j] * t;
}
}
}
#else
for(i=0;i<n;i++)
{
t = x[i] * A->WD->value[i];
for(j=A->U->ptr[i];j<A->U->ptr[i+1];j++)
{
x[A->U->index[j]] -= A->U->value[j] * t;
}
}
for(i=n-1;i>=0;i--)
{
t = x[i] * A->WD->value[i];
x[i] = t;
for(j=A->L->ptr[i];j<A->L->ptr[i+1];j++)
{
x[A->L->index[j]] -= A->L->value[j] * t;
}
}
#endif
#ifdef USE_QUAD_PRECISION
}
else
{
#ifdef _OPENMP
nprocs = omp_get_max_threads();
#ifndef USE_SSE2
#pragma omp parallel private(i,j,jj,is,ie,w1,my_rank,p1,p2,tq,bhi,blo,chi,clo,sh,sl,th,tl,eh,el)
#else
#pragma omp parallel private(i,j,jj,is,ie,w1,my_rank,bh,ch,sh,wh,th,bl,cl,sl,wl,tl,p1,p2,t0,t1,t2,eh)
#endif
{
my_rank = omp_get_thread_num();
LIS_GET_ISIE(my_rank,nprocs,n,is,ie);
for(i=is;i<ie;i++)
{
#ifndef USE_SSE2
LIS_QUAD_MULD(w1.hi,w1.lo,x[i],xl[i],A->WD->value[i]);
#else
LIS_QUAD_MULD_SSE2(w1.hi,w1.lo,x[i],xl[i],A->WD->value[i]);
#endif
/* t = x[i] * A->WD->value[i]; */
for(j=A->U->ptr[i];j<A->U->ptr[i+1];j++)
{
jj = A->U->index[j];
if( jj<is || jj>=ie ) continue;
#ifndef USE_SSE2
LIS_QUAD_FMAD(x[jj],xl[jj],x[jj],xl[jj],w1.hi,w1.lo,-A->U->value[j]);
#else
LIS_QUAD_FMAD_SSE2(x[jj],xl[jj],x[jj],xl[jj],w1.hi,w1.lo,-A->U->value[j]);
#endif
/* x[A->U->index[j]] -= A->U->value[j] * t; */
}
}
for(i=ie-1;i>=is;i--)
{
#ifndef USE_SSE2
LIS_QUAD_MULD(w1.hi,w1.lo,x[i],xl[i],A->WD->value[i]);
#else
LIS_QUAD_MULD_SSE2(w1.hi,w1.lo,x[i],xl[i],A->WD->value[i]);
#endif
x[i] = w1.hi;
xl[i] = w1.lo;
/* t = x[i] * A->WD->value[i]; */
/* x[i] = t; */
for(j=A->L->ptr[i];j<A->L->ptr[i+1];j++)
{
jj = A->L->index[j];
if( jj<is || jj>=ie ) continue;
#ifndef USE_SSE2
LIS_QUAD_FMAD(x[jj],xl[jj],x[jj],xl[jj],w1.hi,w1.lo,-A->L->value[j]);
#else
LIS_QUAD_FMAD_SSE2(x[jj],xl[jj],x[jj],xl[jj],w1.hi,w1.lo,-A->L->value[j]);
#endif
/* x[A->L->index[j]] -= A->L->value[j] * t; */
}
}
}
#else
for(i=0;i<n;i++)
{
#ifndef USE_SSE2
LIS_QUAD_MULD(w1.hi,w1.lo,x[i],xl[i],A->WD->value[i]);
#else
LIS_QUAD_MULD_SSE2(w1.hi,w1.lo,x[i],xl[i],A->WD->value[i]);
#endif
/* t = x[i] * A->WD->value[i]; */
for(j=A->U->ptr[i];j<A->U->ptr[i+1];j++)
{
jj = A->U->index[j];
#ifndef USE_SSE2
LIS_QUAD_FMAD(x[jj],xl[jj],x[jj],xl[jj],w1.hi,w1.lo,-A->U->value[j]);
#else
LIS_QUAD_FMAD_SSE2(x[jj],xl[jj],x[jj],xl[jj],w1.hi,w1.lo,-A->U->value[j]);
#endif
/* x[A->U->index[j]] -= A->U->value[j] * t; */
}
}
for(i=n-1;i>=0;i--)
{
#ifndef USE_SSE2
LIS_QUAD_MULD(w1.hi,w1.lo,x[i],xl[i],A->WD->value[i]);
#else
LIS_QUAD_MULD_SSE2(w1.hi,w1.lo,x[i],xl[i],A->WD->value[i]);
#endif
x[i] = w1.hi;
xl[i] = w1.lo;
/* t = x[i] * A->WD->value[i]; */
/* x[i] = t; */
for(j=A->L->ptr[i];j<A->L->ptr[i+1];j++)
{
jj = A->L->index[j];
#ifndef USE_SSE2
LIS_QUAD_FMAD(x[jj],xl[jj],x[jj],xl[jj],w1.hi,w1.lo,-A->L->value[j]);
#else
LIS_QUAD_FMAD_SSE2(x[jj],xl[jj],x[jj],xl[jj],w1.hi,w1.lo,-A->L->value[j]);
#endif
/* x[A->L->index[j]] -= A->L->value[j] * t; */
}
}
#endif
}
#endif
break;
}
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
}
LIS_INT lis_matvect_ilu(LIS_MATRIX A, LIS_MATRIX_ILU LU, LIS_VECTOR X, LIS_VECTOR Y)
{
LIS_INT i,j,jj,n;
LIS_SCALAR t,*x,*y;
LIS_QUAD_DECLAR;
#ifdef USE_QUAD_PRECISION
LIS_INT j0,j1;
LIS_QUAD_PD tt;
#endif
LIS_DEBUG_FUNC_IN;
n = LU->n;
x = X->value;
y = Y->value;
#ifdef USE_QUAD_PRECISION
if( X->precision==LIS_PRECISION_DEFAULT )
#endif
{
#ifdef USE_MPI
LIS_MATVEC_SENDRECV;
#endif
#ifdef _OPENMP
#pragma omp parallel for private(i,j,jj,t)
#endif
for(i=0;i<n;i++)
{
t = 0.0;
for(j=0;j<LU->nnz[i];j++)
{
jj = LU->index[i][j];
t += LU->value[i][j] * X->value[jj];
}
Y->value[i] = t;
}
}
#ifdef USE_QUAD_PRECISION
else
{
#ifdef USE_MPI
lis_send_recv_mp(A->commtable,X);
#endif
#ifndef USE_FMA2_SSE2
#ifndef USE_SSE2
#pragma omp parallel private(i,j,jj,p1,p2,tq,bhi,blo,chi,clo,sh,sl,th,tl,eh,el)
#else
#pragma omp parallel private(i,j,jj,bh,ch,sh,wh,th,bl,cl,sl,wl,tl,p1,p2,t0,t1,t2,eh)
#endif
for(i=0;i<n;i++)
{
Y->value[i] = Y->value_lo[i] = 0.0;
for(j=0;j<LU->nnz[i];j++)
{
jj = LU->index[i][j];
#ifndef USE_SSE2
LIS_QUAD_FMAD(Y->value[i],Y->value_lo[i],Y->value[i],Y->value_lo[i],X->value[jj],X->value_lo[jj],LU->value[i][j]);
#else
LIS_QUAD_FMAD_SSE2(Y->value[i],Y->value_lo[i],Y->value[i],Y->value_lo[i],X->value[jj],X->value_lo[jj],LU->value[i][j]);
#endif
}
}
#else
#ifdef _OPENMP
#ifndef USE_SSE2
#pragma omp parallel for private(i,j,j0,j1,tt,p1,p2,tq,bhi,blo,chi,clo,sh,sl,th,tl,eh,el)
#else
#pragma omp parallel for private(i,j,j0,j1,tt,bh,ch,sh,wh,th,bl,cl,sl,wl,tl,p1,p2,t0,t1,t2,eh)
#endif
#endif
for(i=0;i<n;i++)
{
tt.hi[0] = tt.hi[1] = tt.lo[0] = tt.lo[1] = 0.0;
for(j=0;j<LU->nnz[i]-1;j+=2)
{
j0 = LU->index[i][j];
j1 = LU->index[i][j+1];
#ifdef USE_SSE2
LIS_QUAD_FMAD2_SSE2_LDSD(tt.hi[0],tt.lo[0],tt.hi[0],tt.lo[0],X->value[j0],X->value_lo[j0],X->value[j1],X->value_lo[j1],LU->value[i][j]);
#endif
}
#ifdef USE_SSE2
LIS_QUAD_ADD_SSE2(Y->value[i],Y->value_lo[i],tt.hi[0],tt.lo[0],tt.hi[1],tt.lo[1]);
#endif
for(;j<LU->nnz[i];j++)
{
j0 = LU->index[i][j];
#ifdef USE_SSE2
LIS_QUAD_FMAD_SSE2(Y->value[i],Y->value_lo[i],Y->value[i],Y->value_lo[i],X->value[j0],X->value_lo[j0],LU->value[i][j]);
#endif
}
}
#endif
}
#endif
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
}
LIS_INT lis_psolvet_ilut_csr(LIS_SOLVER solver, LIS_VECTOR B, LIS_VECTOR X)
{
#ifdef _OPENMP
LIS_INT i,j,jj,n;
LIS_INT is,ie,my_rank,nprocs;
LIS_SCALAR *b,*x;
LIS_MATRIX_ILU L,U;
LIS_VECTOR D;
LIS_PRECON precon;
LIS_QUAD_DECLAR;
#ifdef USE_QUAD_PRECISION
LIS_SCALAR *xl;
#endif
LIS_DEBUG_FUNC_IN;
precon = solver->precon;
L = precon->L;
U = precon->U;
D = precon->D;
b = B->value;
x = X->value;
#ifdef USE_QUAD_PRECISION
xl = X->value_lo;
#endif
n = solver->A->n;
nprocs = omp_get_max_threads();
#ifdef USE_QUAD_PRECISION
if( B->precision==LIS_PRECISION_DEFAULT )
{
#endif
lis_vector_copy(B,X);
#pragma omp parallel private(i,j,jj,is,ie,my_rank)
{
my_rank = omp_get_thread_num();
LIS_GET_ISIE(my_rank,nprocs,n,is,ie);
for(i=is;i<ie;i++)
{
x[i] = D->value[i]*x[i];
for(j=0;j<U->nnz[i];j++)
{
jj = U->index[i][j];
x[jj] -= U->value[i][j] * x[i];
}
}
for(i=ie-1;i>=is;i--)
{
for(j=0;j<L->nnz[i];j++)
{
jj = L->index[i][j];
x[jj] -= L->value[i][j] * x[i];
}
}
}
#ifdef USE_QUAD_PRECISION
}
else
{
lis_vector_copyex_mm(B,X);
nprocs = omp_get_max_threads();
#ifndef USE_SSE2
#pragma omp parallel private(i,j,jj,is,ie,my_rank,p1,p2,tq,bhi,blo,chi,clo,sh,sl,th,tl,eh,el)
#else
#pragma omp parallel private(i,j,jj,is,ie,my_rank,bh,ch,sh,wh,th,bl,cl,sl,wl,tl,p1,p2,t0,t1,t2,eh)
#endif
{
my_rank = omp_get_thread_num();
LIS_GET_ISIE(my_rank,nprocs,n,is,ie);
for(i=is;i<ie;i++)
{
#ifndef USE_SSE2
LIS_QUAD_MULD(x[i],xl[i],x[i],xl[i],D->value[i]);
#else
LIS_QUAD_MULD_SSE2(x[i],xl[i],x[i],xl[i],D->value[i]);
#endif
/* x[i] = D->value[i]*x[i];*/
for(j=0;j<U->nnz[i];j++)
{
jj = U->index[i][j];
#ifndef USE_SSE2
LIS_QUAD_FMAD(x[jj],xl[jj],x[jj],xl[jj],x[i],xl[i],-U->value[i][j]);
#else
LIS_QUAD_FMAD_SSE2(x[jj],xl[jj],x[jj],xl[jj],x[i],xl[i],-U->value[i][j]);
#endif
/* x[jj] -= U->value[i][j] * x[i];*/
}
}
for(i=ie-1;i>=is;i--)
{
for(j=0;j<L->nnz[i];j++)
{
jj = L->index[i][j];
#ifndef USE_SSE2
LIS_QUAD_FMAD(x[jj],xl[jj],x[jj],xl[jj],x[i],xl[i],-L->value[i][j]);
#else
LIS_QUAD_FMAD_SSE2(x[jj],xl[jj],x[jj],xl[jj],x[i],xl[i],-L->value[i][j]);
#endif
/* x[jj] -= L->value[i][j] * x[i];*/
}
}
}
}
#endif
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
#else
LIS_INT i,j,jj,n;
LIS_SCALAR *b,*x;
LIS_MATRIX_ILU L,U;
LIS_VECTOR D;
LIS_PRECON precon;
LIS_QUAD_DECLAR;
#ifdef USE_QUAD_PRECISION
LIS_SCALAR *xl;
#endif
LIS_DEBUG_FUNC_IN;
precon = solver->precon;
L = precon->L;
U = precon->U;
D = precon->D;
b = B->value;
x = X->value;
#ifdef USE_QUAD_PRECISION
xl = X->value_lo;
#endif
n = solver->A->n;
#ifdef USE_QUAD_PRECISION
if( B->precision==LIS_PRECISION_DEFAULT )
{
#endif
lis_vector_copy(B,X);
for(i=0; i<n; i++)
{
x[i] = D->value[i]*x[i];
for(j=0;j<U->nnz[i];j++)
{
jj = U->index[i][j];
x[jj] -= U->value[i][j] * x[i];
}
}
for(i=n-1; i>=0; i--)
{
for(j=0;j<L->nnz[i];j++)
{
jj = L->index[i][j];
x[jj] -= L->value[i][j] * x[i];
}
}
#ifdef USE_QUAD_PRECISION
}
else
{
lis_vector_copy(B,X);
for(i=0; i<n; i++)
{
#ifndef USE_SSE2
LIS_QUAD_MULD(x[i],xl[i],x[i],xl[i],D->value[i]);
#else
LIS_QUAD_MULD_SSE2(x[i],xl[i],x[i],xl[i],D->value[i]);
#endif
/* x[i] = D->value[i]*x[i];*/
for(j=0;j<U->nnz[i];j++)
{
jj = U->index[i][j];
#ifndef USE_SSE2
LIS_QUAD_FMAD(x[jj],xl[jj],x[jj],xl[jj],x[i],xl[i],-U->value[i][j]);
#else
LIS_QUAD_FMAD_SSE2(x[jj],xl[jj],x[jj],xl[jj],x[i],xl[i],-U->value[i][j]);
#endif
/* x[jj] -= U->value[i][j] * x[i];*/
}
}
for(i=n-1; i>=0; i--)
{
for(j=0;j<L->nnz[i];j++)
{
jj = L->index[i][j];
#ifndef USE_SSE2
LIS_QUAD_FMAD(x[jj],xl[jj],x[jj],xl[jj],x[i],xl[i],-L->value[i][j]);
#else
LIS_QUAD_FMAD_SSE2(x[jj],xl[jj],x[jj],xl[jj],x[i],xl[i],-L->value[i][j]);
#endif
/* x[jj] -= L->value[i][j] * x[i];*/
}
}
}
#endif
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
#endif
}
