From 307867e886f6b95b9473d5ba6598e486a93f1e99 Mon Sep 17 00:00:00 2001
From: Thomas Fischer <thomas.fischer@ufz.de>
Date: Wed, 14 Dec 2011 12:54:10 +0100
Subject: [PATCH] try to improve speedup using other OpenMP pragmas, reducing
the number of parameters of function scpr and omit a initialization of a
vector
---
MathLib/LinAlg/Solvers/CGParallel.cpp | 34 +++++++++++++--------------
MathLib/MathTools.cpp | 7 +++---
MathLib/MathTools.h | 3 +--
3 files changed, 21 insertions(+), 23 deletions(-)
diff --git a/MathLib/LinAlg/Solvers/CGParallel.cpp b/MathLib/LinAlg/Solvers/CGParallel.cpp
index 79f1cf26e71..fa2b8153c07 100644
--- a/MathLib/LinAlg/Solvers/CGParallel.cpp
+++ b/MathLib/LinAlg/Solvers/CGParallel.cpp
@@ -84,7 +84,7 @@ unsigned CGParallel(CRSMatrix<double,unsigned> const * mat, double const * const
mat->precondApply(rhat);
// rho = r * r^;
- rho = scpr(r, rhat, N, num_threads);
+ rho = scpr(r, rhat, N);
if (l > 1) {
double beta = rho / rho1;
@@ -102,29 +102,29 @@ unsigned CGParallel(CRSMatrix<double,unsigned> const * mat, double const * const
}
// q = Ap
-// blas::setzero(N, q);
- #pragma omp parallel for
- for (k = 0; k < N; k++) {
- q[k] = 0.0;
- }
mat->amux(D_ONE, p, q);
// alpha = rho / p*q
- double alpha = rho / scpr(p, q, N, num_threads);
+ double alpha = rho / scpr(p, q, N);
- // x += alpha * p
- #pragma omp parallel for
- for (k = 0; k < N; k++) {
- x[k] += alpha * p[k];
- }
+ #pragma omp parallel
+ {
+ // x += alpha * p
+ #pragma omp for nowait
+ for (k = 0; k < N; k++) {
+ x[k] += alpha * p[k];
+ }
- // r -= alpha * q
- #pragma omp parallel for
- for (k = 0; k < N; k++) {
- r[k] -= alpha * q[k];
+ // r -= alpha * q
+ #pragma omp for nowait
+ for (k = 0; k < N; k++) {
+ r[k] -= alpha * q[k];
+ }
+
+ #pragma omp barrier
}
- resid = sqrt(scpr(r, r, N, num_threads));
+ resid = sqrt(scpr(r, r, N));
if (resid <= eps * nrmb) {
eps = resid / nrmb;
diff --git a/MathLib/MathTools.cpp b/MathLib/MathTools.cpp
index ccad1686733..35bd2612dc1 100644
--- a/MathLib/MathTools.cpp
+++ b/MathLib/MathTools.cpp
@@ -10,15 +10,14 @@
namespace MathLib {
#ifdef _OPENMP
-double scpr(double const * const v, double const * const w, unsigned n,
- unsigned num_of_threads)
+double scpr(double const * const v, double const * const w, unsigned n)
{
- long double res (0.0);
+ long double res (v[0]*w[0]);
unsigned k;
#pragma omp parallel
{
#pragma omp parallel for reduction (+:res)
- for (k = 0; k<n; k++) {
+ for (k = 1; k<n; k++) {
res += v[k] * w[k];
}
}
diff --git a/MathLib/MathTools.h b/MathLib/MathTools.h
index 1d88785af6d..4a9d2b804f9 100644
--- a/MathLib/MathTools.h
+++ b/MathLib/MathTools.h
@@ -35,8 +35,7 @@ double scpr(const T* v0, const T* v1, size_t n)
}
#ifdef _OPENMP
-double scpr(double const * const v, double const * const w, unsigned n,
- unsigned num_of_threads);
+double scpr(double const * const v, double const * const w, unsigned n);
#endif
/**
--
GitLab