tdf#137679 Use Kahan summation for scmatrix operations

May also want implement Kahan sum in there:
sc/source/core/tool/arraysumSSE2.cxx
sc/source/core/inc/arraysumfunctor.hxx
Under some conditions the sum of a pointer type vector
may be perforemed by arraysumfunctor on NumericCellAccumulator.
arraysumSSE2 implements part of it.
This code has been left unmodified.

Why the test has been modified:
The error was: 238.89000000000001 != 238.89
|
17 th digit
IEEE 754 double-precision has 53 log10(2) ≈ 15.955 digits.
So it's just noise.

Change-Id: I6f84826bf3875be4f444f5eb61854bc1f95769bb
Reviewed-on: https://gerrit.libreoffice.org/c/core/+/114627
Reviewed-by: Mike Kaganski <mike.kaganski@collabora.com>
Tested-by: Jenkins

1 files changed, 4 insertions, 3 deletions

diff --git a/sc/inc/matrixoperators.hxx b/sc/inc/matrixoperators.hxx
index 540a38e93370..36fa1cabe3d6 100644
--- a/sc/inc/matrixoperators.hxx
+++ b/sc/inc/matrixoperators.hxx
@@ -11,6 +11,7 @@
 
 
 #include <functional>
+#include "math.hxx"
 
 namespace sc::op {
 
@@ -35,19 +36,19 @@ using Op = Op_<std::function<void(double&, double)>>;
 struct Sum
 {
     static const double InitVal;
-    void operator()(double& rAccum, double fVal) const;
+    void operator()(KahanSum& rAccum, double fVal) const;
 };
 
 struct SumSquare
 {
     static const double InitVal;
-    void operator()(double& rAccum, double fVal) const;
+    void operator()(KahanSum& rAccum, double fVal) const;
 };
 
 struct Product
 {
     static const double InitVal;
-    void operator()(double& rAccum, double fVal) const;
+    void operator()(KahanSum& rAccum, double fVal) const;
 };
 
 }