MathUtils: materialise the product in SMatrixGPU's operator*=

Unlike ROOT's SMatrix, where the product of two matrices is a matrix, SMatrixGPU
returns a lazy expression, and every element of that expression reads the whole
of the left operand. Assigning it back element by element therefore reads values
that have already been overwritten. It also did not compile: the expression
matched the generic operator= that copies the representation, which an
expression does not have.

This was unreachable until now, since nothing instantiated a matrix multiply
assignment in device code.

Author: ktf

Common/MathUtils/include/MathUtils/SMatrixGPU.h

@@ -1433,14 +1433,18 @@ template <class T, unsigned int D1, unsigned int D2, class R>
 template <class R2>
 GPUdi() SMatrixGPU<T, D1, D2, R>& SMatrixGPU<T, D1, D2, R>::operator*=(const SMatrixGPU<T, D1, D2, R2>& rhs)
 {
-  return operator=(*this* rhs);
+  // the product is an expression evaluated element by element, and every element
+  // of it reads the whole of *this, so it has to be materialised first
+  const SMatrixGPU<T, D1, D2, R> tmp(*this * rhs);
+  return operator=(tmp);
 }
 
 template <class T, unsigned int D1, unsigned int D2, class R>
 template <class A, class R2>
 GPUdi() SMatrixGPU<T, D1, D2, R>& SMatrixGPU<T, D1, D2, R>::operator*=(const Expr<A, T, D1, D2, R2>& rhs)
 {
-  return operator=(*this* rhs);
+  const SMatrixGPU<T, D1, D2, R> tmp(*this * rhs);
+  return operator=(tmp);
 }
 
 template <class T, unsigned int D1, unsigned int D2, class R>