Segmentation fault for kernel with workspace

[mjschleich]

This following kernel results in a segmentation fault on master:

#include "taco.h"
using namespace taco;

int main(int argc, char* argv[]) {
  Format csr({Dense,Sparse});
  Format sv({Sparse});
  Format dv({Dense});

  int dimension = 1000;
  int sparsity_factor = 3;

  Tensor<double> y("y", {dimension}, sv);
  Tensor<double> A("A", {dimension,dimension}, csr);
  Tensor<double> x("x", {dimension},  dv);
  Tensor<double> beta("b");

  for (int i=0; i<dimension; ++i) {
    for (int j=0; j<dimension; ++j) {
      if ((i + j) % (1 << sparsity_factor) == 0){
        A(i,j) = double(i * j + j);
      }
    }
  }
  for (int i=0; i<dimension; ++i) {
      x(i) = double(i);
  }
  beta() = 1.2;

  A.pack();
  x.pack();

  IndexVar i, j, k;

  IndexExpr precomputedExpr = A(i,k) * x(k);
  y(i) =  beta() * A(i,j) * precomputedExpr;

  IndexStmt stmt = y.getAssignment().concretize();

  // Add workspace for A(i,k) * x(k)
  TensorVar workspace("workspace", Type(Float64, {Dimension(dimension)}), taco::dense);
  stmt = stmt.precompute(precomputedExpr, {i}, {i}, workspace);
  stmt = stmt.concretize();
 
  y.compile(stmt);
  y.assemble();
  y.compute();
}

It works fine if I remove the workspace, also when I use the workspace but remove beta (a constant) from the expression.