Correct dependencies of trialfunction basis evaluations on their input data

Tricky... some grad components depended on the sumfac kernels of the
other directions instead of their input...

python/dune/perftool/sumfact/basis.py

@@ -68,7 +68,6 @@ def pymbolic_trialfunction_gradient(element, restriction, component, visitor):
 
     dim = world_dimension()
     buffers = []
-    insn_dep = None
     for i in range(dim):
         # Construct the matrix sequence for this sum factorization
         a_matrices = construct_amatrix_sequence(derivative=i,
@@ -90,8 +89,7 @@ def pymbolic_trialfunction_gradient(element, restriction, component, visitor):
                                 ).get_temporary(shape=shape,
                                                 name=inp,
                                                 )
-        if insn_dep is None:
-            insn_dep = frozenset({Writes(inp)})
+        insn_dep = frozenset({Writes(inp)})
 
         if get_option('fastdg'):
             # Name of direct input, shape and globalarg is set in sum_factorization_kernel
@@ -161,7 +159,6 @@ def pymbolic_trialfunction(element, restriction, component, visitor):
                                       name=inp,
                                       )
 
-    # TODO: fastdg and vectorization
     if get_option('fastdg'):
         # Name of direct input, shape and globalarg is set in sum_factorization_kernel
         direct_input = name_coefficientcontainer(restriction)