Smaller_knl_with_prob.md

Making the Kernel

knl = lp.make_kernel(
        "{ [i, j, k, k1]: 0<=i, j, k, k1<256 }",
        """
        temp_cnst[k] = 2.0 {id=insn_1}
        temp_cnst_2[k1] = 2*temp_cnst[k1] {id=insn_4}
        c[i, j] = reduce(sum, k, a[i,k]*b[k,j]*temp_cnst_2[k]) {id=insn_2}
        c[i, j] = reduce(sum, k1, a[i,k1]*b[k1,j]*temp_cnst_2[k1]) {id=insn_3}
        """)

Code to resolve it

processed_knl = lp.preprocess.preprocess_kernel(knl)
fixed_knl = processed_knl.copy()
while not lp.has_schedulable_iname_nesting(fixed_knl):
    iname, insn = next(lp.get_iname_duplication_options(fixed_knl))
    fixed_knl = lp.duplicate_inames(fixed_knl, iname, insn)
print('After all this duplication business, we have'
      ' has_schedulable_iname_nesting =',
      lp.has_schedulable_iname_nesting(fixed_knl))
knl = fixed_knl

Error

I am receiving the error message: ERROR: Sorry--loo.py did not find a schedule for your kernel..

Slight difference between this(i.e. smaller example kernel) and the mixed problem kernel

In the case of the smaller kernel the lp.has_schedulable_iname_nesting directly returns True hence it does not enter the while loop. But in mixed kernel case, after doing some duplications it succesfully enters and exits the while loop and still gives out the same scheduling error and performing some iname_duplication transformations.