Mapping failed on nested loops

[HobbitQia]

Related to PR129.

For a simple nested loop with two layers, the mapping pass cannot find a valid mapping and will output [DEBUG] mapping failed for II = 10.

An example:

// RUN: clang++ -S -emit-llvm kernel.cpp -o kernel.ll
// RUN: mlir-translate --import-llvm kernel.ll -o kernel.mlir
// RUN: mlir-neura-opt --assign-accelerator \
// RUN:   --lower-llvm-to-neura \
// RUN:   --canonicalize-live-in \
// RUN:   --leverage-predicated-value \
// RUN:   --transform-ctrl-to-data-flow \
// RUN:   --fold-constant \
// RUN:   --insert-data-mov \
// RUN:   --map-to-accelerator="mapping-strategy=heuristic backtrack-config=simple" --generate-code kernel.mlir | FileCheck %s --check-prefix=CHECK-LLVM2NEURA-MAP


#include <stdio.h>

#define NTAPS 32

int input[NTAPS];
int output[NTAPS];
int coefficients[NTAPS];

void kernel(int input[], int output[], int coefficient[]);

int main()
{
  kernel(input, output, coefficients);

  return 0;
}

void kernel(int input[], int output[], int coefficient[]) {
  int i, j;

   for (i = 0; i < NTAPS; ++i) {
        for (j = 0; j < NTAPS; ++j) {
            
        }
    }
}

I tried to increase the max_location_to_try from 10 to 500, max_backtrack_depth from 3 to 20, maxII from 10 to 100 but it can not work still.