Rocket chip didn't work when using large array

Question

I am simulating baremetal executables on Rocket chip with verilator. When I use a large array like float a[3][224][224] the simulation in verilator will not work successfully.

Here is my main.cpp:

int main(int argc, char** argv)
{
  printf("CPU+cgra execute Resnet1!\n");
  long long unsigned start;
  long long unsigned end;
  float a [1][3][224][224];
  int d0 , d1, d2, d3;
  for(d0 = 0; d0 < 1; d0++){
    for(d1 = 0; d1 < 3; d1++){
      for(d2 = 0; d2 < 224; d2++){
        for(d3 = 0; d3 < 224; d3++){
          a[d0][d1][d2][d3]=d3;
        }
      } 
    }
  } 
  printf("value assign finished!\n");
    
  for(d0 = 0; d0 < 1; d0++ ){
    for(d1 = 0; d1 < 3; d1++ ){
      for(d2 = 0; d2 < 224; d2++ ){
        for(d3 = 0; d3 < 224; d3++ ){
     int I = (int)(a[d0][d1][d2][d3] * 10000);
     printf("a[%d][%d][%d][%d]:%d\n",d0,d1,d2,d3,I );
        }
      }
    }
}

Note: I compile this file with riscv-unknown-elf-gcc in chipyard/.conda-env. Even the first printf will not show.

Increasing the size of L2 cache will not work either. How can I handle this problem? Thanks.

Answer 1

Increasing L2 cache seems to work for my problem, but simulation is slow.