/* 
 * trans.c - Matrix transpose B = A^T
 *
 * Each transpose function must have a prototype of the form:
 * void trans(int M, int N, int A[N][M], int B[M][N]);
 *
 * A transpose function is evaluated by counting the number of misses
 * on a 1KB direct mapped cache with a block size of 32 bytes.
 */ 
#include <stdio.h>
#include "cachelab.h"

int is_transpose(int M, int N, int A[N][M], int B[M][N]);

/* 
 * transpose_submit - This is the solution transpose function that you
 *     will be graded on for Part B of the assignment. Do not change
 *     the description string "Transpose submission", as the driver
 *     searches for that string to identify the transpose function to
 *     be graded. 
 */
char transpose_submit_desc[] = "Transpose submission";
void transpose_sub(int N, int t, int l, int size, int B[][N]) {
  for (int i = 0; i < size; i++) {
    for (int j = i + 1; j < size; j++) {
      B[t + i][l + j] ^= B[t + j][l + i];
      B[t + j][l + i] ^= B[t + i][l + j];
      B[t + i][l + j] ^= B[t + j][l + i];
    }
  }
}
void transpose_submit(int M, int N, int A[N][M], int B[M][N]) {
  int v0, v1, v2, v3, v4, v5, v6, v7;
  if (M == 32) {
    for (int j = 0; j < M; j += 8) {
      for (int i = 0; i < N; i++) {
        v0 = A[i][j];
        v1 = A[i][j + 1];
        v2 = A[i][j + 2];
        v3 = A[i][j + 3];
        v4 = A[i][j + 4];
        v5 = A[i][j + 5];
        v6 = A[i][j + 6];
        v7 = A[i][j + 7];

        B[j][i] = v0;
        B[j + 1][i] = v1;
        B[j + 2][i] = v2;
        B[j + 3][i] = v3;
        B[j + 4][i] = v4;
        B[j + 5][i] = v5;
        B[j + 6][i] = v6;
        B[j + 7][i] = v7;
      }
    }
  } else if (M == 64) {
    for (int i = 0; i < N; i += 8) {
      for (int j = 0; j < M; j += 8) {
        if (i == j) {
          for (int ii = i; ii < i + 4; ii++) {
            v0 = A[ii][j];
            v1 = A[ii][j + 1];
            v2 = A[ii][j + 2];
            v3 = A[ii][j + 3];
            v4 = A[ii][j + 4];
            v5 = A[ii][j + 5];
            v6 = A[ii][j + 6];
            v7 = A[ii][j + 7];

            B[ii][j] = v0;
            B[ii][j + 1] = v1;
            B[ii][j + 2] = v2;
            B[ii][j + 3] = v3;
            B[ii][j + 4] = v4;
            B[ii][j + 5] = v5;
            B[ii][j + 6] = v6;
            B[ii][j + 7] = v7;
          }
          transpose_sub(N, i, j, 4, B);
          transpose_sub(N, i, j + 4, 4, B);

          for (int ii = i + 4; ii < i + 8; ii++) {
            v0 = A[ii][j];
            v1 = A[ii][j + 1];
            v2 = A[ii][j + 2];
            v3 = A[ii][j + 3];
            v4 = A[ii][j + 4];
            v5 = A[ii][j + 5];
            v6 = A[ii][j + 6];
            v7 = A[ii][j + 7];

            B[ii][j] = v0;
            B[ii][j + 1] = v1;
            B[ii][j + 2] = v2;
            B[ii][j + 3] = v3;
            B[ii][j + 4] = v4;
            B[ii][j + 5] = v5;
            B[ii][j + 6] = v6;
            B[ii][j + 7] = v7;
          }
          transpose_sub(N, i + 4, j, 4, B);
          transpose_sub(N, i + 4, j + 4, 4, B);

          for (int ii = i; ii < i + 4; ii++) {
            v0 = B[ii + 4][j];
            v1 = B[ii + 4][j + 1];
            v2 = B[ii + 4][j + 2];
            v3 = B[ii + 4][j + 3];

            v4 = B[ii][j + 4];
            v5 = B[ii][j + 5];
            v6 = B[ii][j + 6];
            v7 = B[ii][j + 7];

            B[ii][j + 4] = v0;
            B[ii][j + 5] = v1;
            B[ii][j + 6] = v2;
            B[ii][j + 7] = v3;

            B[ii + 4][j] = v4;
            B[ii + 4][j + 1] = v5;
            B[ii + 4][j + 2] = v6;
            B[ii + 4][j + 3] = v7;
          }
        } else {
          for (int ii = i; ii < i + 4; ii++) {
            for (int jj = j; jj < j + 4; jj++) {
              B[jj][ii] = A[ii][jj];
            }
            for (int jj = j + 4; jj < j + 8; jj++) {
              B[jj - 4][ii + 4] = A[ii][jj];
            }
          }

          for (int jj = j; jj < j + 4; jj++) {
            v0 = B[jj][i + 4];
            v1 = B[jj][i + 5];
            v2 = B[jj][i + 6];
            v3 = B[jj][i + 7];

            B[jj][i + 4] = A[i + 4][jj];
            B[jj][i + 5] = A[i + 5][jj];
            B[jj][i + 6] = A[i + 6][jj];
            B[jj][i + 7] = A[i + 7][jj];

            B[jj + 4][i] = v0;
            B[jj + 4][i + 1] = v1;
            B[jj + 4][i + 2] = v2;
            B[jj + 4][i + 3] = v3;
          }

          for (int ii = i + 4; ii < i + 8; ii++) {
            for (int jj = j + 4; jj < j + 8; jj++) {
              B[jj][ii] = A[ii][jj];
            }
          }
        }
      }
    }
  } else if (M == 61) {
    for (int j = 0; j < M; j += 17) {
      for (int i = 0; i < N; i++) {
        for (int jj = j; jj < j + 17 && jj < M; jj++) {
          B[jj][i] = A[i][jj];
        }
      }
    }
  }
}

/* 
 * You can define additional transpose functions below. We've defined
 * a simple one below to help you get started. 
 */ 

/* 
 * trans - A simple baseline transpose function, not optimized for the cache.
 */
char trans_desc[] = "Simple row-wise scan transpose";
void trans(int M, int N, int A[N][M], int B[M][N])
{
    int i, j, tmp;

    for (i = 0; i < N; i++) {
        for (j = 0; j < M; j++) {
            tmp = A[i][j];
            B[j][i] = tmp;
        }
    }    

}

/*
 * registerFunctions - This function registers your transpose
 *     functions with the driver.  At runtime, the driver will
 *     evaluate each of the registered functions and summarize their
 *     performance. This is a handy way to experiment with different
 *     transpose strategies.
 */
void registerFunctions()
{
    /* Register your solution function */
    registerTransFunction(transpose_submit, transpose_submit_desc); 

    /* Register any additional transpose functions */
    registerTransFunction(trans, trans_desc); 

}

/* 
 * is_transpose - This helper function checks if B is the transpose of
 *     A. You can check the correctness of your transpose by calling
 *     it before returning from the transpose function.
 */
int is_transpose(int M, int N, int A[N][M], int B[M][N])
{
    int i, j;

    for (i = 0; i < N; i++) {
        for (j = 0; j < M; ++j) {
            if (A[i][j] != B[j][i]) {
                return 0;
            }
        }
    }
    return 1;
}