diff options
Diffstat (limited to 'cachelab-handout/trans.c')
| -rw-r--r-- | cachelab-handout/trans.c | 34 |
1 files changed, 16 insertions, 18 deletions
diff --git a/cachelab-handout/trans.c b/cachelab-handout/trans.c index 2a46f92..833e44f 100644 --- a/cachelab-handout/trans.c +++ b/cachelab-handout/trans.c @@ -10,6 +10,8 @@ #include <stdio.h> #include "cachelab.h" +#include "trans.h" + int is_transpose(int M, int N, int A[N][M], int B[M][N]); /* @@ -116,25 +118,21 @@ void transpose_submit(int M, int N, int A[N][M], int B[M][N]) // solve diagonal issue if (i == j) { - for (j2 = 0; j2 < CASE_64_64_MAGIC_NUMBER * CASE_64_64_BLOCK_SIZE; j2 += CASE_64_64_BLOCK_SIZE) - { - for (t4 = 0; t4 < CASE_64_64_MAGIC_NUMBER * CASE_64_64_BLOCK_SIZE; t4 += CASE_64_64_BLOCK_SIZE) - { - // move 4x4 - for (i2 = 0; i2 < CASE_64_64_BLOCK_SIZE; i2++) - { - t0 = A[i * CASE_64_64_MAGIC_NUMBER * CASE_64_64_BLOCK_SIZE + j2 + i2][j * CASE_64_64_MAGIC_NUMBER * CASE_64_64_BLOCK_SIZE + t4 + 0]; - t1 = A[i * CASE_64_64_MAGIC_NUMBER * CASE_64_64_BLOCK_SIZE + j2 + i2][j * CASE_64_64_MAGIC_NUMBER * CASE_64_64_BLOCK_SIZE + t4 + 1]; - t2 = A[i * CASE_64_64_MAGIC_NUMBER * CASE_64_64_BLOCK_SIZE + j2 + i2][j * CASE_64_64_MAGIC_NUMBER * CASE_64_64_BLOCK_SIZE + t4 + 2]; - t3 = A[i * CASE_64_64_MAGIC_NUMBER * CASE_64_64_BLOCK_SIZE + j2 + i2][j * CASE_64_64_MAGIC_NUMBER * CASE_64_64_BLOCK_SIZE + t4 + 3]; + // A00,01 -> B00,01 + CP_4_BY_8(i*8 + 0*4, j*8 + 0*4, i2); + B_SMALL_TRANSPOSE(i*8 + 0*4, j*8 + 0*4, i2); + B_SMALL_TRANSPOSE(i*8 + 0*4, j*8 + 1*4, i2); + + // A10,11 -> B10,11 + CP_4_BY_8(i*8 + 1*4, j*8 + 0*4, i2); + B_SMALL_TRANSPOSE(i*8 + 1*4, j*8 + 0*4, i2); + B_SMALL_TRANSPOSE(i*8 + 1*4, j*8 + 1*4, i2); + + // swap B01, B10 + // B10 is cached + B_SWAP_4_BY_4(i*8 + 1*4, j*8 + 0*4, + i*8 + 0*4, j*8 + 1*4, i2); - B[j * CASE_64_64_MAGIC_NUMBER * CASE_64_64_BLOCK_SIZE + t4 + 0][i * CASE_64_64_MAGIC_NUMBER * CASE_64_64_BLOCK_SIZE + j2 + i2] = t0; - B[j * CASE_64_64_MAGIC_NUMBER * CASE_64_64_BLOCK_SIZE + t4 + 1][i * CASE_64_64_MAGIC_NUMBER * CASE_64_64_BLOCK_SIZE + j2 + i2] = t1; - B[j * CASE_64_64_MAGIC_NUMBER * CASE_64_64_BLOCK_SIZE + t4 + 2][i * CASE_64_64_MAGIC_NUMBER * CASE_64_64_BLOCK_SIZE + j2 + i2] = t2; - B[j * CASE_64_64_MAGIC_NUMBER * CASE_64_64_BLOCK_SIZE + t4 + 3][i * CASE_64_64_MAGIC_NUMBER * CASE_64_64_BLOCK_SIZE + j2 + i2] = t3; - } - } - } continue; } |