Debugging N_BLOCKS

[WilsonCWu]

Why is the first one failing?

/matmul
N=16, N_BLOCK=2
--------------------  M=4096 N=4096 K=4096  --------------------
Block size: 128x32
Allocated host memory
Initialized matrices
Performed CPU matrix multiplication
Allocated device memory
Copied matrices to device
Launching warmup kernel with grid (132, 1), block (384)
Launching kernel with grid (132, 1), block (384)
Avg Kernel execution time: 742.272 us
Achieved performance: 185.16 TFLOPs
Copied result back to host
Converted result back to float
Error at row 0 col 0: -3.09375 != -1.39006 (ref)
Error at row 0 col 1: -2.82812 != 1.66512 (ref)
Error at row 0 col 2: 8.3125 != 2.44395 (ref)
Error at row 0 col 3: -6.21875 != -8.22833 (ref)
Error at row 0 col 4: -3.35938 != 2.22921 (ref)
Error at row 0 col 5: 0.660156 != 6.12578 (ref)
Error at row 0 col 6: 18.875 != -7.62248 (ref)
Error at row 0 col 7: 7.25 != -1.25781 (ref)
Error at row 0 col 8: 1.91406 != 3.26331 (ref)
Error at row 0 col 9: 0.460938 != 3.74322 (ref)
Error at row 0 col 10: 2.375 != -1.60718 (ref)
Error at row 0 col 11: 0.691406 != 3.00533 (ref)
Error at row 0 col 12: -3.89062 != 6.23455 (ref)
Error at row 0 col 13: -4.125 != 2.55398 (ref)
Error at row 0 col 14: 1.05469 != 6.0381 (ref)
Error at row 0 col 15: -0.236328 != 4.82473 (ref)
Error at row 0 col 17: 5.15625 != -4.56702 (ref)
Error at row 0 col 18: -5.625 != 1.13531 (ref)
Error at row 0 col 19: -1.19531 != 7.92724 (ref)
Error at row 0 col 20: 5.96875 != 0.462465 (ref)
Too many errors to show them all.
Max error: 40.0543
Error count: 15008549
N=16, N_BLOCK=1
--------------------  M=4096 N=4096 K=4096  --------------------
Block size: 128x16
Allocated host memory
Initialized matrices
Performed CPU matrix multiplication
Allocated device memory
Copied matrices to device
Launching warmup kernel with grid (132, 1), block (384)
Launching kernel with grid (132, 1), block (384)
Avg Kernel execution time: 1172.43 us
Achieved performance: 117.226 TFLOPs
Copied result back to host
Converted result back to float
Max error: 0.0982647
Error count: 0
N=32, N_BLOCK=1
--------------------  M=4096 N=4096 K=4096  --------------------
Block size: 128x32
Allocated host memory
Initialized matrices
Performed CPU matrix multiplication
Allocated device memory
Copied matrices to device
Launching warmup kernel with grid (132, 1), block (384)
Launching kernel with grid (132, 1), block (384)
Avg Kernel execution time: 633.43 us
Achieved performance: 216.976 TFLOPs
Copied result back to host
Converted result back to float
Max error: 0.0982647
Error count: 0
N=64, N_BLOCK=2
--------------------  M=4096 N=4096 K=4096  --------------------
Block size: 128x128
Allocated host memory
Initialized matrices
Performed CPU matrix multiplication
Allocated device memory
Copied matrices to device
Launching warmup kernel with grid (132, 1), block (384)
Launching kernel with grid (132, 1), block (384)
Avg Kernel execution time: 229.372 us
Achieved performance: 599.198 TFLOPs
Copied result back to host
Converted result back to float
Max error: 0.0982647
Error count: 0