MMALIB Datasheet
Performance Summary

Performance data was obtained on the AM62A EVM. EVM warm cycle obtained by profiling the kernel's compute code execution after a cold run of the same code. Please refer to the kernel's documentation for more information about the parameters in the tables shown below.



CNN kernels

This section contains tables that depict the expected performance numbers for the CNN kernels.

CNN kernels

This section contains tables that depict the expected performance numbers for the CNN kernels.

Row-wise Convolution (MMALIB_CNN_convolveBias_row_ixX_ixX_oxX)

Bit Width
Stride
Fr
Fc
Ni
No
inWidth
validColsPerRowIn
validRowsIn
ValidColsIn
inChOffset
Bias
quantMethod
Pad left
Pad Right
Pad Top
Pad Bottom
EVM Cycles
8
1
3
3
378
64
128
NA
NA
512
576
Yes
No
1
1
0
0
55388
8
1
5
5
64
64
64
NA
NA
1280
2048
Yes
Yes
2
2
0
0
103360
8
1
1
1
64
128
256
NA
NA
1024
1088
Yes
Yes
0
0
0
0
9105
8
2
3
3
3
32
2048
2048
31
NA
65536
Yes
Yes
1
1
1
0
19282

Depth-wise Convolution (MMALIB_CNN_convolve_col_smallNo_ixX_ixX_oxX)

Bit Width
Kernel Width
Kernel Height
StrideX
strideY
topPad
bottomPad
leftPad
rightPad
padFillValue
Lc
Lr
numInChannels
numOutChannels
numGroupsPerKernel
EVM Cycles
8
3
3
1
1
1
1
1
1
-122
256
128
1
1
4
4848
8
5
5
1
1
2
2
2
2
-112
128
128
1
1
4
2648
8
3
3
2
2
1
1
1
1
-97
256
128
1
1
4
6726
16
3
3
1
1
1
1
1
1
-62
128
128
1
1
4
4850

Depth-wise Convolution with High Precision (MMALIB_CNN_convolve_col_smallNo_highPrecision

Fully-connected Layer (MMALIB_CNN_fullyConnectedBias_ixX_ixX_oxX)

Bit Width
Batch Size
Ni
No
Stride Y
Stride X
Stride H
EVM Warm Cycles
8
64
1280
128
128
1600
40960
10880
16
64
1280
32
128
2624
40960
10762