maeri-project / mRNA

mRNA
https://synergy.ece.gatech.edu/tools/mrna
MIT License
22 stars 9 forks source link

mapping candidates that generate partial sums are used #2

Open francisco-munoz opened 5 years ago

francisco-munoz commented 5 years ago

Hello,

I am experimenting with mRNA and I have noticed an issue I do not understand.

It looks as if the configurations proposed by the tool generate partial sums. However, in the paper, you specify that in order to reduce the search space you prune all the possible mapping that generate partial sum.

Here it is an example. When I execute the command with 512 Ms and the first layer of AlexNet, I obtain this MAERI configuration file:

MAERI Configuration: Layer variables: X = 224 Y = 224 C = 3 K = 96 N = 32 X' = 56 Y' = 56 R = 11 S = 11 Mapping kernel (tile): T_X = 11 T_Y = 11 T_C = 1 T_K = 4 T_N = 1 T_X' = 1 T_Y' = 1 Virtual Neuron : VN_Size = 121 Num_VN = 4 Loops outside the tile: R/T_X = 1 S/T_Y = 1 C/T_C = 3 C/T_K = 24 C/T_N = 32 C/T_OX = 56 C/T_OY = 56 R%T_X = 0 S%T_Y = 0 C%T_C = 0 K%T_K = 0 N%T_N = 0 X'%T_OX = 0 Y'%T_OY = 0 Outer loop order (from outermost to innermost): N->C->K->X'->Y' Total_iteration = 7225344

As you can see, the number of channels in the layer is 3 and at the same time the tile size is 1. So, in the principal configuration, it is generating partial sums and in a second configuration, it finishes the computation. Can you say then that partial sums are allowed?