equations for feature map size resulting from convolution and pooling layers

[Stuart Kerr]

Referencing both:
Stanford's cs231 class and Caffe's Layer's Docs
they note that the size of a feature map resulting from a convolution is:
(W - F +2P)/S  + 1
where W = incoming feature map spatial dimension, F = convolution filter dimension, P = padding and S = stride. However, when training a standard AlexNet where F = 11 and S = 4 on 64x64 image patches, Caffe does not throw an error, and we train successfully. Similar calculations can be done for feature maps resulting from pooling layers.

Please explain this discrepancy...

[Stuart Kerr]

Forgot to mention that caffe does not throw an error, despite the quotient is not an integer...