Commit e183b8d0 authored by Eugene Zhulenev's avatar Eugene Zhulenev Committed by TensorFlower Gardener
Browse files

Custom Conv3DBackprop Input/Filter kernels. 

~2x-3x speedup when compiled with AVX over the Eigen kernels, at the cost of memory overhead (needs to allocate temp buffers).

Memory overhead is constrained. When memory requirements grow too far, fallback on Eigen implementation.

PiperOrigin-RevId: 212734097
parent 35e17c01
Show whitespace changes
Inline Side-by-side
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Please to comment