Commit fbd89bc6 authored by Shucai Xiao's avatar Shucai Xiao
Browse files

backup dot operator changes.

parent 4ffaf3c7
...@@ -807,6 +807,17 @@ struct gather ...@@ -807,6 +807,17 @@ struct gather
} }
}; };
// The dot operation is combination of the onnx GEMM and MatMul operators.
// For GEMM, it support the C matrix in the formula alpha * AB + beta * C,
// in which C is broadcastable to the shape of AB. For the transpose of A
// and B, we add a tranpose operator if the onnx file needs.
// For MatMul, it has the same definition as the numpy.matmul, which means
// A, B could be 1 to N-dims. For 1-dim input of A, it is a vector * matrix,
// for 1-dim of B, it is a matrix * vector. Note that there is not support
// of batch of 1-dim vector. In another word, if A or B is 1-dim, it is a
// vector input; if A or B is 2-dim, it is a matrix (no case of a batch of
// vectors as input). If A or B is 3 or more dims, it is considered as a
// stack(batch) of matrices.
struct dot struct dot
{ {
float alpha = 1.0; float alpha = 1.0;
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment