Which document does this relate to?
Linalg
Describe the issue you see with the spec
FP8 data type is required to be supported for input when vector-matrix operation, does the matrix layout must be optimal for fp8? is it possible row or column and we need load fp8 matrix data from memory directly? or the matrix data in memory must be native type, like f32, f16?
Another question, for vector convert function linalg::Convert, does this convert used by vector-matrix operation only? is it possible used by other operations?
Which document does this relate to?
Linalg
Describe the issue you see with the spec
FP8 data type is required to be supported for input when vector-matrix operation, does the matrix layout must be optimal for fp8? is it possible row or column and we need load fp8 matrix data from memory directly? or the matrix data in memory must be native type, like f32, f16?
Another question, for vector convert function linalg::Convert, does this convert used by vector-matrix operation only? is it possible used by other operations?