Open ftynse opened 6 years ago
Great, thanks!
Actually, memory promotion takes the byte-size from Halide, so it should just work. https://github.com/facebookresearch/TensorComprehensions/blob/f5e3899ddda03c378cc5ce2717e4f28eb92c70d8/src/core/polyhedral/memory_promotion_heuristic.cc#L431
yeah and the byte sizes information is also part of DLTensor
bool, int8, int16, int32, int64, float16, float64