When analyzing scalable expressions, the analyzer will iterate over a series of known vscale values in the range 1-16. However, we can tighten this range to only values that are a power of two, as stated in the LLVM lang ref and more generally the reference manual.
When analyzing scalable expressions, the analyzer will iterate over a series of known vscale values in the range 1-16. However, we can tighten this range to only values that are a power of two, as stated in the LLVM lang ref and more generally the reference manual.
This comes from a discussion in https://github.com/apache/tvm/pull/16921#discussion_r1600048788