4:37:39no-defun-allowedI'm going to have another go at testing that optimisation, but bailing out if the bignum argument could fit in (signed-byte 64) or (unsigned-byte 64); which is probably the "stupid" option.
4:59:32no-defun-allowedStill no change with my interpreter; though now that transform only fires on the comparisons to 2^256 and -2^256.
5:24:12no-defun-allowedIf I use that transform after SBCL has booted but before the interpreter function is compiled, it doesn't slow down the interpreter though.
5:41:57no-defun-allowedBut all the times it fires, in SBCL, libraries and my interpreter, it should be faster, and the bignum is bigger than 64 bits.