<div dir="ltr"><div dir="ltr" class="gmail_msg"><div dir="ltr" class="gmail_msg">I have found some internal test failures due to the wrong constant folding on ppc_fp128.</div><div dir="ltr" class="gmail_msg"><br class="gmail_msg"></div><div dir="ltr" class="gmail_msg">As documented in APFloat::PPCDoubleDouble, APFloat doesn't support PowerPC double-double correctly <<a href="https://github.com/llvm-mirror/llvm/blob/492acdd450bcdf9837494d6da029ed064f14fc33/lib/Support/APFloat.cpp#L74" class="gmail_msg" target="_blank">https://github.com/llvm-mirror/llvm/blob/492acdd450bcdf9837494d6da029ed064f14fc33/lib/Support/APFloat.cpp#L74</a>>.</div><div dir="ltr" class="gmail_msg"><br class="gmail_msg"></div><div class="gmail_msg">To support this, we need to add a second tuple of (sign, exponent, significand) to APFloat. I wonder where should I start to change, so that it's less hacky? I certainly expect refactoring to come. :)</div><div class="gmail_msg"><br class="gmail_msg"></div><div class="gmail_msg">Thanks!</div></div></div>