[ExpandLargeFpConvert] Fix bug in int-to-fp expansion. (llvm#85370)

When deciding whether to perform rounding on the significand,
the generated IR was using (width - leading zeros - 1) rather
than (width - leading zeros). This is different from how the
routine in compiler-rt does it:

    int sd = srcBits - clzSrcT(a);
    int e = sd - 1;
    if (sd > dstMantDig) {

This bug means that the following code, when built on -O0:

    #include <stdio.h>

    _BitInt(233) v_1037 = 0;

    int main(void)
    {
        v_1037 = 18014398509481982wb;
        double d = v_1037;
        printf("d = %f\n", d);

        return 0;
    }

prints "d = 9007199254740992.000000", which is incorrect.
The correct result is "d = 18014398509481982.000000".

(cherry picked from commit f623adb)
Change-Id: Ieb9c0c786a1b781200f40e6229e69231c8fabe15

Author: bevin-hansson

llvm/lib/CodeGen/ExpandLargeFpConvert.cpp

@@ -375,7 +375,7 @@ static void expandIToFP(Instruction *IToFP) {
   Value *Sub2 = Builder.CreateSub(Builder.getIntN(BitWidthNew, BitWidth - 1),
                                   FloatWidth == 128 ? Call : Cast);
   Value *Cmp3 = Builder.CreateICmpSGT(
-      Sub2, Builder.getIntN(BitWidthNew, FPMantissaWidth + 1));
+      Sub1, Builder.getIntN(BitWidthNew, FPMantissaWidth + 1));
   Builder.CreateCondBr(Cmp3, IfThen4, IfElse);
 
   // if.then4: