4 1. Check for denormal and pseudodenormal operands in ALL instructions
5 I hope all instructions return the same values as real CPU.
7 2. Unmasked underflow/overflow should correct the result by magic number
8 for all operations, including float32 and float64.
10 3. Elliminate floa128 use, Intel uses only 67-bit precision calculations
11 when float128 has at least 112-bit. Replacement of float128 with for
12 example 96-bit precision number could significantly speed up