Intel ITANIUM ARCHITECTURE - SOFTWARE DEVELOPERS MANUAL VOLUME 1 REV 2.3 Manual page 1846

RCPPS: Packed Single-FP Reciprocal (Continued)
For input values x which satisfy
1.11111111110100000000001
1.00000000000110000000000
flush-to-zero might or might not occur, depending on the implementation (this interval
contains 6144 + 3072 = 9216 single precision floating-point numbers).
Results are guaranteed to be tiny, and therefore flushed to zero, for input values x
which satisfy
|x| <= 1.00000000000110000000001
The decimal approximations of the single precision numbers that delimit the three
intervals specified above, are as follows:
1.11111111110100000000000
1.11111111110100000000001
1.00000000000110000000000
1.00000000000110000000001
The hexadecimal representations of the single precision numbers that delimit the three
intervals specified above, are as follows:
1.11111111110100000000000
1.11111111110100000000001
1.00000000000110000000000
1.00000000000110000000001
4:544
125
×2 <= |x| <=
B
126
×2
B
126
×2
B
125
×2 ~= 8.5039437×10
B
125
×2 ~= 8.5039443×10
B
126
×2 ~= 4.2550872×10
B
126
×2 ~= 4.2550877×10
B
125
×2 = 0x7e7fe800
B
125
×2 = 0x7e7fe801
B
126
×2 = 0x7e800c00
B
126
×2 = 0x7e800c01
B
37
37
37
37
Volume 4: IA-32 SSE Instruction Reference