Single precision

Article on other languages:

• Einfache_Genauigkeit
• 単精度
• Число_одинарной_точности
• Једнострука_тачност

	del.icio.us
	Digg
	Furl
	Reddit
	Rojo
	Add to OnlyWire

In computing, single precision is a computer numbering format that occupies one storage location in computer memory at a given address. A single-precision number, sometimes simply a single, may be defined to be an integer, fixed point, or floating point.

Modern computers with 32-bit words (single precision) provide 64-bit double precision. Single precision floating point is an IEEE 754 standard for encoding floating point numbers that uses 4 bytes.

Single precision memory format

 Sign bit: 1
 Exponent width: 8  
 Significant precision: 23 (24 implicit)

The format is written with an implicit most-significant bit with value 1 unless the written exponent is all zeros. Thus only 23 bits of the fraction mantissa appear in the memory format but the total precision is 24 bits (better than 7 decimal digits, ).

Exponent encoding

 Emin (0x01) = -126
 Emax (0xfe) = 127
 Exponent bias (0x7f) = 127

The true exponent = written exponent - exponent bias

 0x00 and 0xff  are reserved exponents 
 0x00 is used to represent zero and denormals
 0xff is used to represent infinity and NaNs

All bit patterns are valid encoding.

Single precision examples in hexadecimal

 3f80 0000   = 1

 c000 0000   = -2

 7f7f ffff   ~ 3.4028234 x 1038  (Max Single)

 3eaa aaab   ~ 1/3

By default, 1/3 rounds up instead of down like double precision, because of the even number of bits in the significant. So the bits beyond the rounding point are 1010... which is more than 1/2 of a unit in the last place.

 0000 0000   = 0
 8000 0000   = -0

 7f80 0000   = Infinity
 ff80 0000   = -Infinity

Converting from single precision to human readable form

We start with the hexadecimal representation of the value, 41c80000, in this example, and convert it to binary

 41c8 000016 = 0100 0001 1100 1000 0000 0000 0000 00002

then we break it down into three parts; sign bit, exponent and mantissa.

 Sign bit: 0
 Exponent: 1000 00112 = 8316 = 131
 Mantissa: 100 1000 0000 0000 0000 00002 = 48000016

We then add the implicit 24th bit to the mantissa

 Mantissa: 1100 1000 0000 0000 0000 00002 = C8000016

and decode the exponent value by subtracting 127

 Raw exponent: 8316 = 131
 Decoded exponent: 131 - 127 = 4

Each of the 24 bits of the mantissa, bit 23 to bit 0, represents a value, starting at 1 and halves for each bit, as follows

 bit 23 = 1
 bit 22 = 0.5
 bit 21 = 0.25
 bit 20 = 0.125
 bit 19 = 0.0625
 .
 .

The mantissa in this example has three bits set, bit 23, bit 22 and bit 19. We can now decode the mantissa by adding the values represented by these bits.

 Decoded mantissa: 1 + 0.5 + 0.0625 = 1.5625

Then we need to multiply with the base, 2, to the power of the exponent to get the final result

 1.5625 * 24 = 25

Thus

 41c8 0000   = 25

See also

• half precision – single precision – double precision – quadruple precision
• Floating point