evanw / float-toy Goto Github PK

@evanw
This is an awesome project and I love it!
But lack of some new data format, such as BF16, TF32. Could you add supports of them?

0 11111 0000000000 in 16 bit float should read 65536. This web app claims it is infinity.

Could you please add this peculiar format https://en.wikipedia.org/wiki/Unum_(number_format)

Text description sounds too vague for me)

If you set the 64-bit exponent to all zeros and toggle the least significant (rightmost) bit it doesn’t update the exponent shown below. This is on Firefox 72.0.1.

M gonna boot the tool:

BOOM

https://en.wikipedia.org/wiki/Half-precision_floating-point_format

The accuracy of the values goes downhill from bit 16 onward, with incorrect results.

I.e. using the 32-bit (float) toy:

1. Set only bit 17 'on': 0x3F810000 : value is 1.0078125: correct!
2. Set only bit 16 'on': 0x3F808000 : value is 1.0039062: incorrect! [Should be 1.00390625]
3. Set only bit 15 'on': 0x3F804000 : value is 1.0019531: incorrect! [Should be 1.001953125]
4. Set only bit 14 'on': 0x3F802000 : value is 1.0009766: incorrect! [Should be 1.0009765625]

The inaccuracies accumulate.

My specific egg-on-face incident was "1.28": the toy claims that 0x3FA3D70A correctly represents "1.28", but any C compiler will tell you the value is 1.27999973297.

[Or Google:
1+(2^-2)+(2^-6)+(2^-7)+(2^-8)+(2^-9)+(2^-11)+(2^-13)+(2^-14)+(2^-15)+(2^-20)]

https://en.wikipedia.org/wiki/Signed_zero

When the sign-bit is true, and all other bits are false, the result should be negative zero.