Current implementation is multiplying Float32 number in range (-1,1) by 0xFFFF which is eqvivalent to 65535. I found that with this conversion data is incorrect.
By som googling I found that using this:
num < 0 ? num * 0x8000 : num * 0x7FFF;
Im getting results equivalent to numpy implementation.
Is this a issue or I got something wrong ?
Current implementation is multiplying Float32 number in range (-1,1) by 0xFFFF which is eqvivalent to 65535. I found that with this conversion data is incorrect.
By som googling I found that using this:
Im getting results equivalent to numpy implementation.
Is this a issue or I got something wrong ?