I'm a little confused by this also, but given the level of detail in the linked paper (http://eprint.iacr.org/2014/134.pdf) it's definitely not just an oversight. The issue may be that he's aiming for fastest possible performance on Sandy Bridge (AVX), and 256-bit integer vector multiplication wasn't added until Haswell (AVX2). And even then, 64-bit integer elements can only be done one at a time. But since he's suggesting instruction set changes, he's clearly not thinking only of backwards compatibility.
1XXXX
* 1XXXX
_________
XXXXX
XXXXX
XXXXX
XXXXX
1XXXX
Using a sticky bit means computing the logical-or of all the bits that are sufficiently far to the right in all these partial results, instead of computing them individually and summing them.Like for addition, the idea is that the result you obtain is different from the exact result, but rounds the same. The sticky bit only serves to round (midpoint + a small quantity) correctly up.
1XXXXYZZZZ
To round it we need:(1) the X bits for our answer
(2) the Y bit for rounding
(3) the logical-or of the Z bits for rounding
Ok so there might be a trick to compute (3) without computing ZZZZ, but we still want to know how to compute 1XXXXY without computing ZZZZ.
This explanation assumes that the N partial products are added sequentially. I don't know what techniques are used to achieve the 4- or 5-cycle latencies of modern desktop processors. Implementations with more parallelism and/or that are part of an FMA unit have it harder, but I still expect that they don't naively compute a full width partial result before rounding it (in the case of the FMA, it would be impractical as the partial result could be thousands of bits wide)
For some reason (maximum depth reached?) there is no “reply” link at the bottom of the post of yours at which I understood, so I am replying here.
So the answer to your question is... Floating Point numbers "don't round correctly". They're actually quite complicated, and require careful study.
There are very simple rules on how to round. This makes calculation of errors a bit complicated. http://en.wikipedia.org/wiki/IEEE_floating_point#Rounding_ru...
At best, Intel at one point provided 80-bit rounding (ie: 68-bit mantissa). That is, if you use the x87 floating-point coprocessor. But these technically do not match IEEE specifications for rounding.
In particular, to compute a product of two 53-bit floating-point numbers with correct rounding (as the standard mandates), it is certainly not sufficient in general to compute just a 54-bit approximation and round that.
> of about 25 bits [...]
> Why? Because the multiplier insists on rounding
> its low output bits rather than giving them back to the programmer.
> Why? Because the multiplier is buried inside the floating-point unit.
The reference to the 25-bit success makes it seem like the author is thinking of bits 53-105 of the 106-bit result of multiplying two 53-bit inputs.
The multiplier is not going to “give them back” to the programmer because it never computed them in the first place. Floating-point 53x53->53 multiplication is implemented with a couple of guard bits, the last of which is “sticky”. The multiplier the author is asking for is not “buried”, it does not exist in the processor.
I am not sure why integer multiplication is not preferred. It goes up to 64x64->128, if the author wants all the bits of the product.
> A 53-bit multiplier producing a full 106-bit product won't be much larger than a 53-bit multiplier producing a correctly rounded 53-bit floating-point result.
Yes, it will. It will have to be 106-bit wide instead of 56-bit wide.