Optimize gate count of Subtract to n-1 Toffolis #1057
Conversation
would be useful to include in the description |
|
@NoureldinYosri can you make the requested changes |
|
@mpharrigan done, I was getting test failures when doing the classical simulation test because classical simulation tools couldn't handle negative values and both Add and Subtract bloqs had faulty classical action definitions for the case of signed inputs... I fixed all of these issues |
|
@mpharrigan ptal |
| hN = N >> 1 | ||
| return {'a': a, 'b': int(a + b + hN) % N - hN} |
There was a problem hiding this comment.
h is half? can you provide a line comment explaining the logic here and consider using a more meaningful name than hN. Was the previous behavior using math.fmod wrong? can you add a test that would have failed with the old logic?
|
|
||
| This construction uses `XGate` and `AddK` to compute the twos-compliment of `b` before | ||
| doing a standard `Add`. | ||
| This construction uses the relation `a - b = ~(~a + b)` to turn the operation into addition. |
There was a problem hiding this comment.
are there any quantum references where this construction is used?
There was a problem hiding this comment.
@tanujkhattar you mentioned that this trick is used in some phase gradient papers, right?
There was a problem hiding this comment.
|
What happens if a_bitsize is equal to b_bitsize? This is a good example of a bloq that should support symbolic decomposition; at least for sizeof(a) == sizeof(b) |
|
@NoureldinYosri do take a look at #1161 and #1162, they would help simplify this decomposition |
There was a problem hiding this comment.
can you address @anurudhp 's comment and maybe open an issue to track
will do when the other PR is merged |
This uses the bit-magic
a-b = ~(~a + b)