Propagate split dtype to join #817
Conversation
…ualtran into propagate_split_dtype_to_join
|
Pass == dtype_ is preserved the whole way 
|
|
Post cirq interop split / join fix: |
|
|
fdmalone
requested review from
mpharrigan and
tanujkhattar
and removed request for
mpharrigan
|
^^ I'm not actually sure the cirq-interop splits and joins are correct looking and the picture. |
@fdmalone What are your concerns? One concern I see is that the I think we default to using |
|
Maybe it's unrelated, I was concerned by the integer-like QFxp(4, 0) splitting into a QFxp(2,2), but I think this fine. I was having some issues with this bloq in classical sim with Fxp but probably unrelated. Ignore my previous comment :) |
|
@tanujkhattar PTAL |
|
Safer casting of single qubits into QFxp(1,1) and back to QBit:
|
Debugging. Tidy.
| assert isinstance( | ||
| soq.reg.dtype, QBit | ||
| ), f"Found non-QBit type register which shouldn't happen: {soq.reg.name} {soq.reg.dtype}" | ||
| if not isinstance(in_reg.dtype, QBit): | ||
| err_msg = ( | ||
| "Found non-QBit type register which shouldn't happen: " | ||
| f"{soq.reg.name} {soq.reg.dtype}" | ||
| ) | ||
| assert isinstance(soq.reg.dtype, QBit), err_msg |
There was a problem hiding this comment.
The assertion probably does belon above since in both the if / else cases we assume that soq.reg.dtype is of type QBit ?
…ualtran into propagate_split_dtype_to_join
Fixes #722:
Adds back types to joins to ensure no loss of information when a split is followed by a join. I added a severity level to how strictly we may want to enforce type checking:
I added an autotester to colorize the check to show how strictly enforced our type checking currently is, and updated the cirq interop to handle the splits and joins there