When using crt encoding, some fhe.zero op results will be converted to tensors (crt encoded eint), so should be converted to tfhe.zero_tensor operations instead of tfhe.zero
Printing the mnenomic in the custom printers for RT types leads to a
repetition, since the mnenomic is already printed by the
infrastructure invoking the custom printer (e.g., instead of
RT.future<...>, the printed type is `RT.futurefuture<...>`).
This commit removes the mnenomics from the custom printers and thus
causes them to emit the correct textual representation of types.
client parameter generation was removing last dimension of a tensor when using CRT including plaintext ones, while it should only be done for encrypted ones.
Key serialization for transfers between nodes in clusters was broken
since the changes introduced to separate keys from key parameters and
introduction of support for multi-key (ref
cacffadbd2).
This commit restores functionality for distributing keys to non-shared
memory nodes.
This adds a new pattern to the batching pass that folds operations on
tensors of constants into new tensors of constants. E.g.,
%cst = arith.constant dense<...> : tensor<Nxi9>
%res = scf.for %i = %c0 to %cN {
%cst_i9 = tensor.extract %cst[%i]
%cst_i64 = arith.extui %cst_i9 : i64
...
}
becomes:
%cst = arith.constant dense<...> : tensor<Nxi64>
%res = scf.for %i = %c0 to %cN {
%cst_i64 = tensor.extract %cst[%i]
...
}
The pattern only works for static loops, indexes that are quasi-affine
expressions on single loop induction variables with a constant step
size across iterations and foldable operations that have a single
result.
The current batching pass only supports batching of operations that
have a single batchable operand, that can only be batched in one way
and that operate on scalar values. However, this does not allow for
efficient batching of all arithmetic operations in TFHE, since these
are often applied to pairs of scalar values from tensors, to tensors
and scalars or to tensors that can be grouped in higher-order tensors.
This commit introduces three new features for batching:
1. Support of multiple batchable operands
The operation interface for batching now allows for the
specification of multiple batchable operands. This set can be
composed of any subset of an operation's operands, i.e., it is
not limited to sets of operands with contiguous operand indexes.
2. Support for multiple batching variants
To account for multiple kinds of batching, the batching operation
interface `BatchableOpInterface` now supports variants. The
batching pass attempts to batch an operation by trying the
batching variants expressed via the interface in order until it
succeeds.
3. Support for batching of tensor values
Some operations that could be batched already operate on tensor
values. The new batching pass detects those patterns and groups
the batchable tensors' values into higher-dimensional tensors.
The batching pass erroneously assumes that any expression solely
composed of an induction variable has static bounds. This commit adds
a test for the lower bound, upper bound and step checking that they
are indeed static before attempting to determine their static values.
