v0.0.0-6588-g30773ab2c

Allow configuration of i/o flop on per-channel basis in IR

This adds support for input_flop_kind and output_flop_kind on channels. If present these override the codegen `--flop_inputs[_kind=...]` and `--flop_outputs[_kind=...]` flags for ports associated with those channels. These should be generated by the frontend or added by some other pass. These options have no effect on the semantics of the proc ir and are only a directive for codegen conversion.

PiperOrigin-RevId: 706887241

v0.0.0-6581-g4611acdf4

Update xls/ir subpackage to follow updated StatusOr style guidance at…

v0.0.0-6564-gf999382f2

Merge pull request #1780 from xlsynth:cdleary/2024-12-09-c-apis-for-c…

v0.0.0-6545-g69c1c4510

[XLS] Avoid creating cycles when merging channel operations

It is possible for our mutual-exclusion pass to determine that operations are mergeable in such a way that, if all merges are carried out, we will end up creating a cycle. For instance, if we have mergeable classes {A, C} and {B, D}, where D depends on A and C depends on B, we would end up introducing a cyclic dependency.

To avoid this, we check before merging operations whether the result would add a cyclic dependency. If it would, we prevent the merge; if the channels are using `proven_mutually_exclusive` strictness, we produce a more useful error.

Fixes google/xls#1716

PiperOrigin-RevId: 705261047

v0.0.0-6537-g1dfce581f

Merge pull request #1774 from xlsynth:cdleary/2024-12-06-fuzz-xn

PiperOrigin-RevId: 704973407

v0.0.0-6516-gf59d49721

Added patch IR for applying patches generated by the IR diff on IRs.

PiperOrigin-RevId: 703630687

v0.0.0-6507-g245ca1f62

[xls][mlir] Add `extern_sproc` and `extern_eproc` ops

These ops allow to reference procedures external to the current module. Support has
been added in -elaborate-procs to elaborate from extern_sproc to extern_eproc.

No support has been added for linking modules and "resolving" the externs.

PiperOrigin-RevId: 703196503

v0.0.0-6489-g6c9f5bd17

Move formatting `VerbatimNode` and `ConstAssert` from `Fmt` standalon…

v0.0.0-6485-g4d46c68da

Add setting Codegen NG flag during fuzzer sample generation.

PiperOrigin-RevId: 702185844

v0.0.0-6466-gef2bd29a6

[mlir][xls] Optimize ternary_ops::FromKnownBits()

On workloads with large bits datatypes this function has been observed to be super hot (dominating in
the profile).

In 120s of CPU time in this benchmark, 53s comes directly from this function. After optimization
the function takes 500ms.

PiperOrigin-RevId: 701531383