Convert hexadecimal string \p Input to its binary representation. The return string is half the size of \p Input.

Fuse modules for quantization Fuses only the following sequence of modules: Linear + ReLU -> LinearRelu Conv2d + ReLU -> Conv2dRe

Given two MemRefTypes `aT` and `bT`, return a MemRefType to which both can be cast. If the MemRefTypes don't have the same rank or are not strided, re

Encodes a ral_dispatch op and appends the encoding to the string buffer `out`. The format: encoding = separator.join(target_name, device, inputs_encod

Returns the `edgePaddingLow` for `axis`.

Return the SSA values that represent the data point accessed using a given `indexingMap` for a given point in the iteration space represented by `ivs`

Returns the `InteriorPadding` for `axis`.

Returns a (per-process-level) unique name used for saving compiled result proto on disc.

Returns the number of device consts that have been emitted.

Returns the number of host consts that have been emitted.

TODO: only support the reduce body in the form of one lhlo_instruction and one terminator

Returns the sorted parallel axes.

Encodes a mlir type and appends the encoding to the string buffer `out`.

During unrolling from `originalShape` to `targetShape` return the offset for the slice `index`.

Get all nodes that can reach to the target_idx. Parameters ---------- target_idx : Set[int] The index of target n

True if the cluster has functional control flow like `If` and `While`.

Interpret the given character \p C as a hexadecimal digit and return its value. If \p C is not a valid hex digit, -1U is returned.

Insert disc_shape.tie_shape for each value having RankedTensorType type. before ``` %0 = ... : tensor<?x10xf32> use(%0) ``` after ``` %0 = ... : tenso

Insert the DataCollectObserver to the graph. For convenience, we group all inputs of each fusion group into a tensor list. Afte

convert: %0 = mhlo.constant dense<1> : tensor<f32> %1 = bcast %0, ... : tensor<?x?xf32> %2 = mhlo.mul %1, %arg : tensor<?x?xf32> %use(%2) to: %use(%ar

return true if op in fusion block

Returns true if the dimension `d` is the inner most.

Returns true if this fusion pattern is a kInput fusion.

It does not support transpose of batching dimension and other dimensions together now. TODO: it may support the transpose between batching dimensions

It does not support transpose of batching dimension and other dimensions together now. TODO: it may support the transpose between batching dimensions

Return true if this op is a rank-2 transpose

Returns true if the fusion type is stitch fusion.