Alexandre Eichenberger

Last use of GetRef is associated with Strings

1

Here is the last use of `KrnlGetRefOp` (created via the Krill Builder `getRef()`. ONNXToKrnl...CategoryMapper.cpp ``` Value loadElement(Value memref, ValueRange loopInd, Type elementType, int64_t rank, KrnlBuilder &createKrnl) const { Value inputElem;...

MHLO code in onnx-mlir: Status requested.

6

We would like to gather an honest opinion on how active the MHLO code is within the onnx-mlir code base. As it is a dependence that causes significant amount of...

Our current mapping of ONNX to zAIU ops is as follow 1. Map all zAIU-eligible operations to zAIU greedily, adding stick/unstick for input/outputs while eventually removing redundant stick/unstick or unsick/stick...

Resnet with batch size>1 has issue on NNPA

2

Reporting an error that Cindy Guo experienced when targeting Resnet for NNPA with batch size > 1. She reported the error using this benchmark: > this is one I used:...

Add krnl prefetch

1

Cannot add `affine.prefetch` during ONNX to KRNL lowering because affine prefetch check the enclosing loop nests, which are not yet `affine.for` loops during KRNL lowering. Thus created a placeholder `krnl.prefetch`...

Using StableHLO for ONNX to HLO conversion

5

In light of the presentation on OpenXLA where they touted StableHLO as the preferred representation for exchanges between compilers/tools, would it make sense to migrate our support from ONNX to...

Substitute zdnn calls for stick/unstick late, after most ZLow optimizations are performed

1

The many prior instances used buffered / tiled stick unstick code gen, but the redundant load / store ended up costing more than the simple approach of stick/unstick and store....

Updated the README page with the results for the new term June 2024 - May 2025. Thanks to all contributors

[Dynamic] Linear Quantization Linear test is not friendly to optimization (reciprocal)

13

# Ask a Question ### Question The dynamic quantization is computed as listed below in the ONNX backend tests ``` x_min = np.minimum(0, np.min(X)) x_max = np.maximum(0, np.max(X)) Y_Scale =...