[all-commits] [llvm/llvm-project] 3a3377: [mlir][math]Update `convertPowfOp` `ExpandPatterns...

[Hyunsung Lee via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 3a3377579f137a0a6e14b60d891a9736707e7e8d
      https://github.com/llvm/llvm-project/commit/3a3377579f137a0a6e14b60d891a9736707e7e8d
  Author: Hyunsung Lee <ita9naiwa at gmail.com>
  Date:   2025-01-28 (Tue, 28 Jan 2025)

  Changed paths:
    M mlir/lib/Dialect/Math/Transforms/ExpandPatterns.cpp
    M mlir/test/Dialect/Math/expand-math.mlir
    M mlir/test/mlir-runner/test-expand-math-approx.mlir

  Log Message:
  -----------
  [mlir][math]Update `convertPowfOp` `ExpandPatterns.cpp` (#124402)

The current implementation of `convertPowfOp` requires a calculation of
`a * a` but, max\<fp16\> ~= 65,504, and if `a` is about 16, it will
overflow so get INF in fp8 or fp16 easily.


Remove support when `a < 0`. Overhead of handling negative value of `a`
is large and easy to overflow;

- related issue in iree:
https://github.com/iree-org/iree/issues/15936



To unsubscribe from these emails, change your notification settings at https://github.com/llvm/llvm-project/settings/notifications