[Mlir-commits] [mlir] [mlir][vector] Linearization: push 'bit width' logic out of patterns (PR #136581)

[James Newling]

newling wrote:

> The motivation to use dialect conversion instead of just greedy rewriter was so we can reuse existing generic scf/cf/func conversions to convert vector types through control flow ops.

Can you please give a bit more information, I'm not familiar with this. A pointer to an example of where this is done would be useful. Is this specific to linearization, or do you think it should be approach for the other pattern-adding APIs in [VectorRewritePatterns](https://github.com/llvm/llvm-project/blob/69327c16d17b13cb2bc27968a557276bd82eef9a/mlir/include/mlir/Dialect/Vector/Transforms/VectorRewritePatterns.h#L4) ? 

https://github.com/llvm/llvm-project/pull/136581