[Mlir-commits] [mlir] [mlir] [math] Fix the precision issue of expand math (PR #120865)

[donald chen]

================
@@ -215,31 +215,6 @@ static LogicalResult convertFmaFOp(math::FmaOp op, PatternRewriter &rewriter) {
   return success();
 }
 
-// Converts a floorf() function to the following:
----------------
cxy-1993 wrote:

While adding relevant documentation, I discovered that this is already described in mlir/docs/Canonicalization.md: operations should be recoverable after canonicalization. I haven't found a more suitable place to add this documentation.

https://github.com/llvm/llvm-project/pull/120865