[Mlir-commits] [mlir] [MLIR][ARITH] Adds missing foldings for truncf (PR #128096)

[Zahi Moudallal]

================
@@ -1517,6 +1517,22 @@ LogicalResult arith::TruncIOp::verify() {
 /// Perform safe const propagation for truncf, i.e., only propagate if FP value
 /// can be represented without precision loss.
 OpFoldResult arith::TruncFOp::fold(FoldAdaptor adaptor) {
+  if (auto extOp = getOperand().getDefiningOp<arith::ExtFOp>()) {
----------------
zahimoud wrote:

> I remember needing to go fix a similar folding (that one was for y = trunc(ext(bf16 x to f32) to bf16)) in LLVM because it was causing numerical correctness issues

Can you explain why this one would cause numerical issues ? ext to f32 is just adding 16 zero mantissa bits, then trunc back to bf16 would remove those zero bits, I don't see any rounding error there.

https://github.com/llvm/llvm-project/pull/128096