[Rachit2323]

Fixes #1642.

Summary

On the cutlass-mlir backend, 300.0 as u8 wasn't clamped to 255 like it is in Rust. The exporter was emitting plain arith.fptosi/arith.fptoui, and those are undefined when the value is out of range. Now it uses the .sat intrinsics, the same as mir-lower does.

Changes

• mir/core.rs: float to int casts go through llvm.call_intrinsic with llvm.fptosi.sat / llvm.fptoui.sat. I copied the pattern from the saturating add/sub in mir/memory.rs.
• Ints wider than 64 bits now return an error, same as fix(mir-lower): reject FloatToInt to integer widths wider than 64 bits #878.
• 2 new tests.

Testing

No GPU on my machine, so I couldn't run a kernel.

• The new tests fail on main and pass with this.

• cargo test -p cuda-oxide-mlir-export, fmt and clippy are clean.

• I also pushed the MLIR through mlir-translate and llc by hand. The PTX has the clamp now, and it didn't before.

• just check (can't run it, no CUDA

• cargo oxide run (no GPU)

• New example (not needed)

Checklist

• All commits signed off (git commit -s)
• SPDX headers on new source files (no new files)

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.