test: clean minimal nvvm.inline_ptx test

tests/unit/test_ptx_clean.py

@@ -0,0 +1,40 @@
+"""Minimal nvvm.inline_ptx test - no debug env vars."""
+import torch
+import cutlass.cute as cute
+import cutlass.torch as cutlass_torch
+from cutlass.cutlass_dsl import dsl_user_op, T
+from cutlass._mlir.dialects import nvvm
+from cutlass.cute.typing import Float32, Int32
+
+
+@dsl_user_op
+def f32_to_i32_rni(x: Float32, *, loc=None, ip=None) -> Int32:
+    result = nvvm.inline_ptx(
+        write_only_args=[T.i32()],
+        read_only_args=[Float32(x).ir_value(loc=loc, ip=ip)],
+        ptx_code="cvt.rni.s32.f32 $0, $1;",
+        loc=loc,
+        ip=ip,
+    )
+    return Int32(result)
+
+
+@cute.kernel
+def test_k(inp: cute.Tensor, out: cute.Tensor):
+    tidx, _, _ = cute.arch.thread_idx()
+    if tidx == Int32(0):
+        x = cute.arch.load(inp.iterator, Float32)
+        r = f32_to_i32_rni(x)
+        cute.arch.store(out.iterator, r)
+
+
+if __name__ == "__main__":
+    x = torch.tensor([3.7], dtype=torch.float32, device='cuda')
+    o = torch.zeros(1, dtype=torch.int32, device='cuda')
+    xc = cutlass_torch.from_dlpack(x).mark_layout_dynamic(leading_dim=0)
+    oc = cutlass_torch.from_dlpack(o).mark_layout_dynamic(leading_dim=0)
+    print("Compiling...")
+    compiled = cute.compile(test_k, xc, oc)
+    print("Running...")
+    compiled(xc, oc)
+    print(f"Result: {o.item()} (expected 4)")