diff --git a/dsv4/kernels/attention/fmha.py b/dsv4/kernels/attention/fmha.py
index 20d02f73..300ca8c0 100644
--- a/dsv4/kernels/attention/fmha.py
+++ b/dsv4/kernels/attention/fmha.py
@@ -366,12 +366,8 @@ class FmhaKernel:
             # O rescale atoms (hand-constructed, using composition layout like CUTLASS correction_rescale)
             # Only needed when there are multiple KV tiles (O must be rescaled per-kt).
             # With n_kv_tiles=1, no rescale is needed (kt is always 0).
-            # Define placeholder values unconditionally for CuTeDSL scoping.
             corr_tile_size = 16
             n_corr_tiles = self.pv_n_tile // corr_tile_size
-            tTMrO = cute.make_rmem_tensor(
-                (cute.make_layout((1,)), 1), self.acc_dtype
-            )
             if const_expr(self.n_kv_tiles > 1):
                 tOcO = pv_thr.partition_C(cS)
                 tOtO_i_layout = cute.composition(tOtO0.layout, cute.make_layout((128, corr_tile_size)))