diff --git a/dsv4/kernels/attention/fmha.py b/dsv4/kernels/attention/fmha.py
index dc71a968..e0cb3555 100644
--- a/dsv4/kernels/attention/fmha.py
+++ b/dsv4/kernels/attention/fmha.py
@@ -205,7 +205,8 @@ class FmhaKernel:
             gQ = tC_gQ[None, None, head_cta_idx, None, None, None]
             c_tiler = cute.slice_(self.pv_mma_tiler, (None, None, 0))  # (128, hd)
             tC_gC = cute.flat_divide(mC, c_tiler)
-            gC = tC_gC[None, None, head_cta_idx, None, None, None]
+            print(f"D2 DEBUG: tC_gC shape={cute.shape(tC_gC)}")
+            gC = tC_gC[None, None, head_cta_idx, None, None]
         else:
             gQ = cute.local_tile(mQ,cute.slice_(self.qk_mma_tiler,(None,0,None)),(None,None,None))
             gC = cute.local_tile(mC,cute.slice_(self.pv_mma_tiler,(None,None,0)),(None,None,None))