SMEM-P: add iterator offset debug print

dsv4/kernels/attention/fmha.py

@@ -186,6 +186,12 @@ class FmhaKernel:
         tCrP = pv_mma.make_fragment_A(sP)
         if self.use_smem_p:
             print(f"[SMEM-P DEBUG] tCrP shape: {cute.shape(tCrP)} layout: {tCrP.layout}")
+            # DEBUG: compute iterator offset between tCrP and sP
+            try:
+                offset_elems = tCrP.iterator - sP.iterator
+                print(f"[SMEM-P DEBUG] tCrP iterator offset: {offset_elems}")
+            except:
+                print(f"[SMEM-P DEBUG] iterator offset not available")
         # tOrP0 always defined as tOrP. The TMEM-P path in the MMA warp applies
         # the p0 column offset inline when constructing the gemm arguments.
         tOrP0 = tOrP