diff --git a/tests/unit/test_fmha_v3_stage_c.py b/tests/unit/test_fmha_v3_stage_c.py
index 4c79cab9..02fbacb3 100644
--- a/tests/unit/test_fmha_v3_stage_c.py
+++ b/tests/unit/test_fmha_v3_stage_c.py
@@ -148,6 +148,7 @@ class FmhaV3StageCMulti:
             s_bar: cute.struct.MemRange[cutlass.Int64, 2]
             acc_bar: cute.struct.MemRange[cutlass.Int64, self.num_acc_stage*2]
             tmem_dealloc: cutlass.Int64; holding: cutlass.Int32
+            kv_coord_smem: cute.struct.MemRange[Int32, 1]
         smem = utils.SmemAllocator(); st = smem.allocate(SS)
 
         qp,qc = pipeline.PipelineTmaUmma.create(barrier_storage=st.q_bar.data_ptr(),num_stages=self.q_stage,producer_group=pipeline.CooperativeGroup(pipeline.Agent.Thread),consumer_group=pipeline.CooperativeGroup(pipeline.Agent.Thread,1),tx_count=self.q_tx_bytes,cta_layout_vmnk=cl_vmnk,defer_sync=True).make_participants()
@@ -224,11 +225,15 @@ class FmhaV3StageCMulti:
             cute.copy(tma_q, tAgQ[(None, Int32(0))], tAsQ[(None, qh.index)], tma_bar_ptr=qh.barrier)
             qp.tail()
             kvp.reset()
-            # DEBUG: Use constant Int32(1) to test if TMA can read from tile 1 at all
+            # Use SMEM-backed counter — the JIT can't constant-fold SMEM reads.
+            # Initialize to 0, read before each copy, increment after.
+            st.kv_coord_smem[0] = Int32(0)
             for kt in range(n_kv_tiles):
+                kv_coord = st.kv_coord_smem[0]  # Dynamic read from SMEM
                 kvh = kvp.acquire_and_advance()
-                cute.copy(tma_k, tBgK[(None, Int32(1))], tBsK[(None, kvh.index)], tma_bar_ptr=kvh.barrier)
-                cute.copy(tma_v, tVgV[(None, Int32(1))], tVsV[(None, kvh.index)], tma_bar_ptr=kvh.barrier)
+                cute.copy(tma_k, tBgK[(None, kv_coord)], tBsK[(None, kvh.index)], tma_bar_ptr=kvh.barrier)
+                cute.copy(tma_v, tVgV[(None, kv_coord)], tVsV[(None, kvh.index)], tma_bar_ptr=kvh.barrier)
+                st.kv_coord_smem[0] = kv_coord + 1  # Write back to SMEM
             kvp.tail()
 
         # ===== MMA warp =====