fix: use gC not tCgC for TMA partition, group modes 0-3

tests/unit/test_fmha_v3_stage_c.py

@@ -460,7 +460,7 @@ class FmhaV3StageCMulti:
 
             cute.arch.fence_proxy("async.shared", space="cta")
 
-            # TMA store: SMEM → GMEM (reuse existing tma_c from kernel setup)
+            # TMA store: SMEM → GMEM
             # Sync all softmax warps before TMA store
             softmax_all_bar = pipeline.NamedBarrier(
                 barrier_id=5, num_threads=32 * len(self.epilogue_warp_id)
@@ -468,10 +468,11 @@ class FmhaV3StageCMulti:
             softmax_all_bar.arrive_and_wait()
 
             # Partition SMEM and GMEM for TMA store
+            epi_s = cute.select(self.c_smem_s, mode=[0, 1])
             tCsC, tCgC_tma = cpasync.tma_partition(
                 tma_c, 0, cute.make_layout((1,)),
                 cute.group_modes(sC, 0, 2),
-                cute.group_modes(tCgC, 0, 3),
+                cute.group_modes(gC, 0, 3),
             )
             cute.copy(tma_c, tCsC, tCgC_tma)
             cute.arch.cp_async_bulk_commit_group()