Update BMM_Theory: bmm_theory/branches/stream_k.py (fix review issues #4-#10)

BMM/BMM_Theory/bmm_theory/branches/stream_k.py

@@ -145,11 +145,12 @@ class StreamKBranch(Branch):
         t_mmad = t_mmad_tile / grid_k
         t_mte2 = t_mte2_tile / grid_k
 
-        # 归约: 部分和 4B 驻留 L2, AIV 归约
+        # 归约: 部分和 4B 驻留 L2, AIV 归约 (含最终按 C dtype 写回)
         t_reduce = eval_streamk_reduce(tile_elems, grid_k, out_b, s)
 
-        # 写出: 最终归约结果按 C dtype; 中间部分和按 4B
-        fix_bytes = tile_elems * out_b
+        # Fixpipe: 部分和写出按 4B (L0C dtype, 防精度丢失), 驻留 L2.
+        # 最终归约结果的 C dtype 写回已在 t_reduce 内计, 此处不重复 (issue#9 口径对齐).
+        fix_bytes = grid_k * tile_elems * 4
         t_fix = fix_bytes / s.bw_l2_pc
 
         flops_pc = 2.0 * tile_elems * k / grid_k
@@ -158,7 +159,8 @@ class StreamKBranch(Branch):
         return assemble_timing(
             t_mte2_gm=t_mte2, t_mte2_l2=0.0, t_dma_cmd=0.0,
             t_mmad=t_mmad, t_fixpipe=t_fix, t_reduce=t_reduce,
-            t_drain=t_reduce,  # 归约串行追加
+            t_drain=t_reduce,  # 归约串行追加 (reduce_serial 默认 True, 不进稳态 max)
             gm_read_bytes=gm_bytes, l2_read_bytes=0.0, dma_cmd_count=0.0,
             cube_flops=flops_pc, fixpipe_bytes=fix_bytes,
+            reduce_serial=True,
         )