【AutoParallel】Promote fuselinear pass in auto-parallel

python/paddle/distributed/auto_parallel/constants.py

@@ -145,6 +145,12 @@ def set_field_default_config(category, field, default_value):
 set_field_default_config(DATASET, "enable", False)
 set_field_default_config(DATASET, "num_shards", 1)
 
+# #########################################
+# # offload configuration
+# #########################################
+FUSEDLINEARPROMOTION = "fused_linear_promotion"
+set_field_default_config(FUSEDLINEARPROMOTION, "enable", False)
+
 #########################################
 # fused passes configuration
 #########################################

python/paddle/distributed/auto_parallel/static/parallelizer_v2.py

@@ -348,6 +348,35 @@ def _apply_post_optimization(
             )
             sp_pass.apply([main_program], [startup_program], self._pass_context)
 
+        # apply fused linear promotion pass
+        if (
+            self.is_train
+            and self._strategy.fused_linear_promotion.enable
+            and self._strategy.fused_passes.enable
+        ):
+            if (
+                len(self._strategy.fused_passes.fused_passes_list) > 0
+                and "fuse_gemm_epilogue"
+                in self._strategy.fused_passes.fused_passes_list
+            ):
+                amp_config = None
+                if self._strategy.amp.enable:
+                    amp_config = copy.deepcopy(self._strategy.amp.to_dict())
+                config = {}
+                config["dist_context"] = self._dist_context
+                config["global_rank"] = rank
+                config["enable_sp"] = self._strategy.sp_optimization.enable
+                config["params_grads"] = params_grads
+                config["amp_level"] = (
+                    amp_config['level'] if amp_config is not None else "o0"
+                )
+                fused_linear_promotion_pass = new_pass(
+                    "auto_parallel_fused_linear_promotion", config
+                )
+                fused_linear_promotion_pass.apply(
+                    [main_program], [startup_program], self._pass_context
+                )
+
         # data parallel optimization
         if self._strategy.dp_optimization.enable:
             config = copy.deepcopy(self._strategy.dp_optimization.to_dict())

python/paddle/distributed/auto_parallel/strategy.py

@@ -82,6 +82,12 @@ def __init__(self, config_dict=None):
         super().__init__(category, config_dict)
 
 
+class FusedPromotionConfig(BaseConfig):
+    def __init__(self, config_dict=None):
+        category = constants.FUSEDPROMOTION
+        super().__init__(category, config_dict)
+
+
 class AMPConfig(BaseConfig):
     def __init__(self, config_dict=None):
         category = constants.AMP
@@ -224,6 +230,9 @@ def __init__(self, config=None):
         config_dict = self._config_dict.get(constants.FUSED_PASSES, None)
         self.fused_passes = FusedPassesConfig(config_dict)
 
+        config_dict = self._config_dict.get(constants.FUSEDPROMOTION, None)
+        self.fused_linear_promotion = FusedPromotionConfig(config_dict)
+
         config_dict = self._config_dict.get(constants.DP_OPTIMIZATION, None)
         self.dp_optimization = DPOptimizationConfig(config_dict)
 

python/paddle/distributed/passes/__init__.py

@@ -22,6 +22,7 @@
 from .auto_parallel_quantization import *  # noqa: F403
 from .auto_parallel_data_parallel_optimization import *  # noqa: F403
 from .auto_parallel_grad_clip import *  # noqa: F403
+from .auto_parallel_fused_linear_promotion import *  # noqa: F403
 from .auto_parallel_supplement_explicit_dependencies import *  # noqa: F403
 from .auto_parallel_pipeline import *  # noqa: F403
 from .auto_parallel_sequence_parallel_optimization import *  # noqa: F403