Revert "[PT2][fusion] ban fusions with large accumulated reads (#157563)"

pytorchmergebot · pytorchmergebot · commit 26807dcf277f · 2025-07-15T16:35:55.000Z
This reverts commit c062550. Reverted #157563 on behalf of https://github.com/clee2000 due to broke test_linear_and_cel on main https://hud.pytorch.org/pytorch/pytorch/commit/c062550a3598d27c2d6572db7c0f4ff90a84cc84, caused OOM? Also broken on PR, Dr. CI classification is wrong (claims the test is disabled by an issue but the issue is for a different test). Also I'm pretty sure the expected results json is supposed to have a ton of empty lines, its to prevent merge conflicts, I will add it to the linter ([comment](#157563 (comment)))
diff --git a/benchmarks/dynamo/pr_time_benchmarks/expected_results.csv b/benchmarks/dynamo/pr_time_benchmarks/expected_results.csv
@@ -1,23 +1,89 @@
-add_loop_eager,compile_time_instruction_count,2996000000,0.015
+add_loop_eager,compile_time_instruction_count,3017000000,0.015
+
+
+
 add_loop_eager_dynamic,compile_time_instruction_count,4352000000,0.025
-add_loop_inductor,compile_time_instruction_count,33090000000,0.015
-add_loop_inductor_dynamic_gpu,compile_time_instruction_count,42660000000,0.025
-add_loop_inductor_gpu,compile_time_instruction_count,29690000000,0.015
+
+
+
+add_loop_inductor,compile_time_instruction_count,29490000000,0.015
+
+
+
+add_loop_inductor_dynamic_gpu,compile_time_instruction_count,38760000000,0.025
+
+
+
+add_loop_inductor_gpu,compile_time_instruction_count,26000000000,0.015
+
+
+
 basic_modules_ListOfLinears_eager,compile_time_instruction_count,947600000,0.015
-basic_modules_ListOfLinears_inductor,compile_time_instruction_count,18830000000,0.015
-basic_modules_ListOfLinears_inductor_gpu_force_shape_pad,compile_time_instruction_count,17460000000,0.015
-basic_modules_ListOfLinears_inductor_gpu,compile_time_instruction_count,11020000000,0.2
+
+
+
+basic_modules_ListOfLinears_inductor,compile_time_instruction_count,18490000000,0.015
+
+
+
+basic_modules_ListOfLinears_inductor_gpu_force_shape_pad,compile_time_instruction_count,17020000000,0.015
+
+
+
+basic_modules_ListOfLinears_inductor_gpu,compile_time_instruction_count,10297683772,0.2
+
+
+
 update_hint_regression,compile_time_instruction_count,1673000000,0.02
+
+
+
 sum_floordiv_regression,compile_time_instruction_count,986800000,0.015
-symint_sum,compile_time_instruction_count,3184000000,0.015
+
+
+
+symint_sum,compile_time_instruction_count,3166000000,0.015
+
+
+
 symint_sum_loop,compile_time_instruction_count,4202000000,0.015
+
+
+
 aotdispatcher_inference_nosubclass_cpu,compile_time_instruction_count,2103000000,0.015
+
+
+
 aotdispatcher_inference_subclass_cpu,compile_time_instruction_count,6004000000,0.015
+
+
+
 aotdispatcher_partitioner_cpu,compile_time_instruction_count,8783000000,0.015
+
+
+
 aotdispatcher_partitioner_cpu2,compile_time_instruction_count,1940000000,0.015
+
+
+
 aotdispatcher_training_nosubclass_cpu,compile_time_instruction_count,3885000000,0.015
+
+
+
 aotdispatcher_training_subclass_cpu,compile_time_instruction_count,10470000000,0.015
-mm_loop_inductor_gpu,compile_time_instruction_count,4365000000,0.015
-mm_loop_inductor_dynamic_gpu,compile_time_instruction_count,8184000000,0.015
+
+
+
+mm_loop_inductor_gpu,compile_time_instruction_count,4324000000,0.015
+
+
+
+mm_loop_inductor_dynamic_gpu,compile_time_instruction_count,8116000000,0.015
+
+
+
 basic_NestedModule_eager,compile_time_instruction_count,8152524390,0.015
+
+
+
 basic_InlineMod_eager,compile_time_instruction_count,7255000000,0.015
diff --git a/test/inductor/test_memory.py b/test/inductor/test_memory.py
@@ -306,57 +306,6 @@ def f(a, b, c):
         expected_bound = a.size(0) * c.size(1) * a.dtype.itemsize * 2
         self.assertLess(peak_mem, expected_bound)
 
-    def test_fusion_acc_large_reads(self):
-        def f(x, y, z):
-            res = torch.zeros_like(x[0])
-            for i in range(4):
-                temp = torch.matmul(x, y) + z
-                res = res + temp
-            return res
-
-        N = 128
-        x = torch.rand(N, N, dtype=torch.float32, device=GPU_TYPE)
-        y = torch.rand(N, N, dtype=torch.float32, device=GPU_TYPE)
-        z = torch.rand(N, N, dtype=torch.float32, device=GPU_TYPE)
-
-        # CASE 1: no restriction on the amount of accumulation
-        with config.patch({"realize_acc_reads_size_threshold": float("inf")}):
-            f_compiled = torch.compile(f)
-            code = run_and_get_triton_code(f_compiled, x, y, z)
-            (
-                FileCheck()
-                .check("triton_poi_fused_add_0.run(buf4, arg2_1, buf1, buf2, buf3")
-                .run(code)
-            )
-
-        # CASE 2: for tensors with the same size as x (which is 4 * N**2 bytes)
-        # at most 12 / 4 = 3 reads can be accumulated during fusion
-        with config.patch({"realize_acc_reads_size_threshold": 12 * N**2}):
-            f_compiled = torch.compile(f)
-            code = run_and_get_triton_code(f_compiled, x, y, z)
-            (
-                FileCheck()
-                .check("triton_poi_fused_add_0.run(buf3, arg2_1, buf1, buf2,")
-                .check("triton_poi_fused_add_1.run(buf5, buf4, arg2_1,")
-                .run(code)
-            )
-
-        # CASE 3: no such fusion allowed
-        with config.patch({"realize_acc_reads_size_threshold": N**2}):
-            f_compiled = torch.compile(f)
-            code = run_and_get_triton_code(f_compiled, x, y, z)
-            (
-                FileCheck()
-                .check("triton_poi_fused_add_0.run(buf1, arg2_1,")
-                .check("triton_poi_fused_add_0.run(buf3, arg2_1,")
-                .check("triton_poi_fused_add_0.run(buf4, buf3,")
-                .check("triton_poi_fused_add_0.run(buf6, arg2_1,")
-                .check("triton_poi_fused_add_0.run(buf7, buf6,")
-                .check("triton_poi_fused_add_0.run(buf9, arg2_1,")
-                .check("triton_poi_fused_add_0.run(buf10, buf9,")
-                .run(code)
-            )
-
 
 if __name__ == "__main__":
     from torch._inductor.test_case import run_tests
diff --git a/test/inductor/test_online_softmax.py b/test/inductor/test_online_softmax.py
@@ -13,7 +13,6 @@
     instantiate_parametrized_tests,
     IS_LINUX,
     parametrize,
-    serialTest,
 )
 from torch.testing._internal.inductor_utils import GPU_TYPE, HAS_CUDA
 
@@ -78,17 +77,12 @@ def f(x):
         out, source_codes = run_and_get_code(f, x)
         return source_codes[0]
 
-    @serialTest()
     def test_codegen_3pass_softmax_due_to_disable(self):
-        with inductor_config.patch(
-            online_softmax=False,
-            realize_acc_reads_size_threshold=float("inf"),
-        ):
+        with inductor_config.patch(online_softmax=False):
             wrapper_code = self.get_softmax_wrapper()
 
         self.assertEqual(wrapper_code.count("for r0_offset in"), 3)
 
-    @serialTest()
     @parametrize("V", [2048, 50304])
     @parametrize("use_log_softmax", [False, True])
     def test_codegen_online_softmax(self, use_log_softmax, V):
diff --git a/torch/_inductor/choices.py b/torch/_inductor/choices.py
@@ -365,10 +365,6 @@ def can_fuse(
             WhyNoFuse(node1, node2)("Fusion will increase peak memory")
             return False
 
-        if scheduler.fusion_accumulate_large_reads(node1, node2):
-            WhyNoFuse(node1, node2)("Fusion accumulate large amount of reads")
-            return False
-
         return True
 
     @staticmethod
diff --git a/torch/_inductor/config.py b/torch/_inductor/config.py
@@ -574,7 +574,6 @@ def use_autoheuristic(name: str) -> bool:
 
 # Threshold to prevent excessive accumulation of ops in one buffer during lowering
 realize_acc_reads_threshold = 8
-realize_acc_reads_size_threshold = 3 * (1024**3)
 
 # fallback to eager for random/dropout, this is slow but useful for debugging
 fallback_random = False
diff --git a/torch/_inductor/graph.py b/torch/_inductor/graph.py
@@ -123,7 +123,6 @@
     from torch.fx.graph import Graph
 
     from .codegen.wrapper import PythonWrapperCodegen
-    from .dependencies import Dep
     from .scheduler import BaseSchedulerNode
 
     CompiledModule = Union[ModuleType, FileBackedGraphModule]
@@ -486,9 +485,6 @@ def __init__(
 
         self.bw_donated_idxs = get_donated_idxs()
 
-        # Cache for dep size hints to avoid expensive recomputation
-        self.dep_size_hint_cache: dict[Dep, int] = {}
-
     def freeze_runtime_asserts(self) -> None:
         self._shape_env.freeze_runtime_asserts()
 
@@ -574,23 +570,6 @@ def has_feature(
         assert isinstance(feature, BackendFeature), feature
         return feature in self.get_backend_features(get_device_type(device))
 
-    def get_dep_size_hint(self, dep: Dep) -> int:
-        """
-        Get the size hint for a dependency with caching to avoid expensive recomputation.
-        """
-        if dep not in self.dep_size_hint_cache:
-            res = 0
-            try:
-                if not dep.has_unbacked_symbols():
-                    res = dep.numbytes_hint()
-            except KeyError:
-                # In at least one test (test/inductor/test_torchbind.py) we
-                # create a StarDep that doesn't exist in the graph and calling
-                # `has_unbacked_symbols()` throws an error.
-                pass
-            self.dep_size_hint_cache[dep] = res
-        return self.dep_size_hint_cache[dep]
-
     def get_current_device_or_throw(self) -> torch.device:
         if device := self.current_device:
             return device
diff --git a/torch/_inductor/ir.py b/torch/_inductor/ir.py
@@ -7829,10 +7829,6 @@ def create(data: IRNode) -> Union[TensorBox, ShapeAsConstantBuffer]:
 
 
 class StorageBox(MutableBox):
-    """
-    StorageBox allow in-place mutation of Tensors
-    """
-
     def is_input_buffer(self) -> bool:
         if isinstance(self.data, (InputBuffer, ReinterpretView)):
             return self.data.get_name() in V.graph.graph_inputs
@@ -7882,17 +7878,10 @@ def realize_hint(self) -> None:
         ):
             self.realize()
 
-    def has_accumulated_enough_reads_by_size(self) -> bool:
-        return (
-            sum(V.graph.get_dep_size_hint(dep) for dep in self.get_reads())
-            > config.realize_acc_reads_size_threshold
-        )
-
     def has_exceeded_max_reads(self) -> bool:
         return isinstance(self.data, Pointwise) and (
             self.num_reads() > config.realize_acc_reads_threshold
             or self.has_large_inner_fn()
-            or self.has_accumulated_enough_reads_by_size()
         )
 
     def should_realize_on_reuse(self, users: int) -> bool:
diff --git a/torch/_inductor/memory.py b/torch/_inductor/memory.py
@@ -78,8 +78,19 @@ def get_freeable_input_buf(
         A dictionary containing all freeble input buffers, keyed by their names.
     """
 
+    # this function is copied from torch/_inductor/scheduler.py
+    # TODO: would be nice to remove the try/except block for both places
     def _dep_size_hint(dep: Dep) -> int:
-        return V.graph.get_dep_size_hint(dep)
+        res = 0
+        try:
+            if not dep.has_unbacked_symbols():
+                res = dep.numbytes_hint()
+        except KeyError:
+            # In at least one test (test/inductor/test_torchbind.py) we
+            # create a StarDep that doesn't exist in the graph and calling
+            # `has_unbacked_symbols()` throws an error.
+            pass
+        return res
 
     # get freeable input buffers' successor nodes and their sizes
     # note that different deps can have the same name, so we use name as keys
diff --git a/torch/_inductor/scheduler.py b/torch/_inductor/scheduler.py
@@ -2051,12 +2051,15 @@ class Scheduler:
     optimizations such as fusion, reorder, and graph partition.
     """
 
+    __dep_size_hint_cache: dict[Dep, int]
+
     def __init__(self, nodes: list[ir.Operation]) -> None:
         with dynamo_timed("Scheduler.__init__"):
             self._init(nodes)
 
     def _init(self, nodes: list[ir.Operation]) -> None:
         super().__init__()
+        self.__dep_size_hint_cache = {}
         V.graph.scheduler = self
         self.backends: dict[torch.device, BaseScheduling] = {}
         self.post_grad_graph_id = next(_post_grad_graph_counter)
@@ -3502,17 +3505,6 @@ def _find_single_user_inputs(
             return True
         return False
 
-    def fusion_accumulate_large_reads(
-        self, node1: BaseSchedulerNode, node2: BaseSchedulerNode
-    ) -> bool:
-        all_reads = (node1.read_writes.reads | node2.read_writes.reads) - (
-            node1.read_writes.writes | node2.read_writes.writes
-        )
-        return (
-            sum(self.dep_size_hint(dep) for dep in all_reads)
-            > config.realize_acc_reads_size_threshold
-        )
-
     def are_long_distant_nodes(
         self, node1: BaseSchedulerNode, node2: BaseSchedulerNode
     ) -> bool:
@@ -4018,7 +4010,20 @@ def fusable_read_and_write(self, read: Dep, write: MemoryDep) -> bool:
         return False
 
     def dep_size_hint(self, dep: Dep) -> int:
-        return V.graph.get_dep_size_hint(dep)
+        res = 0
+        if dep not in self.__dep_size_hint_cache:
+            try:
+                if not dep.has_unbacked_symbols():
+                    res = dep.numbytes_hint()
+            except KeyError:
+                # In at least one test (test/inductor/test_torchbind.py) we
+                # create a StarDep that doesn't exist in the graph and calling
+                # `has_unbacked_symbols()` throws an error.
+                pass
+            self.__dep_size_hint_cache[dep] = res
+        else:
+            res = self.__dep_size_hint_cache[dep]
+        return res
 
     def score_fusion_memory(
         self, node1: BaseSchedulerNode, node2: BaseSchedulerNode
