[inductor] don't try to reorder loops for template (#166910)

pytorchbot · shunting314 · web-flow · commit fc5612a499b0 · 2025-11-05T14:49:20.000-05:00
[inductor] don't try to reorder loops for template (#165601) fix #165579 Pull Request resolved: #165601 Approved by: https://github.com/yushangdi (cherry picked from commit a303d6d) Co-authored-by: Shunting Zhang <shunting@fb.com>
diff --git a/test/inductor/test_loop_ordering.py b/test/inductor/test_loop_ordering.py
@@ -592,6 +592,31 @@ def f(x, y):
             ".run(", 1 + int(inductor_config.benchmark_kernel), exactly=True
         ).run(code[0])
 
+    @inductor_config.patch(
+        {
+            "max_autotune": True,
+            "max_autotune_gemm_backends": "TRITON",
+            "test_configs.max_mm_configs": 4,
+        }
+    )
+    @skipUnless(HAS_GPU and is_big_gpu(), "Need big gpu for max-autotune")
+    def test_interaction_with_multi_template(self):
+        """
+        Skip MultiTemplateBuffer during loop reordering
+        """
+
+        @torch.compile
+        def f(x, y):
+            return (x @ y), x + 1
+
+        N = 2
+        x = torch.randn([N, N], device=GPU_TYPE, dtype=torch.bfloat16)
+        y = torch.randn([N, N], device=GPU_TYPE, dtype=torch.bfloat16)
+
+        out, code = run_and_get_code(f, x, y)
+        # didn't fuse due to small savings
+        FileCheck().check_count("@triton.jit", 2, exactly=True).run(code[0])
+
     def test_fuse_with_scalar_shared_memory(self):
         """
         Make sure if we can fuse two nodes sharing a scalar before,
diff --git a/torch/_inductor/scheduler.py b/torch/_inductor/scheduler.py
@@ -3953,6 +3953,12 @@ def shared_data_after_reordering_loop(
         ):
             return -1
 
+        # in some rare case, a template can be passed in.
+        # Check test_interaction_with_multi_template in test_loop_ordering.py
+        # and https://github.com/pytorch/pytorch/issues/165579
+        if node1.is_template() or node2.is_template():
+            return -1
+
         node1_buffer_names = node1.read_writes.buffer_names()
         node2_buffer_names = node2.read_writes.buffer_names()
         # Fast path: no common buffers.
