Make user defined Triton kernels serializable for fx_graph_runnable (#160002)

PaulZhang12 · pytorchmergebot · commit cf0a0dcb0afa · 2025-08-11T20:54:33.000Z
Resolves issue #153475 where `fx_graph_runnable` didn't work with user defined triton kernels. Pull Request resolved: #160002 Approved by: https://github.com/eellison
diff --git a/test/dynamo/test_fx_graph_runnable.py b/test/dynamo/test_fx_graph_runnable.py
@@ -11,12 +11,65 @@
 from torch._inductor.codecache import WritableTempFile
 from torch._inductor.test_case import TestCase
 from torch.testing._internal.common_utils import IS_FBCODE, IS_SANDCASTLE
+from torch.utils._triton import has_triton
 
 
 if torch.distributed.is_available():
     from torch.distributed._tensor import DeviceMesh, DTensor, Replicate, Shard
     from torch.testing._internal.distributed.fake_pg import FakeStore
 
+if has_triton():
+    import triton
+    import triton.language as tl
+
+    def init_to_zero(name):
+        return lambda nargs: nargs[name].zero_()
+
+    @triton.jit
+    def add_kernel(x_ptr, y_ptr, output_ptr, n_elements, BLOCK_SIZE: tl.constexpr):
+        pid = tl.program_id(axis=0)
+
+        block_start = pid * BLOCK_SIZE
+        offsets = block_start + tl.arange(0, BLOCK_SIZE)
+        mask = offsets < n_elements
+
+        x = tl.load(x_ptr + offsets, mask=mask)
+        y = tl.load(y_ptr + offsets, mask=mask)
+        output = x + y
+        tl.atomic_add(output_ptr + offsets, output, mask=mask)
+
+    @triton.autotune(
+        configs=[
+            triton.Config(
+                {"BLOCK_SIZE": 1024},
+                num_warps=4,
+                num_stages=2,
+                pre_hook=init_to_zero("output_ptr"),
+            )
+        ],
+        pre_hook=init_to_zero("output_ptr"),
+        post_hook=init_to_zero("output_ptr"),
+        key=["n_elements"],
+    )
+    @triton.jit
+    def add_kernel_autotune(
+        x_ptr, y_ptr, output_ptr, n_elements, BLOCK_SIZE: tl.constexpr
+    ):
+        pid = tl.program_id(axis=0)
+
+        block_start = pid * BLOCK_SIZE
+        offsets = block_start + tl.arange(0, BLOCK_SIZE)
+        mask = offsets < n_elements
+
+        x = tl.load(x_ptr + offsets, mask=mask)
+        y = tl.load(y_ptr + offsets, mask=mask)
+        output = x + y
+        tl.atomic_add(output_ptr + offsets, output, mask=mask)
+
+
+from torch.testing._internal.inductor_utils import GPU_TYPE
+from torch.testing._internal.triton_utils import requires_gpu
+
 
 class FxGraphRunnableArtifactFilter(logging.Filter):
     def filter(self, record):
@@ -100,6 +153,41 @@ def f(x):
         torch.compile(f)(torch.randn(4))
         self._exec_and_verify_payload()
 
+    @unittest.skipUnless(has_triton(), "Triton not available")
+    def test_user_defined_triton_kernel_autotune(self):
+        def add(x: torch.Tensor, y: torch.Tensor) -> torch.Tensor:
+            output = torch.ones(x.shape, device=x.device, dtype=x.dtype)
+            n_elements = output.numel()
+
+            def grid(
+                meta,
+            ):
+                return (triton.cdiv(n_elements, meta["BLOCK_SIZE"]),)
+
+            add_kernel_autotune[grid](x, y, output, n_elements)
+            return output
+
+        x = torch.ones((4096,), device=GPU_TYPE, dtype=torch.float16)
+        y = torch.ones((4096,), device=GPU_TYPE, dtype=torch.float16)
+
+        torch.compile(add)(x, y)
+        self._exec_and_verify_payload()
+
+    @unittest.skipUnless(has_triton(), "Triton not available")
+    @requires_gpu
+    def test_user_defined_triton_kernel(self):
+        def add(x: torch.Tensor, y: torch.Tensor) -> torch.Tensor:
+            output = torch.ones(x.shape, device=x.device, dtype=x.dtype)
+            n_elements = x.numel()
+            add_kernel[n_elements,](x, y, output, n_elements, BLOCK_SIZE=4)
+            return output
+
+        x = torch.ones((4096,), device=GPU_TYPE, dtype=torch.float16)
+        y = torch.ones((4096,), device=GPU_TYPE, dtype=torch.float16)
+
+        torch.compile(add)(x, y)
+        self._exec_and_verify_payload()
+
     def test_two_inputs_matmul(self):
         def f(a, b):
             return (a @ b).relu()
diff --git a/torch/_dynamo/repro/after_aot.py b/torch/_dynamo/repro/after_aot.py
@@ -34,6 +34,24 @@
 from typing import Any, Callable, IO, Optional, TYPE_CHECKING, Union
 from typing_extensions import Unpack
 
+from torch.utils._triton import has_triton
+
+
+if has_triton():
+    from triton.runtime.autotuner import Autotuner, Heuristics
+    from triton.runtime.jit import JITFunction
+else:
+
+    class Autotuner:  # type: ignore[no-redef]
+        pass
+
+    class JITFunction:  # type: ignore[no-redef]
+        pass
+
+    class Heuristics:  # type: ignore[no-redef]
+        pass
+
+
 import torch
 import torch.fx as fx
 import torch.nn as nn
@@ -58,6 +76,7 @@
 )
 from torch._dynamo.utils import clone_inputs, counters, same
 from torch._environment import is_fbcode
+from torch._higher_order_ops.triton_kernel_wrap import kernel_side_table
 from torch._inductor.cpp_builder import normalize_path_separator
 from torch._inductor.output_code import OutputCode
 from torch._library.fake_class_registry import FakeScriptObject
@@ -302,6 +321,16 @@ def generate_compiler_repro_string(
         """
         ).strip()
 
+    triton_imports = ""
+
+    if len(kernel_side_table.id_to_kernel) > 0:
+        triton_imports = textwrap.dedent(
+            """
+import triton
+import triton.language as tl
+        """
+        ).strip()
+
     model_str = textwrap.dedent(
         f"""
 {generate_env_vars_string(stable_output=stable_output)}
@@ -312,6 +341,7 @@ def generate_compiler_repro_string(
 from math import inf
 import torch._inductor.inductor_prims
 {distributed_imports}
+{triton_imports}
 
 {generate_config_string(stable_output=stable_output)}
 
@@ -330,6 +360,53 @@ def generate_compiler_repro_string(
             model_str += f"# torch git version: {torch.version.git_version}\n\n\n"
         model_str += _cuda_system_info_comment()
 
+    kernel_side_table_prefix = (
+        "torch._higher_order_ops.triton_kernel_wrap.kernel_side_table"
+    )
+    # Track which grid entry corresponds to the best config
+    for id in kernel_side_table.id_to_kernel:
+        kernel = kernel_side_table.get_kernel(id)
+
+        if isinstance(kernel, Autotuner):
+            if isinstance(kernel.fn, Heuristics):
+                model_str += "ERROR: Repro will not work as intended, "
+                model_str += (
+                    "triton.runtime.autotuner.Heuristics is not currently supported\n"
+                )
+                break
+
+            config_strs = []
+            for kernel_config in kernel.configs:
+                config_strs.append(f"""triton.Config(
+                        {str(kernel_config.kwargs)},
+                        num_warps={kernel_config.num_warps},
+                        num_stages={kernel_config.num_stages},
+                    )""")
+
+            config_str = ",".join(config_strs)
+            model_str += textwrap.dedent(f"""
+            @triton.autotune(
+                configs=[
+                    {config_str}
+                ],
+                key=[]
+            )
+            """).strip()
+
+        model_str += "\n@triton.jit\n"
+        src_code = kernel.src if isinstance(kernel, JITFunction) else kernel.fn.src
+        fn_name = (
+            kernel._fn_name if isinstance(kernel, JITFunction) else kernel.fn._fn_name
+        )
+        fn_name = fn_name.split(".")[-1]
+
+        model_str += src_code
+        model_str += "\n"
+        model_str += f"{kernel_side_table_prefix}.add_kernel({fn_name})\n"
+
+    if len(kernel_side_table.constant_args) > 0:
+        model_str += f"{kernel_side_table_prefix}.constant_args={kernel_side_table.constant_args}\n"
+
     model_str += NNModuleToString.convert(gm)
 
     writer = InputWriter(save_dir, stable_hash=stable_hash)
