[rfc] add hint_override kwarg to mark_dynamic (#161007)

The motivation for this change can be seen through the following example:

```
import torch

GPU_TYPE = "cuda"

@torch.compile
def no_override(x):
    return x.sum(dim=0)

@torch.compile
def override(x):
    return x.sum(dim=0)

x_small = torch.randn(4096, 512, device=GPU_TYPE)
no_override(x_small)
torch._dynamo.decorators.mark_dynamic(x_small, 0, hint_override=4096 * 1000)
override(x_small)
```

Previously, when reductions were split, codegen relied only on the first observed shape. With a small input, this resulted in a small split size:

```
def triton_red_fused_sum_0(in_ptr0, out_ptr0, ks0, xnumel, r0_numel, XBLOCK : tl.constexpr, R0_BLOCK : tl.constexpr):
    xnumel = 16384
    rnumel = r0_numel
```

With the new scheme, inductor honors hint_override during codegen, producing larger and more appropriate split sizes:

```
def triton_red_fused_sum_0(in_ptr0, out_ptr0, ks0, xnumel, r0_numel, XBLOCK : tl.constexpr, R0_BLOCK : tl.constexpr):
    xnumel = 1024000
    rnumel = r0_numel
```

This addresses a broader problem with dynamism: performance and numerics previously depended on whichever shape was seen first. For example:

```
f(s0) -> f(s2)
f(s1) -> f(s2)
```

could generate different kernels. With the new approach, an explicit override pins the chosen configuration:

```
f(s0, hint_override=s0) -> f(s2)
f(s1, hint_override=s0) -> f(s2)
```

ensuring consistent kernel generation regardless of input order.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/161007
Approved by: https://github.com/jansel

test/dynamo/test_structured_trace.py

@@ -245,7 +245,7 @@ class StructuredTraceTest(TestCase):
             """\
 {"dynamo_start": {"stack": "STACK"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"describe_storage": {"id": 0, "describer_id": "ID", "size": 4000000}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
-{"describe_tensor": {"id": 0, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cuda', index=0)", "size": [1000, 1000], "is_leaf": true, "stride": [1000, 1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
+{"describe_tensor": {"id": 0, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cuda', index=0)", "size": [1000, 1000], "dynamo_hint_overrides": {}, "is_leaf": true, "stride": [1000, 1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"describe_source": {"describer_id": "ID", "id": 0, "source": "L['a']"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"dynamo_output_graph": {"sizes": {"l_a_": [1000, 1000], "ones": [1000, 1000], "output": [1000, 1000]}}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0, "has_payload": "HASH"}
 {"artifact": {"name": "before_pre_grad_graph", "encoding": "string"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0, "has_payload": "HASH"}
@@ -279,7 +279,7 @@ class StructuredTraceTest(TestCase):
             """\
 {"dynamo_start": {"stack": "STACK"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"describe_storage": {"id": 0, "describer_id": "ID", "size": 4000000}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
-{"describe_tensor": {"id": 0, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cuda', index=0)", "size": [1000, 1000], "is_leaf": true, "stride": [1000, 1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
+{"describe_tensor": {"id": 0, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cuda', index=0)", "size": [1000, 1000], "dynamo_hint_overrides": {}, "is_leaf": true, "stride": [1000, 1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"describe_source": {"describer_id": "ID", "id": 0, "source": "L['a']"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"dynamo_output_graph": {"sizes": {"l_a_": [1000, 1000], "ones": [1000, 1000], "output": [1000, 1000]}}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0, "has_payload": "HASH"}
 {"artifact": {"name": "before_pre_grad_graph", "encoding": "string"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0, "has_payload": "HASH"}
@@ -318,10 +318,10 @@ class StructuredTraceTest(TestCase):
             """\
 {"dynamo_start": {"stack": "STACK"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"describe_storage": {"id": 0, "describer_id": "ID", "size": 4000000}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
-{"describe_tensor": {"id": 0, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [1000, 1000], "is_leaf": true, "stride": [1000, 1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
+{"describe_tensor": {"id": 0, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [1000, 1000], "dynamo_hint_overrides": {}, "is_leaf": true, "stride": [1000, 1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"describe_source": {"describer_id": "ID", "id": 0, "source": "L['x']"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"describe_storage": {"id": 1, "describer_id": "ID", "size": 4000000}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
-{"describe_tensor": {"id": 1, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [1000, 1000], "is_leaf": true, "stride": [1000, 1], "storage": 1, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
+{"describe_tensor": {"id": 1, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [1000, 1000], "dynamo_hint_overrides": {}, "is_leaf": true, "stride": [1000, 1], "storage": 1, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"describe_source": {"describer_id": "ID", "id": 1, "source": "L['y']"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"dynamo_output_graph": {"sizes": {"l_x_": [1000, 1000], "l_y_": [1000, 1000], "add": [1000, 1000]}}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0, "has_payload": "HASH"}
 {"artifact": {"name": "before_pre_grad_graph", "encoding": "string"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0, "has_payload": "HASH"}
@@ -342,7 +342,7 @@ class StructuredTraceTest(TestCase):
 {"artifact": {"name": "recompile_reasons", "encoding": "json"}, "frame_id": 0, "frame_compile_id": 1, "attempt": 0, "has_payload": "HASH"}
 {"dynamo_start": {"stack": "STACK"}, "frame_id": 0, "frame_compile_id": 1, "attempt": 0}
 {"describe_storage": {"id": 0, "describer_id": "ID", "size": 4000000}, "frame_id": 0, "frame_compile_id": 1, "attempt": 0}
-{"describe_tensor": {"id": 0, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [1000, 1000], "is_leaf": true, "stride": [1000, 1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 1, "attempt": 0}
+{"describe_tensor": {"id": 0, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [1000, 1000], "dynamo_hint_overrides": {}, "is_leaf": true, "stride": [1000, 1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 1, "attempt": 0}
 {"describe_source": {"describer_id": "ID", "id": 0, "source": "L['x']"}, "frame_id": 0, "frame_compile_id": 1, "attempt": 0}
 {"create_symbol": {"symbol": "s48", "val": "1", "vr": "[-int_oo, int_oo]", "source": "L['y']", "user_stack": "STACK", "stack": "STACK"}, "frame_id": 0, "frame_compile_id": 1, "attempt": 0}
 {"dynamo_output_graph": {"sizes": {"l_x_": [1000, 1000], "add": [1000, 1000]}}, "frame_id": 0, "frame_compile_id": 1, "attempt": 0, "has_payload": "HASH"}
@@ -375,7 +375,7 @@ class StructuredTraceTest(TestCase):
             """\
 {"dynamo_start": {"stack": "STACK"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"describe_storage": {"id": 0, "describer_id": "ID", "size": 4000000}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
-{"describe_tensor": {"id": 0, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [1000, 1000], "is_leaf": true, "stride": [1000, 1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
+{"describe_tensor": {"id": 0, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [1000, 1000], "dynamo_hint_overrides": {}, "is_leaf": true, "stride": [1000, 1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"describe_source": {"describer_id": "ID", "id": 0, "source": "L['a']"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"dynamo_output_graph": {"sizes": {"l_a_": [1000, 1000], "ones": [1000, 1000], "output": [1000, 1000], "ones_1": [1000, 1000], "output_1": [1000, 1000]}}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0, "has_payload": "HASH"}
 {"artifact": {"name": "before_pre_grad_graph", "encoding": "string"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0, "has_payload": "HASH"}
@@ -411,28 +411,28 @@ class StructuredTraceTest(TestCase):
             """\
 {"dynamo_start": {"stack": "STACK"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"describe_storage": {"id": 0, "describer_id": "ID", "size": 4000000}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
-{"describe_tensor": {"id": 0, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [1000, 1000], "is_leaf": true, "requires_grad": true, "stride": [1000, 1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
+{"describe_tensor": {"id": 0, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [1000, 1000], "dynamo_hint_overrides": {}, "is_leaf": true, "requires_grad": true, "stride": [1000, 1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"describe_source": {"describer_id": "ID", "id": 0, "source": "L['a']"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"artifact": {"name": "dynamo_graph_break_reason", "encoding": "string"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0, "has_payload": "HASH"}
 {"describe_storage": {"id": 0, "describer_id": "ID", "size": 4000000}, "frame_id": 0, "frame_compile_id": 0, "attempt": 1}
-{"describe_tensor": {"id": 0, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [1000, 1000], "is_leaf": true, "requires_grad": true, "stride": [1000, 1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 1}
+{"describe_tensor": {"id": 0, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [1000, 1000], "dynamo_hint_overrides": {}, "is_leaf": true, "requires_grad": true, "stride": [1000, 1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 1}
 {"describe_source": {"describer_id": "ID", "id": 0, "source": "L['a']"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 1}
 {"dynamo_cpp_guards_str": {}, "frame_id": 0, "frame_compile_id": 0, "attempt": 1, "has_payload": "HASH"}
 {"compilation_metrics": "METRICS", "frame_id": 0, "frame_compile_id": 0, "attempt": 1}
 {"dynamo_start": {"stack": "STACK"}, "frame_id": 1, "frame_compile_id": 0, "attempt": 0}
 {"artifact": {"name": "dynamo_graph_break_reason", "encoding": "string"}, "frame_id": 1, "frame_compile_id": 0, "attempt": 0, "has_payload": "HASH"}
 {"describe_storage": {"id": 0, "describer_id": "ID", "size": 4000000}, "frame_id": 1, "frame_compile_id": 0, "attempt": 1}
-{"describe_tensor": {"id": 0, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [1000, 1000], "is_leaf": true, "requires_grad": true, "stride": [1000, 1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 1, "frame_compile_id": 0, "attempt": 1}
+{"describe_tensor": {"id": 0, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [1000, 1000], "dynamo_hint_overrides": {}, "is_leaf": true, "requires_grad": true, "stride": [1000, 1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 1, "frame_compile_id": 0, "attempt": 1}
 {"describe_source": {"describer_id": "ID", "id": 0, "source": "L['___stack1']"}, "frame_id": 1, "frame_compile_id": 0, "attempt": 1}
 {"dynamo_cpp_guards_str": {}, "frame_id": 1, "frame_compile_id": 0, "attempt": 1, "has_payload": "HASH"}
 {"compilation_metrics": "METRICS", "frame_id": 1, "frame_compile_id": 0, "attempt": 1}
 {"dynamo_start": {"stack": "STACK"}, "frame_id": 2, "frame_compile_id": 0, "attempt": 0}
 {"describe_storage": {"id": 0, "describer_id": "ID", "size": 4000000}, "frame_id": 2, "frame_compile_id": 0, "attempt": 0}
-{"describe_tensor": {"id": 0, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [1000, 1000], "requires_grad": true, "stride": [1000, 1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 2, "frame_compile_id": 0, "attempt": 0}
+{"describe_tensor": {"id": 0, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [1000, 1000], "dynamo_hint_overrides": {}, "requires_grad": true, "stride": [1000, 1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 2, "frame_compile_id": 0, "attempt": 0}
 {"describe_source": {"describer_id": "ID", "id": 0, "source": "L['___stack0']"}, "frame_id": 2, "frame_compile_id": 0, "attempt": 0}
 {"artifact": {"name": "dynamo_graph_break_reason", "encoding": "string"}, "frame_id": 2, "frame_compile_id": 0, "attempt": 0, "has_payload": "HASH"}
 {"describe_storage": {"id": 0, "describer_id": "ID", "size": 4000000}, "frame_id": 2, "frame_compile_id": 0, "attempt": 1}
-{"describe_tensor": {"id": 0, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [1000, 1000], "requires_grad": true, "stride": [1000, 1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 2, "frame_compile_id": 0, "attempt": 1}
+{"describe_tensor": {"id": 0, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [1000, 1000], "dynamo_hint_overrides": {}, "requires_grad": true, "stride": [1000, 1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 2, "frame_compile_id": 0, "attempt": 1}
 {"describe_source": {"describer_id": "ID", "id": 0, "source": "L['___stack0']"}, "frame_id": 2, "frame_compile_id": 0, "attempt": 1}
 {"dynamo_output_graph": {"sizes": {"l_stack0_": [1000, 1000], "ones": [1000, 1000], "output": [1000, 1000], "sum_1": []}}, "frame_id": 2, "frame_compile_id": 0, "attempt": 1, "has_payload": "HASH"}
 {"artifact": {"name": "before_pre_grad_graph", "encoding": "string"}, "frame_id": 2, "frame_compile_id": 0, "attempt": 1, "has_payload": "HASH"}
@@ -460,7 +460,7 @@ class StructuredTraceTest(TestCase):
 {"bwd_compilation_metrics": "METRICS", "frame_id": 2, "frame_compile_id": 0, "attempt": 1}
 {"dynamo_start": {"stack": "STACK"}, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
 {"describe_storage": {"id": 0, "describer_id": "ID", "size": 4000000}, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
-{"describe_tensor": {"id": 0, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [1000, 1000], "requires_grad": true, "stride": [1000, 1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
+{"describe_tensor": {"id": 0, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [1000, 1000], "dynamo_hint_overrides": {}, "requires_grad": true, "stride": [1000, 1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
 {"describe_source": {"describer_id": "ID", "id": 0, "source": "L['output']"}, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
 {"compilation_metrics": "METRICS", "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
 """,  # noqa: B950
@@ -480,7 +480,7 @@ class StructuredTraceTest(TestCase):
             """\
 {"dynamo_start": {"stack": "STACK"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"describe_storage": {"id": 0, "describer_id": "ID", "size": 4000000}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
-{"describe_tensor": {"id": 0, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [1000, 1000], "is_leaf": true, "requires_grad": true, "stride": [1000, 1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
+{"describe_tensor": {"id": 0, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [1000, 1000], "dynamo_hint_overrides": {}, "is_leaf": true, "requires_grad": true, "stride": [1000, 1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"describe_source": {"describer_id": "ID", "id": 0, "source": "L['a']"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"artifact": {"name": "dynamo_error", "encoding": "string"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0, "has_payload": "HASH"}
 {"compilation_metrics": "METRICS", "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
@@ -514,7 +514,7 @@ class StructuredTraceTest(TestCase):
             """\
 {"dynamo_start": {"stack": "STACK"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"describe_storage": {"id": 0, "describer_id": "ID", "size": 4000000}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
-{"describe_tensor": {"id": 0, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [1000, 1000], "is_leaf": true, "requires_grad": true, "stride": [1000, 1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
+{"describe_tensor": {"id": 0, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [1000, 1000], "dynamo_hint_overrides": {}, "is_leaf": true, "requires_grad": true, "stride": [1000, 1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"describe_source": {"describer_id": "ID", "id": 0, "source": "L['a']"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"dynamo_output_graph": {"sizes": {"l_a_": [1000, 1000], "output": [1000, 1000]}}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0, "has_payload": "HASH"}
 {"artifact": {"name": "before_pre_grad_graph", "encoding": "string"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0, "has_payload": "HASH"}
@@ -625,7 +625,7 @@ class StructuredTraceTest(TestCase):
 {"dynamo_start": {"stack": "STACK"}, "rank": 0, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"artifact": {"name": "dynamo_graph_break_reason", "encoding": "string"}, "rank": 0, "frame_id": 0, "frame_compile_id": 0, "attempt": 0, "has_payload": "HASH"}
 {"describe_storage": {"id": 0, "describer_id": "ID", "size": 4194304}, "rank": 0, "frame_id": 0, "frame_compile_id": 0, "attempt": 1}
-{"describe_tensor": {"id": 0, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cuda', index=0)", "size": [1024, 1024], "is_leaf": true, "stride": [1024, 1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "rank": 0, "frame_id": 0, "frame_compile_id": 0, "attempt": 1}
+{"describe_tensor": {"id": 0, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cuda', index=0)", "size": [1024, 1024], "dynamo_hint_overrides": {}, "is_leaf": true, "stride": [1024, 1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "rank": 0, "frame_id": 0, "frame_compile_id": 0, "attempt": 1}
 {"describe_source": {"describer_id": "ID", "id": 0, "source": "L['args'][0]"}, "rank": 0, "frame_id": 0, "frame_compile_id": 0, "attempt": 1}
 {"dynamo_cpp_guards_str": {}, "rank": 0, "frame_id": 0, "frame_compile_id": 0, "attempt": 1, "has_payload": "HASH"}
 {"compilation_metrics": "METRICS", "rank": 0, "frame_id": 0, "frame_compile_id": 0, "attempt": 1}
@@ -641,32 +641,32 @@ class StructuredTraceTest(TestCase):
 {"compilation_metrics": "METRICS", "rank": 0, "frame_id": 3, "frame_compile_id": 0, "attempt": 0}
 {"dynamo_start": {"stack": "STACK"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
 {"describe_storage": {"id": 0, "describer_id": "ID", "size": 4194304}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
-{"describe_tensor": {"id": 0, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cuda', index=0)", "size": [1024, 1024], "is_leaf": true, "requires_grad": true, "is_parameter": true, "stride": [1024, 1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
+{"describe_tensor": {"id": 0, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cuda', index=0)", "size": [1024, 1024], "dynamo_hint_overrides": {}, "is_leaf": true, "requires_grad": true, "is_parameter": true, "stride": [1024, 1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
 {"describe_source": {"describer_id": "ID", "id": 0, "source": "L['self']._modules['layers']._modules['0']._parameters['weight']"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
 {"describe_storage": {"id": 1, "describer_id": "ID", "size": 4096}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
-{"describe_tensor": {"id": 1, "ndim": 1, "dtype": "torch.float32", "device": "device(type='cuda', index=0)", "size": [1024], "is_leaf": true, "requires_grad": true, "is_parameter": true, "stride": [1], "storage": 1, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
+{"describe_tensor": {"id": 1, "ndim": 1, "dtype": "torch.float32", "device": "device(type='cuda', index=0)", "size": [1024], "dynamo_hint_overrides": {}, "is_leaf": true, "requires_grad": true, "is_parameter": true, "stride": [1], "storage": 1, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
 {"describe_source": {"describer_id": "ID", "id": 1, "source": "L['self']._modules['layers']._modules['0']._parameters['bias']"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
 {"describe_storage": {"id": 2, "describer_id": "ID", "size": 4194304}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
-{"describe_tensor": {"id": 2, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cuda', index=0)", "size": [1024, 1024], "is_leaf": true, "stride": [1024, 1], "storage": 2, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
+{"describe_tensor": {"id": 2, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cuda', index=0)", "size": [1024, 1024], "dynamo_hint_overrides": {}, "is_leaf": true, "stride": [1024, 1], "storage": 2, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
 {"describe_source": {"describer_id": "ID", "id": 2, "source": "L['x']"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
 {"describe_storage": {"id": 3, "describer_id": "ID", "size": 4194304}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
-{"describe_tensor": {"id": 8, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cuda', index=0)", "size": [1024, 1024], "is_leaf": true, "requires_grad": true, "is_parameter": true, "stride": [1024, 1], "storage": 3, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
+{"describe_tensor": {"id": 8, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cuda', index=0)", "size": [1024, 1024], "dynamo_hint_overrides": {}, "is_leaf": true, "requires_grad": true, "is_parameter": true, "stride": [1024, 1], "storage": 3, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
 {"describe_source": {"describer_id": "ID", "id": 8, "source": "L['self']._modules['layers']._modules['1']._parameters['weight']"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
 {"describe_storage": {"id": 4, "describer_id": "ID", "size": 4096}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
-{"describe_tensor": {"id": 9, "ndim": 1, "dtype": "torch.float32", "device": "device(type='cuda', index=0)", "size": [1024], "is_leaf": true, "requires_grad": true, "is_parameter": true, "stride": [1], "storage": 4, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
+{"describe_tensor": {"id": 9, "ndim": 1, "dtype": "torch.float32", "device": "device(type='cuda', index=0)", "size": [1024], "dynamo_hint_overrides": {}, "is_leaf": true, "requires_grad": true, "is_parameter": true, "stride": [1], "storage": 4, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
 {"describe_source": {"describer_id": "ID", "id": 9, "source": "L['self']._modules['layers']._modules['1']._parameters['bias']"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
 {"dynamo_output_graph": {"sizes": {"l_self_modules_layers_modules_0_parameters_weight_": [1024, 1024], "l_self_modules_layers_modules_0_parameters_bias_": [1024], "l_x_": [1024, 1024], "l_self_modules_layers_modules_1_parameters_weight_": [1024, 1024], "l_self_modules_layers_modules_1_parameters_bias_": [1024], "input_1": [1024, 1024], "input_2": [1024, 1024]}}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0, "has_payload": "HASH"}
 {"optimize_ddp_split_graph": {}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0, "has_payload": "HASH"}
 {"optimize_ddp_split_child": {"name": "submod_0"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0, "has_payload": "HASH"}
 {"optimize_ddp_split_child": {"name": "submod_1"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0, "has_payload": "HASH"}
 {"describe_storage": {"id": 0, "describer_id": "ID", "size": 4194304}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
-{"describe_tensor": {"id": 0, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cuda', index=0)", "size": [1024, 1024], "is_leaf": true, "stride": [1024, 1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
+{"describe_tensor": {"id": 0, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cuda', index=0)", "size": [1024, 1024], "dynamo_hint_overrides": {}, "is_leaf": true, "stride": [1024, 1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
 {"describe_source": {"describer_id": "ID", "id": 0, "source": "L['x']"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
 {"describe_storage": {"id": 1, "describer_id": "ID", "size": 4194304}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
-{"describe_tensor": {"id": 1, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cuda', index=0)", "size": [1024, 1024], "is_leaf": true, "requires_grad": true, "is_parameter": true, "stride": [1024, 1], "storage": 1, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
+{"describe_tensor": {"id": 1, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cuda', index=0)", "size": [1024, 1024], "dynamo_hint_overrides": {}, "is_leaf": true, "requires_grad": true, "is_parameter": true, "stride": [1024, 1], "storage": 1, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
 {"describe_source": {"describer_id": "ID", "id": 1, "source": "L['self']._modules['layers']._modules['0']._parameters['weight']"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
 {"describe_storage": {"id": 2, "describer_id": "ID", "size": 4096}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
-{"describe_tensor": {"id": 2, "ndim": 1, "dtype": "torch.float32", "device": "device(type='cuda', index=0)", "size": [1024], "is_leaf": true, "requires_grad": true, "is_parameter": true, "stride": [1], "storage": 2, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
+{"describe_tensor": {"id": 2, "ndim": 1, "dtype": "torch.float32", "device": "device(type='cuda', index=0)", "size": [1024], "dynamo_hint_overrides": {}, "is_leaf": true, "requires_grad": true, "is_parameter": true, "stride": [1], "storage": 2, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
 {"describe_source": {"describer_id": "ID", "id": 2, "source": "L['self']._modules['layers']._modules['0']._parameters['bias']"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
 {"artifact": {"name": "before_pre_grad_graph", "encoding": "string"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0, "has_payload": "HASH"}
 {"artifact": {"name": "after_pre_grad_graph", "encoding": "string"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0, "has_payload": "HASH"}
@@ -682,10 +682,10 @@ class StructuredTraceTest(TestCase):
 {"inductor_output_code": {"filename": "FILENAME"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0, "has_payload": "HASH"}
 {"artifact": {"name": "fx_graph_cache_miss", "encoding": "json"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0, "has_payload": "HASH"}
 {"describe_storage": {"id": 16, "describer_id": "ID", "size": 4194304}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
-{"describe_tensor": {"id": 29, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cuda', index=0)", "size": [1024, 1024], "is_leaf": true, "requires_grad": true, "is_parameter": true, "stride": [1024, 1], "storage": 16, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
+{"describe_tensor": {"id": 29, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cuda', index=0)", "size": [1024, 1024], "dynamo_hint_overrides": {}, "is_leaf": true, "requires_grad": true, "is_parameter": true, "stride": [1024, 1], "storage": 16, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
 {"describe_source": {"describer_id": "ID", "id": 29, "source": "L['self']._modules['layers']._modules['1']._parameters['weight']"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
 {"describe_storage": {"id": 17, "describer_id": "ID", "size": 4096}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
-{"describe_tensor": {"id": 30, "ndim": 1, "dtype": "torch.float32", "device": "device(type='cuda', index=0)", "size": [1024], "is_leaf": true, "requires_grad": true, "is_parameter": true, "stride": [1], "storage": 17, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
+{"describe_tensor": {"id": 30, "ndim": 1, "dtype": "torch.float32", "device": "device(type='cuda', index=0)", "size": [1024], "dynamo_hint_overrides": {}, "is_leaf": true, "requires_grad": true, "is_parameter": true, "stride": [1], "storage": 17, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
 {"describe_source": {"describer_id": "ID", "id": 30, "source": "L['self']._modules['layers']._modules['1']._parameters['bias']"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0}
 {"artifact": {"name": "before_pre_grad_graph", "encoding": "string"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0, "has_payload": "HASH"}
 {"artifact": {"name": "after_pre_grad_graph", "encoding": "string"}, "rank": 0, "frame_id": 4, "frame_compile_id": 0, "attempt": 0, "has_payload": "HASH"}
@@ -725,7 +725,7 @@ class StructuredTraceTest(TestCase):
 {"compilation_metrics": "METRICS", "frame_id": 0, "frame_compile_id": 0, "attempt": 1}
 {"dynamo_start": {"stack": "STACK"}, "frame_id": 1, "frame_compile_id": 0, "attempt": 0}
 {"describe_storage": {"id": 0, "describer_id": "ID", "size": 4}, "frame_id": 1, "frame_compile_id": 0, "attempt": 0}
-{"describe_tensor": {"id": 0, "ndim": 1, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [1], "is_leaf": true, "stride": [1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 1, "frame_compile_id": 0, "attempt": 0}
+{"describe_tensor": {"id": 0, "ndim": 1, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [1], "dynamo_hint_overrides": {}, "is_leaf": true, "stride": [1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 1, "frame_compile_id": 0, "attempt": 0}
 {"describe_source": {"describer_id": "ID", "id": 0, "source": "L['x']"}, "frame_id": 1, "frame_compile_id": 0, "attempt": 0}
 {"dynamo_output_graph": {"sizes": {"l_x_": [1], "add": [1]}}, "frame_id": 1, "frame_compile_id": 0, "attempt": 0, "has_payload": "HASH"}
 {"artifact": {"name": "before_pre_grad_graph", "encoding": "string"}, "frame_id": 1, "frame_compile_id": 0, "attempt": 0, "has_payload": "HASH"}
@@ -766,10 +766,10 @@ class StructuredTraceTest(TestCase):
             """\
 {"dynamo_start": {"stack": "STACK"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"describe_storage": {"id": 0, "describer_id": "ID", "size": 800}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
-{"describe_tensor": {"id": 0, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [10, 20], "is_leaf": true, "stride": [20, 1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
+{"describe_tensor": {"id": 0, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [10, 20], "dynamo_hint_overrides": {}, "is_leaf": true, "stride": [20, 1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"describe_source": {"describer_id": "ID", "id": 0, "source": "L['a']"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"describe_storage": {"id": 1, "describer_id": "ID", "size": 2400}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
-{"describe_tensor": {"id": 1, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [20, 30], "is_leaf": true, "stride": [30, 1], "storage": 1, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
+{"describe_tensor": {"id": 1, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [20, 30], "dynamo_hint_overrides": {}, "is_leaf": true, "stride": [30, 1], "storage": 1, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"describe_source": {"describer_id": "ID", "id": 1, "source": "L['b']"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"dynamo_output_graph": {"sizes": {"l_a_": [10, 20], "l_b_": [20, 30], "matmul": [10, 30]}}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0, "has_payload": "HASH"}
 {"dynamo_cpp_guards_str": {}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0, "has_payload": "HASH"}
@@ -777,12 +777,12 @@ class StructuredTraceTest(TestCase):
 {"artifact": {"name": "recompile_reasons", "encoding": "json"}, "frame_id": 0, "frame_compile_id": 1, "attempt": 0, "has_payload": "HASH"}
 {"dynamo_start": {"stack": "STACK"}, "frame_id": 0, "frame_compile_id": 1, "attempt": 0}
 {"describe_storage": {"id": 0, "describer_id": "ID", "size": 200}, "frame_id": 0, "frame_compile_id": 1, "attempt": 0}
-{"describe_tensor": {"id": 0, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [5, 10], "is_leaf": true, "stride": [10, 1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 1, "attempt": 0}
+{"describe_tensor": {"id": 0, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [5, 10], "dynamo_hint_overrides": {}, "is_leaf": true, "stride": [10, 1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 1, "attempt": 0}
 {"describe_source": {"describer_id": "ID", "id": 0, "source": "L['a']"}, "frame_id": 0, "frame_compile_id": 1, "attempt": 0}
 {"create_symbol": {"symbol": "s97", "val": "5", "vr": "[2, int_oo]", "source": "L['a'].size()[0]", "user_stack": "STACK", "stack": "STACK"}, "frame_id": 0, "frame_compile_id": 1, "attempt": 0}
 {"create_symbol": {"symbol": "s98", "val": "10", "vr": "[2, int_oo]", "source": "L['a'].size()[1]", "user_stack": "STACK", "stack": "STACK"}, "frame_id": 0, "frame_compile_id": 1, "attempt": 0}
 {"describe_storage": {"id": 1, "describer_id": "ID", "size": 600}, "frame_id": 0, "frame_compile_id": 1, "attempt": 0}
-{"describe_tensor": {"id": 1, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [10, 15], "is_leaf": true, "stride": [15, 1], "storage": 1, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 1, "attempt": 0}
+{"describe_tensor": {"id": 1, "ndim": 2, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [10, 15], "dynamo_hint_overrides": {}, "is_leaf": true, "stride": [15, 1], "storage": 1, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 1, "attempt": 0}
 {"describe_source": {"describer_id": "ID", "id": 1, "source": "L['b']"}, "frame_id": 0, "frame_compile_id": 1, "attempt": 0}
 {"create_symbol": {"symbol": "s52", "val": "10", "vr": "[2, int_oo]", "source": "L['b'].size()[0]", "user_stack": "STACK", "stack": "STACK"}, "frame_id": 0, "frame_compile_id": 1, "attempt": 0}
 {"create_symbol": {"symbol": "s20", "val": "15", "vr": "[2, int_oo]", "source": "L['b'].size()[1]", "user_stack": "STACK", "stack": "STACK"}, "frame_id": 0, "frame_compile_id": 1, "attempt": 0}
@@ -818,7 +818,7 @@ class StructuredTraceTest(TestCase):
             """\
 {"dynamo_start": {"stack": "STACK"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"describe_storage": {"id": 0, "describer_id": "ID", "size": 4}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
-{"describe_tensor": {"id": 0, "ndim": 1, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [1], "is_leaf": true, "stride": [1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
+{"describe_tensor": {"id": 0, "ndim": 1, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [1], "dynamo_hint_overrides": {}, "is_leaf": true, "stride": [1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"describe_source": {"describer_id": "ID", "id": 0, "source": "L['x']"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"dynamo_output_graph": {"sizes": {"l_x_": [1], "x": [1]}}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0, "has_payload": "HASH"}
 {"dynamo_cpp_guards_str": {}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0, "has_payload": "HASH"}
@@ -826,7 +826,7 @@ class StructuredTraceTest(TestCase):
 {"artifact": {"name": "recompile_reasons", "encoding": "json"}, "frame_id": 0, "frame_compile_id": 1, "attempt": 0, "has_payload": "HASH"}
 {"dynamo_start": {"stack": "STACK"}, "frame_id": 0, "frame_compile_id": 1, "attempt": 0}
 {"describe_storage": {"id": 0, "describer_id": "ID", "size": 4}, "frame_id": 0, "frame_compile_id": 1, "attempt": 0}
-{"describe_tensor": {"id": 0, "ndim": 1, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [1], "is_leaf": true, "stride": [1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 1, "attempt": 0}
+{"describe_tensor": {"id": 0, "ndim": 1, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [1], "dynamo_hint_overrides": {}, "is_leaf": true, "stride": [1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 1, "attempt": 0}
 {"describe_source": {"describer_id": "ID", "id": 0, "source": "L['x']"}, "frame_id": 0, "frame_compile_id": 1, "attempt": 0}
 {"dynamo_output_graph": {"sizes": {"l_x_": [1], "x": [1]}}, "frame_id": 0, "frame_compile_id": 1, "attempt": 0, "has_payload": "HASH"}
 {"dynamo_cpp_guards_str": {}, "frame_id": 0, "frame_compile_id": 1, "attempt": 0, "has_payload": "HASH"}
@@ -856,10 +856,10 @@ def forward(self, x, y):
     return add
 
 {"describe_storage": {"id": 0, "describer_id": "ID", "size": 12}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
-{"describe_tensor": {"id": 0, "ndim": 1, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [3], "is_leaf": true, "stride": [1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
+{"describe_tensor": {"id": 0, "ndim": 1, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [3], "dynamo_hint_overrides": {}, "is_leaf": true, "stride": [1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"describe_source": {"describer_id": "ID", "id": 0, "source": "L['x']"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"describe_storage": {"id": 1, "describer_id": "ID", "size": 12}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
-{"describe_tensor": {"id": 1, "ndim": 1, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [3], "is_leaf": true, "stride": [1], "storage": 1, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
+{"describe_tensor": {"id": 1, "ndim": 1, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [3], "dynamo_hint_overrides": {}, "is_leaf": true, "stride": [1], "storage": 1, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"describe_source": {"describer_id": "ID", "id": 1, "source": "L['y']"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"dynamo_output_graph": {"sizes": {"l_x_": [3], "l_y_": [3], "add": [3]}}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0, "has_payload": "HASH"}
 {"dynamo_cpp_guards_str": {}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0, "has_payload": "HASH"}
@@ -886,7 +886,7 @@ def forward(self, x, y):
             """\
 {"dynamo_start": {"stack": "STACK"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"describe_storage": {"id": 0, "describer_id": "ID", "size": 4}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
-{"describe_tensor": {"id": 0, "ndim": 1, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [1], "is_leaf": true, "stride": [1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
+{"describe_tensor": {"id": 0, "ndim": 1, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [1], "dynamo_hint_overrides": {}, "is_leaf": true, "stride": [1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"describe_source": {"describer_id": "ID", "id": 0, "source": "L['a']"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"dynamo_output_graph": {"sizes": {"l_a_": [1], "sin": [1]}}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0, "has_payload": "HASH"}
 {"artifact": {"name": "before_pre_grad_graph", "encoding": "string"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0, "has_payload": "HASH"}
@@ -906,7 +906,7 @@ def forward(self, x, y):
 {"compilation_metrics": "METRICS", "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"dynamo_start": {"stack": "STACK"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"describe_storage": {"id": 0, "describer_id": "ID", "size": 4}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
-{"describe_tensor": {"id": 0, "ndim": 1, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [1], "is_leaf": true, "stride": [1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
+{"describe_tensor": {"id": 0, "ndim": 1, "dtype": "torch.float32", "device": "device(type='cpu')", "size": [1], "dynamo_hint_overrides": {}, "is_leaf": true, "stride": [1], "storage": 0, "view_func": "VIEW_FUNC", "describer_id": "ID"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"describe_source": {"describer_id": "ID", "id": 0, "source": "L['a']"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0}
 {"dynamo_output_graph": {"sizes": {"l_a_": [1], "sin": [1]}}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0, "has_payload": "HASH"}
 {"artifact": {"name": "before_pre_grad_graph", "encoding": "string"}, "frame_id": 0, "frame_compile_id": 0, "attempt": 0, "has_payload": "HASH"}

test/inductor/test_torchinductor.py

@@ -10585,6 +10585,33 @@ def forward(self, arg0_1: "Sym(s77)", arg1_1: "Sym(s27)", arg2_1: "Sym(s53)", ar
         dynamic_specialized = inductor_matmul(dynamic_specialized_a, b)
         self.assertEqual(dynamic, dynamic_specialized)
 
+    @requires_gpu()
+    @skip_if_not_triton
+    @unittest.skipIf(
+        not IS_BIG_GPU, "Skipping triton backend only since not big GPU (not enough SM)"
+    )
+    @config.patch({"force_disable_caches": True})
+    def test_mark_dynamic_with_hint_override(self):
+        @torch.compile
+        def no_override(x):
+            return x.sum(dim=0)
+
+        @torch.compile
+        def override(x):
+            return x.sum(dim=0)
+
+        x_small = torch.randn(4096, 512, device=GPU_TYPE)
+        torch._dynamo.decorators.mark_dynamic(x_small, 0)
+        code1 = run_and_get_triton_code(no_override, x_small)
+
+        torch._dynamo.reset_code_caches()
+
+        torch._dynamo.decorators.mark_dynamic(x_small, 0, hint_override=4096 * 10)
+        code2 = run_and_get_triton_code(override, x_small)
+        self.assertNotEqual(code1, code2)
+
+        self.assertEqual(no_override(x_small), override(x_small))
+
     @requires_gpu()
     def test_stride_preservation_with_stride_modifying_fx_pass(self):
         def f(x):

torch/_dynamo/decorators.py

@@ -585,6 +585,7 @@ def mark_dynamic(
     t: Any,
     index: Union[int, list[Any], tuple[Any]],
     *,
+    hint_override: Optional[int] = None,
     min: Optional[int] = None,
     max: Optional[int] = None,
     specialize_on: Optional[list[Any]] = None,
@@ -637,10 +638,13 @@ def mark_dynamic(
         if not hasattr(t, "_dynamo_dynamic_indices"):
             t._dynamo_dynamic_indices = set()
             t._dynamo_dynamic_range = set()
+            t._dynamo_hint_overrides = {}
 
         if not hasattr(t, "_specialize_on"):
             t._specialize_on = {}
 
+        if hint_override:
+            t._dynamo_hint_overrides[index] = hint_override
         # TODO(voz): Should we bounds check?
         t._dynamo_dynamic_indices.add(index)
         t._dynamo_dynamic_range.add(_DimRange(index, min, max))  # type: ignore[arg-type]

torch/_subclasses/meta_utils.py

@@ -417,6 +417,7 @@ class MetaTensorDescriber:
             stride=stride,
             storage_offset=storage_offset,
             dynamo_dynamic_indices=list(getattr(t, "_dynamo_dynamic_indices", set())),
+            dynamo_hint_overrides=getattr(t, "_dynamo_hint_overrides", {}),
             sparse_dim=(
                 t.sparse_dim() if t.is_sparse or is_sparse_compressed(t) else None
             ),
@@ -614,6 +615,7 @@ class MetaTensorDesc(Generic[_TensorT]):
     # defined on NJT
     size: tuple[int, ...]
     dynamo_dynamic_indices: list[int]
+    dynamo_hint_overrides: dict[int, int]
 
     layout: torch.layout = torch.strided
     is_inference: bool = False
@@ -956,6 +958,7 @@ class MetaConverter(Generic[_TensorT]):
                         [d in t.dynamo_dynamic_indices for d in range(t.ndim)],
                         src,
                         symbolic_context=symbolic_context,
+                        hint_overrides=t.dynamo_hint_overrides,
                     )
             else:
                 return (t.size, t.stride, t.storage_offset)

torch/fx/experimental/symbolic_shapes.py

@@ -4370,19 +4370,23 @@ class ShapeEnv:
         tensor_size: Sequence[IntLikeType],
         source: Source,
         symbolic_context: SymbolicContext,
+        hint_overrides: Optional[dict[int, int]] = None,
     ) -> list[sympy.Expr]:
         assert all(not is_symbolic(val) for val in tensor_size), (
             f"Expect size to be a plain tuple of ints but got {tensor_size}"
         )
         from torch._dynamo.source import TensorProperty, TensorPropertySource
 
+        if not hint_overrides:
+            hint_overrides = {}
+
         _assert_symbol_context(symbolic_context)
         dynamic_dims = symbolic_context.dynamic_sizes  # type: ignore[attr-defined]
         constraint_dims = symbolic_context.constraint_sizes  # type: ignore[attr-defined]
         size = []
         for i, val in enumerate(tensor_size):
             sym = self.create_symbol(
-                val,
+                val if i not in hint_overrides else hint_overrides[i],
                 TensorPropertySource(source, TensorProperty.SIZE, i),
                 dynamic_dims[i],
                 constraint_dims[i],
@@ -4502,6 +4506,7 @@ class ShapeEnv:
         source: Source,
         *,
         symbolic_context: Optional[SymbolicContext] = None,
+        hint_overrides: Optional[dict[int, int]] = None,
     ) -> tuple[
         tuple[IntLikeType, ...],
         tuple[IntLikeType, ...],
@@ -4509,6 +4514,9 @@ class ShapeEnv:
     ]:
         dim = len(ex_size)
 
+        if not hint_overrides:
+            hint_overrides = {}
+
         # Reimplement the legacy behavior
         if symbolic_context is None:
             constraint_sizes: list[DimConstraint] = [None] * dim
@@ -4563,7 +4571,7 @@ class ShapeEnv:
         from torch._dynamo.source import TensorProperty, TensorPropertySource
 
         size: list[sympy.Expr] = self._produce_dyn_sizes_from_int_tuple(
-            ex_size, source, symbolic_context
+            ex_size, source, symbolic_context, hint_overrides=hint_overrides
         )
         stride = self._compute_symbolic_stride(
             source,
@@ -4579,7 +4587,7 @@ class ShapeEnv:
         sym_sizes = [
             self.create_symintnode(
                 sym,
-                hint=hint,
+                hint=hint if i not in hint_overrides else hint_overrides[i],
                 source=TensorPropertySource(source, TensorProperty.SIZE, i),
             )
             for i, (sym, hint) in enumerate(zip(size, ex_size))