Add files using upload-large-folder tool

lib/python3.12/site-packages/torch/include/ATen/ATen.h

@@ -0,0 +1,37 @@
+#pragma once
+
+#if !defined(_MSC_VER) && __cplusplus < 201703L
+#error C++17 or later compatible compiler is required to use ATen.
+#endif
+
+#include <ATen/Context.h>
+#include <ATen/Device.h>
+#include <ATen/DeviceGuard.h>
+#include <ATen/DimVector.h>
+#include <ATen/Dispatch.h>
+#include <ATen/Formatting.h>
+#include <ATen/Functions.h>
+#include <ATen/NamedTensor.h>
+#include <ATen/ScalarOps.h>
+#include <ATen/Tensor.h>
+#include <ATen/TensorGeometry.h>
+#include <ATen/TensorIndexing.h>
+#include <ATen/TensorOperators.h>
+#include <ATen/Version.h>
+#include <ATen/core/ATenGeneral.h>
+#include <ATen/core/Generator.h>
+#include <ATen/core/Reduction.h>
+#include <ATen/core/Scalar.h>
+#include <ATen/core/UnsafeFromTH.h>
+#include <ATen/core/ivalue.h>
+#include <ATen/core/jit_type.h>
+#include <c10/core/Allocator.h>
+#include <c10/core/InferenceMode.h>
+#include <c10/core/Layout.h>
+#include <c10/core/Storage.h>
+#include <c10/core/TensorOptions.h>
+#include <c10/util/Exception.h>
+
+// TODO: try to remove this
+// There is some back story, see https://github.com/pytorch/pytorch/issues/48684
+#include <ATen/NativeFunctions.h>

lib/python3.12/site-packages/torch/include/ATen/ArrayRef.h

@@ -0,0 +1,2 @@
+#pragma once
+#include <c10/util/ArrayRef.h>

lib/python3.12/site-packages/torch/include/ATen/Backend.h

@@ -0,0 +1,2 @@
+#pragma once
+#include <c10/core/Backend.h>

lib/python3.12/site-packages/torch/include/ATen/Backtrace.h

@@ -0,0 +1,2 @@
+#pragma once
+#include <ATen/core/Backtrace.h>

lib/python3.12/site-packages/torch/include/ATen/BlasBackend.h

@@ -0,0 +1,29 @@
+#pragma once
+
+#include <c10/util/Exception.h>
+
+#include <ostream>
+#include <string>
+
+namespace at {
+
+enum class BlasBackend : int8_t { Cublas, Cublaslt, Ck };
+
+inline std::string BlasBackendToString(at::BlasBackend backend) {
+  switch (backend) {
+    case BlasBackend::Cublas:
+      return "at::BlasBackend::Cublas";
+    case BlasBackend::Cublaslt:
+      return "at::BlasBackend::Cublaslt";
+    case BlasBackend::Ck:
+      return "at::BlasBackend::Ck";
+    default:
+      TORCH_CHECK(false, "Unknown blas backend");
+  }
+}
+
+inline std::ostream& operator<<(std::ostream& stream, at::BlasBackend backend) {
+  return stream << BlasBackendToString(backend);
+}
+
+} // namespace at

lib/python3.12/site-packages/torch/include/ATen/CPUApplyUtils.h

@@ -0,0 +1,352 @@
+#pragma once
+
+#include <ATen/CollapseDims.h>
+#include <ATen/Parallel.h>
+#include <ATen/TensorUtils.h>
+#include <c10/util/irange.h>
+#include <cstring>
+#include <limits>
+
+namespace at {
+
+/*
+ * The basic strategy for apply is as follows:
+ *
+ * 1. Starting with the outermost index, loop until we reach a dimension where
+ * the data is no longer contiguous, i.e. the stride at that dimension is not
+ * equal to the size of the tensor defined by the outer dimensions. Let's call
+ * this outer (contiguous) tensor A. Note that if the Tensor is contiguous, then
+ * A is equal to the entire Tensor. Let's call the inner tensor B.
+ *
+ * 2. We loop through the indices in B, starting at its outermost dimension. For
+ * example, if B is a 2x2 matrix, then we do:
+ *
+ * B[0][0]
+ * B[0][1]
+ * B[1][0]
+ * B[1][1]
+ *
+ * We set the offset into the underlying storage as (storageOffset + stride_B *
+ * index_B), i.e. basically we compute the offset into the storage as we would
+ * normally for a Tensor. But because we are guaranteed the subsequent data is
+ * contiguous in memory, we can simply loop for sizeof(A) iterations and perform
+ * the operation, without having to follow the order described by the strides of
+ * A.
+ *
+ * 3. As an optimization, we merge dimensions of A that are contiguous in
+ * memory. For example, if A is a 3x3x3x3 tensor narrowed from a 3x3x4x3 tensor,
+ * then the first two dimensions can be merged for the purposes of APPLY,
+ * reducing the number of nested loops.
+ */
+
+inline Tensor sort_strides(Tensor& tensor_) {
+  IntArrayRef strides = tensor_.strides();
+  std::vector<int64_t> indices;
+  indices.reserve(tensor_.ndimension());
+  for (const auto i : c10::irange(tensor_.ndimension())) {
+    indices.push_back(i);
+  }
+  std::sort(indices.begin(), indices.end(), [&strides](int64_t i1, int64_t i2) {
+    return strides[i1] > strides[i2];
+  });
+  Tensor tensor = tensor_.permute(indices);
+  return tensor;
+}
+
+template <typename T, int N>
+struct strided_tensor_iter_fixed {
+ public:
+  T* data_ = NULL;
+  int64_t dim_ = 0;
+
+  // NOLINTNEXTLINE(*array*)
+  int64_t counter_[N] = {0};
+  // NOLINTNEXTLINE(*array*)
+  int64_t sizes_[N] = {0};
+  // NOLINTNEXTLINE(*array*)
+  int64_t strides_[N] = {0};
+
+  strided_tensor_iter_fixed(strided_tensor_iter_fixed const&) = delete;
+  strided_tensor_iter_fixed& operator=(strided_tensor_iter_fixed const& x) =
+      delete;
+  strided_tensor_iter_fixed(strided_tensor_iter_fixed&&) noexcept = default;
+  strided_tensor_iter_fixed& operator=(strided_tensor_iter_fixed&& x) noexcept =
+      default;
+  ~strided_tensor_iter_fixed() noexcept = default;
+  strided_tensor_iter_fixed(
+      Tensor& tensor,
+      [[maybe_unused]] bool sort_strides = false)
+      : data_(tensor.data_ptr<T>()) {
+    std::memset(counter_, 0, sizeof(int64_t) * N);
+    if (tensor.dim() > 0) {
+      std::memcpy(
+          sizes_, tensor.sizes().data(), tensor.dim() * sizeof(int64_t));
+      std::memcpy(
+          strides_, tensor.strides().data(), tensor.dim() * sizeof(int64_t));
+    }
+    dim_ = std::get<1>(collapse_dims(sizes_, strides_, tensor.ndimension()));
+  }
+};
+
+template <typename T>
+struct strided_tensor_iter {
+ private:
+ public:
+  T* data_ = NULL;
+  int64_t dim_;
+
+  std::vector<int64_t> counter_;
+  std::vector<int64_t> sizes_;
+  std::vector<int64_t> strides_;
+
+  strided_tensor_iter(strided_tensor_iter const&) = delete;
+  strided_tensor_iter& operator=(strided_tensor_iter const& x) = delete;
+  strided_tensor_iter(strided_tensor_iter&&) noexcept = default;
+  strided_tensor_iter& operator=(strided_tensor_iter&&) noexcept = default;
+  ~strided_tensor_iter() noexcept = default;
+  strided_tensor_iter(Tensor& tensor)
+      : data_(tensor.data_ptr<T>()),
+        dim_(tensor.ndimension()),
+        counter_(dim_, 0),
+        sizes_(tensor.sizes().vec()),
+        strides_(tensor.strides().vec()) {
+    dim_ = std::get<1>(collapse_dims(sizes_.data(), strides_.data(), dim_));
+  }
+};
+
+inline bool _all_equal_numel(at::ArrayRef<Tensor> tensors) {
+  if (tensors.empty())
+    return true;
+  int64_t all_numel = tensors[0].numel();
+  for (const auto i : c10::irange(1, tensors.size())) {
+    if (tensors[i].numel() != all_numel)
+      return false;
+  }
+  return true;
+}
+
+inline std::string _all_equal_numel_error(at::ArrayRef<Tensor> tensors) {
+  std::ostringstream oss;
+  oss << "inconsistent tensor size, expected ";
+  for (size_t i = 0; i < tensors.size() - 1; i++) {
+    oss << tensors[i].sizes() << ", ";
+  }
+  oss << "and " << tensors[tensors.size() - 1].sizes()
+      << " to have the same number of elements, but got ";
+  for (size_t i = 0; i < tensors.size() - 1; i++) {
+    oss << tensors[i].numel() << ", ";
+  }
+  oss << "and " << tensors[tensors.size() - 1].numel()
+      << " elements respectively";
+  return oss.str();
+}
+
+inline bool _apply_preamble(ArrayRef<Tensor> tensors) {
+  checkDeviceType("CPU_tensor_apply", tensors, kCPU);
+  checkLayout("CPU_tensor_apply", tensors, kStrided);
+  if (!_all_equal_numel(tensors))
+    TORCH_CHECK(false, _all_equal_numel_error(tensors));
+  // An empty tensor has no elements
+  for (auto& t : tensors)
+    if (t.numel() == 0)
+      return false;
+  return true;
+}
+
+inline int64_t _max_dim_tensors(ArrayRef<Tensor> tensors) {
+  int64_t dim = 0;
+  for (auto& t : tensors)
+    dim = std::max(dim, t.ndimension());
+  return dim;
+}
+
+inline void iterate(int64_t /*size*/) {}
+
+template <typename Arg, typename... Args>
+inline void iterate(int64_t size, Arg& iter, Args&... iter_tail) {
+  iter.counter_[iter.dim_ - 1] += size;
+  iter.data_ = iter.data_ + size * iter.strides_[iter.dim_ - 1];
+  iterate(size, iter_tail...);
+}
+
+inline bool iterate_continue() {
+  return true;
+}
+
+template <typename Arg, typename... Args>
+inline bool iterate_continue(Arg& iter, Args&... iter_tail) {
+  return iter.counter_[iter.dim_ - 1] < iter.sizes_[iter.dim_ - 1] &&
+      iterate_continue(iter_tail...);
+}
+
+inline int64_t max_iterate_size() {
+  return std::numeric_limits<int64_t>::max();
+}
+
+template <typename Arg, typename... Args>
+inline int64_t max_iterate_size(Arg& iter, Args&... iter_tail) {
+  return std::min(
+      (iter.sizes_[iter.dim_ - 1] - iter.counter_[iter.dim_ - 1]),
+      max_iterate_size(iter_tail...));
+}
+
+inline void iterate_overflow() {}
+
+template <typename Arg, typename... Args>
+inline void iterate_overflow(Arg& iter, Args&... iter_tail) {
+  if (iter.counter_[iter.dim_ - 1] == iter.sizes_[iter.dim_ - 1]) {
+    for (int64_t i = iter.dim_ - 1; i > 0; i--) {
+      if (iter.counter_[i] == iter.sizes_[i]) {
+        iter.counter_[i] = 0;
+        iter.counter_[i - 1]++;
+        iter.data_ = iter.data_ - (iter.sizes_[i] * iter.strides_[i]) +
+            iter.strides_[i - 1];
+      }
+    }
+  }
+  iterate_overflow(iter_tail...);
+}
+
+inline void forward(int64_t /*offset*/) {}
+
+template <typename Arg, typename... Args>
+inline void forward(int64_t offset, Arg& iter, Args&... iter_tail) {
+  int64_t multi = offset;
+  for (int64_t i = iter.dim_ - 1; i >= 0; i--) {
+    int64_t inc = multi % iter.sizes_[i];
+    multi = multi / iter.sizes_[i];
+    iter.data_ = iter.data_ + inc * iter.strides_[i];
+    iter.counter_[i] += inc;
+  }
+  forward(offset, iter_tail...);
+}
+
+inline int64_t max_dim() {
+  return 0;
+}
+
+template <typename Arg, typename... Args>
+inline int64_t max_dim(Arg& iter, Args&... iter_tail) {
+  return std::max(iter.dim_, max_dim(iter_tail...));
+}
+
+inline void apply_op() {}
+
+template <typename Op, typename... Args>
+inline void apply_op(
+    int64_t numel,
+    int64_t offset,
+    const Op& op,
+    Args... iters) {
+  // For 0-dim tensors
+  if (numel == 1 && max_dim(iters...) == 0) {
+    op(*iters.data_...);
+    return;
+  }
+  if (offset > 0)
+    forward(offset, iters...);
+  // Splitting this into chunks helps the compiler create faster assembly
+  for (int64_t i = 0; i < numel;) {
+    for (; iterate_continue(iters...) && i < numel;) {
+      op(*iters.data_...);
+      iterate(1, iters...);
+      i++;
+    }
+    iterate_overflow(iters...);
+  }
+}
+
+/*
+  Apply a pointwise operator to sequence of tensors
+
+  The calling convention for op is a function/functor that takes the same
+  number of pointers of type scalar as the number of given tensors. For example,
+  to compute a = b * c, op would be of the form:
+  [](scalar* a_val, const scalar* b_val, const scalar* c_val) { a_val[0] =
+  b_val[0] * c_val[0]; };
+*/
+
+template <typename scalar1, typename scalar2, typename Op>
+inline void CPU_tensor_apply2(Tensor tensor1, Tensor tensor2, const Op op) {
+  if (!_apply_preamble({tensor1, tensor2}))
+    return;
+  if (_max_dim_tensors({tensor1, tensor2}) <= 8) {
+    apply_op(
+        tensor1.numel(),
+        0,
+        op,
+        strided_tensor_iter_fixed<scalar1, 8>(tensor1),
+        strided_tensor_iter_fixed<scalar2, 8>(tensor2));
+  } else {
+    apply_op(
+        tensor1.numel(),
+        0,
+        op,
+        strided_tensor_iter<scalar1>(tensor1),
+        strided_tensor_iter<scalar2>(tensor2));
+  }
+}
+
+template <typename scalar1, typename scalar2, typename scalar3, typename Op>
+inline void CPU_tensor_apply3(
+    Tensor tensor1,
+    Tensor tensor2,
+    Tensor tensor3,
+    const Op op) {
+  if (!_apply_preamble({tensor1, tensor2, tensor3}))
+    return;
+  if (_max_dim_tensors({tensor1, tensor2, tensor3}) <= 8) {
+    apply_op(
+        tensor1.numel(),
+        0,
+        op,
+        strided_tensor_iter_fixed<scalar1, 8>(tensor1),
+        strided_tensor_iter_fixed<scalar2, 8>(tensor2),
+        strided_tensor_iter_fixed<scalar3, 8>(tensor3));
+  } else {
+    apply_op(
+        tensor1.numel(),
+        0,
+        op,
+        strided_tensor_iter<scalar1>(tensor1),
+        strided_tensor_iter<scalar2>(tensor2),
+        strided_tensor_iter<scalar3>(tensor3));
+  }
+}
+
+template <
+    typename scalar1,
+    typename scalar2,
+    typename scalar3,
+    typename scalar4,
+    typename Op>
+inline void CPU_tensor_apply4(
+    Tensor tensor1,
+    Tensor tensor2,
+    Tensor tensor3,
+    Tensor tensor4,
+    const Op op) {
+  if (!_apply_preamble({tensor1, tensor2, tensor3, tensor4}))
+    return;
+  if (_max_dim_tensors({tensor1, tensor2, tensor3, tensor4}) <= 8) {
+    apply_op(
+        tensor1.numel(),
+        0,
+        op,
+        strided_tensor_iter_fixed<scalar1, 8>(tensor1),
+        strided_tensor_iter_fixed<scalar2, 8>(tensor2),
+        strided_tensor_iter_fixed<scalar3, 8>(tensor3),
+        strided_tensor_iter_fixed<scalar4, 8>(tensor4));
+  } else {
+    apply_op(
+        tensor1.numel(),
+        0,
+        op,
+        strided_tensor_iter<scalar1>(tensor1),
+        strided_tensor_iter<scalar2>(tensor2),
+        strided_tensor_iter<scalar3>(tensor3),
+        strided_tensor_iter<scalar4>(tensor4));
+  }
+}
+
+} // namespace at

lib/python3.12/site-packages/torch/include/ATen/CPUFixedAllocator.h

@@ -0,0 +1,33 @@
+#pragma once
+
+#include <c10/core/Allocator.h>
+#include <c10/util/Exception.h>
+
+// This file creates a fake allocator that just throws exceptions if
+// it is actually used.
+
+// state passed to the allocator is the std::function<void(void*)> called
+// when the blob is release by ATen
+
+namespace at {
+
+static void* cpu_fixed_malloc(void*, ptrdiff_t) {
+  TORCH_CHECK(false, "attempting to resize a tensor view of an external blob");
+}
+
+static void* cpu_fixed_realloc(void*, void*, ptrdiff_t) {
+  TORCH_CHECK(false, "attempting to resize a tensor view of an external blob");
+}
+
+static void cpu_fixed_free(void* state, void* allocation) {
+  auto on_release = static_cast<std::function<void(void*)>*>(state);
+  (*on_release)(allocation);
+  delete on_release;
+}
+
+static Allocator CPU_fixed_allocator = {
+    cpu_fixed_malloc,
+    cpu_fixed_realloc,
+    cpu_fixed_free};
+
+} // namespace at

lib/python3.12/site-packages/torch/include/ATen/CPUFunctions_inl.h

@@ -0,0 +1,540 @@
+#pragma once
+// @generated by torchgen/gen.py from DispatchKeyFunctions_inl.h
+
+// NB: The implementing C++ file is RegisterDispatchKey.cpp
+
+// The only #includes we need are for custom classes that have defaults in the C++ API
+#include <c10/core/MemoryFormat.h>
+#include <c10/core/Scalar.h>
+#include <ATen/core/Reduction.h>
+
+#if defined(AT_PER_OPERATOR_HEADERS) && defined(TORCH_ASSERT_ONLY_METHOD_OPERATORS)
+#error This change adds a dependency on all pytorch operators, meaning the     \
+  file will need to be re-compiled every time an operator is changed or added. \
+  Consider including a specific operator from                                  \
+  <ATen/ops/{my_operator}_cpu_dispatch.h>.                   \
+  See NOTE [TORCH_ASSERT_ONLY_METHOD_OPERATORS].
+#endif
+
+#include <ATen/ops/_adaptive_avg_pool2d_cpu_dispatch.h>
+#include <ATen/ops/_adaptive_avg_pool2d_backward_cpu_dispatch.h>
+#include <ATen/ops/_adaptive_avg_pool3d_cpu_dispatch.h>
+#include <ATen/ops/_adaptive_avg_pool3d_backward_cpu_dispatch.h>
+#include <ATen/ops/_add_relu_cpu_dispatch.h>
+#include <ATen/ops/_addmm_activation_cpu_dispatch.h>
+#include <ATen/ops/_aminmax_cpu_dispatch.h>
+#include <ATen/ops/_amp_foreach_non_finite_check_and_unscale_cpu_dispatch.h>
+#include <ATen/ops/_amp_update_scale_cpu_dispatch.h>
+#include <ATen/ops/_assert_async_cpu_dispatch.h>
+#include <ATen/ops/_batch_norm_with_update_cpu_dispatch.h>
+#include <ATen/ops/_cdist_backward_cpu_dispatch.h>
+#include <ATen/ops/_cdist_forward_cpu_dispatch.h>
+#include <ATen/ops/_cholesky_solve_helper_cpu_dispatch.h>
+#include <ATen/ops/_compute_linear_combination_cpu_dispatch.h>
+#include <ATen/ops/_convert_indices_from_coo_to_csr_cpu_dispatch.h>
+#include <ATen/ops/_convert_indices_from_csr_to_coo_cpu_dispatch.h>
+#include <ATen/ops/_convert_weight_to_int4pack_for_cpu_cpu_dispatch.h>
+#include <ATen/ops/_ctc_loss_cpu_dispatch.h>
+#include <ATen/ops/_ctc_loss_backward_cpu_dispatch.h>
+#include <ATen/ops/_cummax_helper_cpu_dispatch.h>
+#include <ATen/ops/_cummin_helper_cpu_dispatch.h>
+#include <ATen/ops/_dirichlet_grad_cpu_dispatch.h>
+#include <ATen/ops/_efficientzerotensor_cpu_dispatch.h>
+#include <ATen/ops/_embedding_bag_cpu_dispatch.h>
+#include <ATen/ops/_embedding_bag_backward_cpu_dispatch.h>
+#include <ATen/ops/_embedding_bag_dense_backward_cpu_dispatch.h>
+#include <ATen/ops/_embedding_bag_forward_only_cpu_dispatch.h>
+#include <ATen/ops/_embedding_bag_per_sample_weights_backward_cpu_dispatch.h>
+#include <ATen/ops/_empty_affine_quantized_cpu_dispatch.h>
+#include <ATen/ops/_empty_per_channel_affine_quantized_cpu_dispatch.h>
+#include <ATen/ops/_fake_quantize_learnable_per_channel_affine_cpu_dispatch.h>
+#include <ATen/ops/_fake_quantize_learnable_per_channel_affine_backward_cpu_dispatch.h>
+#include <ATen/ops/_fake_quantize_learnable_per_tensor_affine_cpu_dispatch.h>
+#include <ATen/ops/_fake_quantize_learnable_per_tensor_affine_backward_cpu_dispatch.h>
+#include <ATen/ops/_fake_quantize_per_tensor_affine_cachemask_tensor_qparams_cpu_dispatch.h>
+#include <ATen/ops/_fft_c2c_cpu_dispatch.h>
+#include <ATen/ops/_fft_c2r_cpu_dispatch.h>
+#include <ATen/ops/_fft_r2c_cpu_dispatch.h>
+#include <ATen/ops/_foobar_cpu_dispatch.h>
+#include <ATen/ops/_functional_assert_async_cpu_dispatch.h>
+#include <ATen/ops/_fused_adagrad_cpu_dispatch.h>
+#include <ATen/ops/_fused_adam_cpu_dispatch.h>
+#include <ATen/ops/_fused_adamw_cpu_dispatch.h>
+#include <ATen/ops/_fused_moving_avg_obs_fq_helper_cpu_dispatch.h>
+#include <ATen/ops/_fused_sdp_choice_cpu_dispatch.h>
+#include <ATen/ops/_fused_sgd_cpu_dispatch.h>
+#include <ATen/ops/_histogramdd_bin_edges_cpu_dispatch.h>
+#include <ATen/ops/_histogramdd_from_bin_cts_cpu_dispatch.h>
+#include <ATen/ops/_histogramdd_from_bin_tensors_cpu_dispatch.h>
+#include <ATen/ops/_index_put_impl_cpu_dispatch.h>
+#include <ATen/ops/_int_mm_cpu_dispatch.h>
+#include <ATen/ops/_jagged_to_padded_dense_forward_cpu_dispatch.h>
+#include <ATen/ops/_linalg_det_cpu_dispatch.h>
+#include <ATen/ops/_linalg_eigh_cpu_dispatch.h>
+#include <ATen/ops/_linalg_eigvals_cpu_dispatch.h>
+#include <ATen/ops/_linalg_slogdet_cpu_dispatch.h>
+#include <ATen/ops/_linalg_solve_ex_cpu_dispatch.h>
+#include <ATen/ops/_linalg_svd_cpu_dispatch.h>
+#include <ATen/ops/_local_scalar_dense_cpu_dispatch.h>
+#include <ATen/ops/_log_softmax_cpu_dispatch.h>
+#include <ATen/ops/_log_softmax_backward_data_cpu_dispatch.h>
+#include <ATen/ops/_logcumsumexp_cpu_dispatch.h>
+#include <ATen/ops/_make_dep_token_cpu_dispatch.h>
+#include <ATen/ops/_make_per_channel_quantized_tensor_cpu_dispatch.h>
+#include <ATen/ops/_make_per_tensor_quantized_tensor_cpu_dispatch.h>
+#include <ATen/ops/_masked_softmax_cpu_dispatch.h>
+#include <ATen/ops/_masked_softmax_backward_cpu_dispatch.h>
+#include <ATen/ops/_native_batch_norm_legit_cpu_dispatch.h>
+#include <ATen/ops/_native_multi_head_attention_cpu_dispatch.h>
+#include <ATen/ops/_nested_compute_contiguous_strides_offsets_cpu_dispatch.h>
+#include <ATen/ops/_nested_from_padded_cpu_dispatch.h>
+#include <ATen/ops/_nested_tensor_from_mask_cpu_dispatch.h>
+#include <ATen/ops/_nested_tensor_from_mask_left_aligned_cpu_dispatch.h>
+#include <ATen/ops/_nested_view_from_buffer_cpu_dispatch.h>
+#include <ATen/ops/_padded_dense_to_jagged_forward_cpu_dispatch.h>
+#include <ATen/ops/_pdist_backward_cpu_dispatch.h>
+#include <ATen/ops/_pdist_forward_cpu_dispatch.h>
+#include <ATen/ops/_prelu_kernel_cpu_dispatch.h>
+#include <ATen/ops/_prelu_kernel_backward_cpu_dispatch.h>
+#include <ATen/ops/_reshape_alias_cpu_dispatch.h>
+#include <ATen/ops/_sample_dirichlet_cpu_dispatch.h>
+#include <ATen/ops/_scaled_dot_product_flash_attention_for_cpu_cpu_dispatch.h>
+#include <ATen/ops/_scaled_dot_product_flash_attention_for_cpu_backward_cpu_dispatch.h>
+#include <ATen/ops/_segment_reduce_backward_cpu_dispatch.h>
+#include <ATen/ops/_slow_conv2d_backward_cpu_dispatch.h>
+#include <ATen/ops/_slow_conv2d_forward_cpu_dispatch.h>
+#include <ATen/ops/_softmax_cpu_dispatch.h>
+#include <ATen/ops/_softmax_backward_data_cpu_dispatch.h>
+#include <ATen/ops/_spdiags_cpu_dispatch.h>
+#include <ATen/ops/_stack_cpu_dispatch.h>
+#include <ATen/ops/_standard_gamma_cpu_dispatch.h>
+#include <ATen/ops/_standard_gamma_grad_cpu_dispatch.h>
+#include <ATen/ops/_test_functorch_fallback_cpu_dispatch.h>
+#include <ATen/ops/_test_optional_filled_intlist_cpu_dispatch.h>
+#include <ATen/ops/_test_optional_floatlist_cpu_dispatch.h>
+#include <ATen/ops/_test_optional_intlist_cpu_dispatch.h>
+#include <ATen/ops/_to_sparse_cpu_dispatch.h>
+#include <ATen/ops/_to_sparse_bsc_cpu_dispatch.h>
+#include <ATen/ops/_to_sparse_bsr_cpu_dispatch.h>
+#include <ATen/ops/_to_sparse_csc_cpu_dispatch.h>
+#include <ATen/ops/_to_sparse_csr_cpu_dispatch.h>
+#include <ATen/ops/_transform_bias_rescale_qkv_cpu_dispatch.h>
+#include <ATen/ops/_transformer_encoder_layer_fwd_cpu_dispatch.h>
+#include <ATen/ops/_unique_cpu_dispatch.h>
+#include <ATen/ops/_unique2_cpu_dispatch.h>
+#include <ATen/ops/_upsample_bicubic2d_aa_cpu_dispatch.h>
+#include <ATen/ops/_upsample_bicubic2d_aa_backward_cpu_dispatch.h>
+#include <ATen/ops/_upsample_bilinear2d_aa_cpu_dispatch.h>
+#include <ATen/ops/_upsample_bilinear2d_aa_backward_cpu_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact1d_cpu_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact1d_backward_cpu_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact2d_cpu_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact2d_backward_cpu_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact3d_cpu_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact3d_backward_cpu_dispatch.h>
+#include <ATen/ops/_validate_compressed_sparse_indices_cpu_dispatch.h>
+#include <ATen/ops/_weight_int4pack_mm_for_cpu_cpu_dispatch.h>
+#include <ATen/ops/_weight_int8pack_mm_cpu_dispatch.h>
+#include <ATen/ops/_weight_norm_interface_cpu_dispatch.h>
+#include <ATen/ops/_weight_norm_interface_backward_cpu_dispatch.h>
+#include <ATen/ops/abs_cpu_dispatch.h>
+#include <ATen/ops/acos_cpu_dispatch.h>
+#include <ATen/ops/acosh_cpu_dispatch.h>
+#include <ATen/ops/adaptive_avg_pool2d_cpu_dispatch.h>
+#include <ATen/ops/adaptive_avg_pool3d_cpu_dispatch.h>
+#include <ATen/ops/adaptive_avg_pool3d_backward_cpu_dispatch.h>
+#include <ATen/ops/adaptive_max_pool2d_cpu_dispatch.h>
+#include <ATen/ops/adaptive_max_pool2d_backward_cpu_dispatch.h>
+#include <ATen/ops/adaptive_max_pool3d_cpu_dispatch.h>
+#include <ATen/ops/adaptive_max_pool3d_backward_cpu_dispatch.h>
+#include <ATen/ops/add_cpu_dispatch.h>
+#include <ATen/ops/addbmm_cpu_dispatch.h>
+#include <ATen/ops/addcdiv_cpu_dispatch.h>
+#include <ATen/ops/addcmul_cpu_dispatch.h>
+#include <ATen/ops/addmm_cpu_dispatch.h>
+#include <ATen/ops/addmv_cpu_dispatch.h>
+#include <ATen/ops/addr_cpu_dispatch.h>
+#include <ATen/ops/all_cpu_dispatch.h>
+#include <ATen/ops/amax_cpu_dispatch.h>
+#include <ATen/ops/amin_cpu_dispatch.h>
+#include <ATen/ops/aminmax_cpu_dispatch.h>
+#include <ATen/ops/angle_cpu_dispatch.h>
+#include <ATen/ops/any_cpu_dispatch.h>
+#include <ATen/ops/arange_cpu_dispatch.h>
+#include <ATen/ops/argmax_cpu_dispatch.h>
+#include <ATen/ops/argmin_cpu_dispatch.h>
+#include <ATen/ops/as_strided_cpu_dispatch.h>
+#include <ATen/ops/asin_cpu_dispatch.h>
+#include <ATen/ops/asinh_cpu_dispatch.h>
+#include <ATen/ops/atan_cpu_dispatch.h>
+#include <ATen/ops/atan2_cpu_dispatch.h>
+#include <ATen/ops/atanh_cpu_dispatch.h>
+#include <ATen/ops/avg_pool2d_cpu_dispatch.h>
+#include <ATen/ops/avg_pool2d_backward_cpu_dispatch.h>
+#include <ATen/ops/avg_pool3d_cpu_dispatch.h>
+#include <ATen/ops/avg_pool3d_backward_cpu_dispatch.h>
+#include <ATen/ops/baddbmm_cpu_dispatch.h>
+#include <ATen/ops/batch_norm_backward_cpu_dispatch.h>
+#include <ATen/ops/batch_norm_update_stats_cpu_dispatch.h>
+#include <ATen/ops/bernoulli_cpu_dispatch.h>
+#include <ATen/ops/binary_cross_entropy_cpu_dispatch.h>
+#include <ATen/ops/binary_cross_entropy_backward_cpu_dispatch.h>
+#include <ATen/ops/bincount_cpu_dispatch.h>
+#include <ATen/ops/binomial_cpu_dispatch.h>
+#include <ATen/ops/bitwise_and_cpu_dispatch.h>
+#include <ATen/ops/bitwise_left_shift_cpu_dispatch.h>
+#include <ATen/ops/bitwise_not_cpu_dispatch.h>
+#include <ATen/ops/bitwise_or_cpu_dispatch.h>
+#include <ATen/ops/bitwise_right_shift_cpu_dispatch.h>
+#include <ATen/ops/bitwise_xor_cpu_dispatch.h>
+#include <ATen/ops/bmm_cpu_dispatch.h>
+#include <ATen/ops/bucketize_cpu_dispatch.h>
+#include <ATen/ops/cat_cpu_dispatch.h>
+#include <ATen/ops/cauchy_cpu_dispatch.h>
+#include <ATen/ops/ceil_cpu_dispatch.h>
+#include <ATen/ops/channel_shuffle_cpu_dispatch.h>
+#include <ATen/ops/cholesky_cpu_dispatch.h>
+#include <ATen/ops/cholesky_inverse_cpu_dispatch.h>
+#include <ATen/ops/clamp_cpu_dispatch.h>
+#include <ATen/ops/clamp_max_cpu_dispatch.h>
+#include <ATen/ops/clamp_min_cpu_dispatch.h>
+#include <ATen/ops/col2im_cpu_dispatch.h>
+#include <ATen/ops/complex_cpu_dispatch.h>
+#include <ATen/ops/conj_physical_cpu_dispatch.h>
+#include <ATen/ops/copysign_cpu_dispatch.h>
+#include <ATen/ops/cos_cpu_dispatch.h>
+#include <ATen/ops/cosh_cpu_dispatch.h>
+#include <ATen/ops/count_nonzero_cpu_dispatch.h>
+#include <ATen/ops/cumprod_cpu_dispatch.h>
+#include <ATen/ops/cumsum_cpu_dispatch.h>
+#include <ATen/ops/dequantize_cpu_dispatch.h>
+#include <ATen/ops/digamma_cpu_dispatch.h>
+#include <ATen/ops/div_cpu_dispatch.h>
+#include <ATen/ops/dot_cpu_dispatch.h>
+#include <ATen/ops/elu_cpu_dispatch.h>
+#include <ATen/ops/elu_backward_cpu_dispatch.h>
+#include <ATen/ops/embedding_dense_backward_cpu_dispatch.h>
+#include <ATen/ops/embedding_renorm_cpu_dispatch.h>
+#include <ATen/ops/empty_cpu_dispatch.h>
+#include <ATen/ops/empty_strided_cpu_dispatch.h>
+#include <ATen/ops/eq_cpu_dispatch.h>
+#include <ATen/ops/equal_cpu_dispatch.h>
+#include <ATen/ops/erf_cpu_dispatch.h>
+#include <ATen/ops/erfc_cpu_dispatch.h>
+#include <ATen/ops/erfinv_cpu_dispatch.h>
+#include <ATen/ops/exp_cpu_dispatch.h>
+#include <ATen/ops/exp2_cpu_dispatch.h>
+#include <ATen/ops/expm1_cpu_dispatch.h>
+#include <ATen/ops/exponential_cpu_dispatch.h>
+#include <ATen/ops/eye_cpu_dispatch.h>
+#include <ATen/ops/fake_quantize_per_channel_affine_cachemask_cpu_dispatch.h>
+#include <ATen/ops/fake_quantize_per_tensor_affine_cachemask_cpu_dispatch.h>
+#include <ATen/ops/fill_cpu_dispatch.h>
+#include <ATen/ops/flip_cpu_dispatch.h>
+#include <ATen/ops/floor_cpu_dispatch.h>
+#include <ATen/ops/floor_divide_cpu_dispatch.h>
+#include <ATen/ops/fmax_cpu_dispatch.h>
+#include <ATen/ops/fmin_cpu_dispatch.h>
+#include <ATen/ops/fmod_cpu_dispatch.h>
+#include <ATen/ops/frac_cpu_dispatch.h>
+#include <ATen/ops/fractional_max_pool2d_cpu_dispatch.h>
+#include <ATen/ops/fractional_max_pool2d_backward_cpu_dispatch.h>
+#include <ATen/ops/fractional_max_pool3d_cpu_dispatch.h>
+#include <ATen/ops/fractional_max_pool3d_backward_cpu_dispatch.h>
+#include <ATen/ops/frexp_cpu_dispatch.h>
+#include <ATen/ops/from_file_cpu_dispatch.h>
+#include <ATen/ops/gather_cpu_dispatch.h>
+#include <ATen/ops/gcd_cpu_dispatch.h>
+#include <ATen/ops/ge_cpu_dispatch.h>
+#include <ATen/ops/gelu_cpu_dispatch.h>
+#include <ATen/ops/gelu_backward_cpu_dispatch.h>
+#include <ATen/ops/geometric_cpu_dispatch.h>
+#include <ATen/ops/geqrf_cpu_dispatch.h>
+#include <ATen/ops/glu_cpu_dispatch.h>
+#include <ATen/ops/glu_backward_cpu_dispatch.h>
+#include <ATen/ops/glu_backward_jvp_cpu_dispatch.h>
+#include <ATen/ops/glu_jvp_cpu_dispatch.h>
+#include <ATen/ops/grid_sampler_2d_cpu_dispatch.h>
+#include <ATen/ops/grid_sampler_2d_backward_cpu_dispatch.h>
+#include <ATen/ops/grid_sampler_3d_cpu_dispatch.h>
+#include <ATen/ops/grid_sampler_3d_backward_cpu_dispatch.h>
+#include <ATen/ops/gt_cpu_dispatch.h>
+#include <ATen/ops/hardshrink_cpu_dispatch.h>
+#include <ATen/ops/hardshrink_backward_cpu_dispatch.h>
+#include <ATen/ops/hardsigmoid_cpu_dispatch.h>
+#include <ATen/ops/hardsigmoid_backward_cpu_dispatch.h>
+#include <ATen/ops/hardswish_cpu_dispatch.h>
+#include <ATen/ops/hardswish_backward_cpu_dispatch.h>
+#include <ATen/ops/hardtanh_cpu_dispatch.h>
+#include <ATen/ops/hardtanh_backward_cpu_dispatch.h>
+#include <ATen/ops/heaviside_cpu_dispatch.h>
+#include <ATen/ops/histc_cpu_dispatch.h>
+#include <ATen/ops/histogram_cpu_dispatch.h>
+#include <ATen/ops/huber_loss_cpu_dispatch.h>
+#include <ATen/ops/huber_loss_backward_cpu_dispatch.h>
+#include <ATen/ops/hypot_cpu_dispatch.h>
+#include <ATen/ops/i0_cpu_dispatch.h>
+#include <ATen/ops/igamma_cpu_dispatch.h>
+#include <ATen/ops/igammac_cpu_dispatch.h>
+#include <ATen/ops/im2col_cpu_dispatch.h>
+#include <ATen/ops/index_cpu_dispatch.h>
+#include <ATen/ops/index_add_cpu_dispatch.h>
+#include <ATen/ops/index_copy_cpu_dispatch.h>
+#include <ATen/ops/index_fill_cpu_dispatch.h>
+#include <ATen/ops/index_reduce_cpu_dispatch.h>
+#include <ATen/ops/index_select_cpu_dispatch.h>
+#include <ATen/ops/is_set_to_cpu_dispatch.h>
+#include <ATen/ops/isin_cpu_dispatch.h>
+#include <ATen/ops/isnan_cpu_dispatch.h>
+#include <ATen/ops/isneginf_cpu_dispatch.h>
+#include <ATen/ops/isposinf_cpu_dispatch.h>
+#include <ATen/ops/kthvalue_cpu_dispatch.h>
+#include <ATen/ops/lcm_cpu_dispatch.h>
+#include <ATen/ops/le_cpu_dispatch.h>
+#include <ATen/ops/leaky_relu_cpu_dispatch.h>
+#include <ATen/ops/leaky_relu_backward_cpu_dispatch.h>
+#include <ATen/ops/lerp_cpu_dispatch.h>
+#include <ATen/ops/lgamma_cpu_dispatch.h>
+#include <ATen/ops/linalg_cholesky_ex_cpu_dispatch.h>
+#include <ATen/ops/linalg_cross_cpu_dispatch.h>
+#include <ATen/ops/linalg_eig_cpu_dispatch.h>
+#include <ATen/ops/linalg_eigvals_cpu_dispatch.h>
+#include <ATen/ops/linalg_householder_product_cpu_dispatch.h>
+#include <ATen/ops/linalg_inv_ex_cpu_dispatch.h>
+#include <ATen/ops/linalg_ldl_factor_ex_cpu_dispatch.h>
+#include <ATen/ops/linalg_ldl_solve_cpu_dispatch.h>
+#include <ATen/ops/linalg_lstsq_cpu_dispatch.h>
+#include <ATen/ops/linalg_lu_cpu_dispatch.h>
+#include <ATen/ops/linalg_lu_factor_ex_cpu_dispatch.h>
+#include <ATen/ops/linalg_lu_solve_cpu_dispatch.h>
+#include <ATen/ops/linalg_matrix_exp_cpu_dispatch.h>
+#include <ATen/ops/linalg_qr_cpu_dispatch.h>
+#include <ATen/ops/linalg_solve_triangular_cpu_dispatch.h>
+#include <ATen/ops/linalg_vector_norm_cpu_dispatch.h>
+#include <ATen/ops/linspace_cpu_dispatch.h>
+#include <ATen/ops/log_cpu_dispatch.h>
+#include <ATen/ops/log10_cpu_dispatch.h>
+#include <ATen/ops/log1p_cpu_dispatch.h>
+#include <ATen/ops/log2_cpu_dispatch.h>
+#include <ATen/ops/log_normal_cpu_dispatch.h>
+#include <ATen/ops/log_sigmoid_backward_cpu_dispatch.h>
+#include <ATen/ops/log_sigmoid_forward_cpu_dispatch.h>
+#include <ATen/ops/logaddexp_cpu_dispatch.h>
+#include <ATen/ops/logaddexp2_cpu_dispatch.h>
+#include <ATen/ops/logical_and_cpu_dispatch.h>
+#include <ATen/ops/logical_not_cpu_dispatch.h>
+#include <ATen/ops/logical_or_cpu_dispatch.h>
+#include <ATen/ops/logical_xor_cpu_dispatch.h>
+#include <ATen/ops/logit_cpu_dispatch.h>
+#include <ATen/ops/logit_backward_cpu_dispatch.h>
+#include <ATen/ops/logspace_cpu_dispatch.h>
+#include <ATen/ops/lshift_cpu_dispatch.h>
+#include <ATen/ops/lt_cpu_dispatch.h>
+#include <ATen/ops/lu_unpack_cpu_dispatch.h>
+#include <ATen/ops/masked_fill_cpu_dispatch.h>
+#include <ATen/ops/masked_scatter_cpu_dispatch.h>
+#include <ATen/ops/masked_select_cpu_dispatch.h>
+#include <ATen/ops/max_cpu_dispatch.h>
+#include <ATen/ops/max_pool2d_with_indices_cpu_dispatch.h>
+#include <ATen/ops/max_pool2d_with_indices_backward_cpu_dispatch.h>
+#include <ATen/ops/max_pool3d_with_indices_cpu_dispatch.h>
+#include <ATen/ops/max_pool3d_with_indices_backward_cpu_dispatch.h>
+#include <ATen/ops/max_unpool2d_cpu_dispatch.h>
+#include <ATen/ops/max_unpool3d_cpu_dispatch.h>
+#include <ATen/ops/maximum_cpu_dispatch.h>
+#include <ATen/ops/mean_cpu_dispatch.h>
+#include <ATen/ops/median_cpu_dispatch.h>
+#include <ATen/ops/min_cpu_dispatch.h>
+#include <ATen/ops/minimum_cpu_dispatch.h>
+#include <ATen/ops/mish_cpu_dispatch.h>
+#include <ATen/ops/mish_backward_cpu_dispatch.h>
+#include <ATen/ops/mkldnn_rnn_layer_cpu_dispatch.h>
+#include <ATen/ops/mkldnn_rnn_layer_backward_cpu_dispatch.h>
+#include <ATen/ops/mm_cpu_dispatch.h>
+#include <ATen/ops/mode_cpu_dispatch.h>
+#include <ATen/ops/mse_loss_cpu_dispatch.h>
+#include <ATen/ops/mse_loss_backward_cpu_dispatch.h>
+#include <ATen/ops/mul_cpu_dispatch.h>
+#include <ATen/ops/multi_margin_loss_cpu_dispatch.h>
+#include <ATen/ops/multi_margin_loss_backward_cpu_dispatch.h>
+#include <ATen/ops/multilabel_margin_loss_backward_cpu_dispatch.h>
+#include <ATen/ops/multilabel_margin_loss_forward_cpu_dispatch.h>
+#include <ATen/ops/multinomial_cpu_dispatch.h>
+#include <ATen/ops/mvlgamma_cpu_dispatch.h>
+#include <ATen/ops/nan_to_num_cpu_dispatch.h>
+#include <ATen/ops/nanmedian_cpu_dispatch.h>
+#include <ATen/ops/nansum_cpu_dispatch.h>
+#include <ATen/ops/narrow_copy_cpu_dispatch.h>
+#include <ATen/ops/native_batch_norm_cpu_dispatch.h>
+#include <ATen/ops/native_batch_norm_backward_cpu_dispatch.h>
+#include <ATen/ops/native_channel_shuffle_cpu_dispatch.h>
+#include <ATen/ops/native_dropout_cpu_dispatch.h>
+#include <ATen/ops/native_dropout_backward_cpu_dispatch.h>
+#include <ATen/ops/native_group_norm_cpu_dispatch.h>
+#include <ATen/ops/native_group_norm_backward_cpu_dispatch.h>
+#include <ATen/ops/native_layer_norm_cpu_dispatch.h>
+#include <ATen/ops/native_layer_norm_backward_cpu_dispatch.h>
+#include <ATen/ops/ne_cpu_dispatch.h>
+#include <ATen/ops/neg_cpu_dispatch.h>
+#include <ATen/ops/nextafter_cpu_dispatch.h>
+#include <ATen/ops/nll_loss2d_backward_cpu_dispatch.h>
+#include <ATen/ops/nll_loss2d_forward_cpu_dispatch.h>
+#include <ATen/ops/nll_loss_backward_cpu_dispatch.h>
+#include <ATen/ops/nll_loss_forward_cpu_dispatch.h>
+#include <ATen/ops/nonzero_cpu_dispatch.h>
+#include <ATen/ops/nonzero_static_cpu_dispatch.h>
+#include <ATen/ops/norm_cpu_dispatch.h>
+#include <ATen/ops/normal_cpu_dispatch.h>
+#include <ATen/ops/ormqr_cpu_dispatch.h>
+#include <ATen/ops/pixel_shuffle_cpu_dispatch.h>
+#include <ATen/ops/pixel_unshuffle_cpu_dispatch.h>
+#include <ATen/ops/poisson_cpu_dispatch.h>
+#include <ATen/ops/polar_cpu_dispatch.h>
+#include <ATen/ops/polygamma_cpu_dispatch.h>
+#include <ATen/ops/pow_cpu_dispatch.h>
+#include <ATen/ops/prod_cpu_dispatch.h>
+#include <ATen/ops/put_cpu_dispatch.h>
+#include <ATen/ops/quantize_per_channel_cpu_dispatch.h>
+#include <ATen/ops/quantize_per_tensor_cpu_dispatch.h>
+#include <ATen/ops/quantize_per_tensor_dynamic_cpu_dispatch.h>
+#include <ATen/ops/random_cpu_dispatch.h>
+#include <ATen/ops/randperm_cpu_dispatch.h>
+#include <ATen/ops/range_cpu_dispatch.h>
+#include <ATen/ops/reciprocal_cpu_dispatch.h>
+#include <ATen/ops/reflection_pad1d_cpu_dispatch.h>
+#include <ATen/ops/reflection_pad1d_backward_cpu_dispatch.h>
+#include <ATen/ops/reflection_pad2d_cpu_dispatch.h>
+#include <ATen/ops/reflection_pad2d_backward_cpu_dispatch.h>
+#include <ATen/ops/reflection_pad3d_cpu_dispatch.h>
+#include <ATen/ops/reflection_pad3d_backward_cpu_dispatch.h>
+#include <ATen/ops/relu_cpu_dispatch.h>
+#include <ATen/ops/remainder_cpu_dispatch.h>
+#include <ATen/ops/renorm_cpu_dispatch.h>
+#include <ATen/ops/repeat_interleave_cpu_dispatch.h>
+#include <ATen/ops/replication_pad1d_cpu_dispatch.h>
+#include <ATen/ops/replication_pad1d_backward_cpu_dispatch.h>
+#include <ATen/ops/replication_pad2d_cpu_dispatch.h>
+#include <ATen/ops/replication_pad2d_backward_cpu_dispatch.h>
+#include <ATen/ops/replication_pad3d_cpu_dispatch.h>
+#include <ATen/ops/replication_pad3d_backward_cpu_dispatch.h>
+#include <ATen/ops/resize_cpu_dispatch.h>
+#include <ATen/ops/roll_cpu_dispatch.h>
+#include <ATen/ops/round_cpu_dispatch.h>
+#include <ATen/ops/rrelu_with_noise_cpu_dispatch.h>
+#include <ATen/ops/rshift_cpu_dispatch.h>
+#include <ATen/ops/rsqrt_cpu_dispatch.h>
+#include <ATen/ops/rsub_cpu_dispatch.h>
+#include <ATen/ops/scatter_cpu_dispatch.h>
+#include <ATen/ops/scatter_add_cpu_dispatch.h>
+#include <ATen/ops/scatter_reduce_cpu_dispatch.h>
+#include <ATen/ops/searchsorted_cpu_dispatch.h>
+#include <ATen/ops/segment_reduce_cpu_dispatch.h>
+#include <ATen/ops/set_cpu_dispatch.h>
+#include <ATen/ops/sgn_cpu_dispatch.h>
+#include <ATen/ops/sigmoid_cpu_dispatch.h>
+#include <ATen/ops/sigmoid_backward_cpu_dispatch.h>
+#include <ATen/ops/sign_cpu_dispatch.h>
+#include <ATen/ops/signbit_cpu_dispatch.h>
+#include <ATen/ops/silu_cpu_dispatch.h>
+#include <ATen/ops/silu_backward_cpu_dispatch.h>
+#include <ATen/ops/sin_cpu_dispatch.h>
+#include <ATen/ops/sinc_cpu_dispatch.h>
+#include <ATen/ops/sinh_cpu_dispatch.h>
+#include <ATen/ops/slow_conv3d_forward_cpu_dispatch.h>
+#include <ATen/ops/slow_conv_dilated2d_cpu_dispatch.h>
+#include <ATen/ops/slow_conv_dilated3d_cpu_dispatch.h>
+#include <ATen/ops/slow_conv_transpose2d_cpu_dispatch.h>
+#include <ATen/ops/slow_conv_transpose3d_cpu_dispatch.h>
+#include <ATen/ops/smooth_l1_loss_cpu_dispatch.h>
+#include <ATen/ops/smooth_l1_loss_backward_cpu_dispatch.h>
+#include <ATen/ops/softplus_cpu_dispatch.h>
+#include <ATen/ops/softplus_backward_cpu_dispatch.h>
+#include <ATen/ops/softshrink_cpu_dispatch.h>
+#include <ATen/ops/softshrink_backward_cpu_dispatch.h>
+#include <ATen/ops/sort_cpu_dispatch.h>
+#include <ATen/ops/special_airy_ai_cpu_dispatch.h>
+#include <ATen/ops/special_bessel_j0_cpu_dispatch.h>
+#include <ATen/ops/special_bessel_j1_cpu_dispatch.h>
+#include <ATen/ops/special_bessel_y0_cpu_dispatch.h>
+#include <ATen/ops/special_bessel_y1_cpu_dispatch.h>
+#include <ATen/ops/special_chebyshev_polynomial_t_cpu_dispatch.h>
+#include <ATen/ops/special_chebyshev_polynomial_u_cpu_dispatch.h>
+#include <ATen/ops/special_chebyshev_polynomial_v_cpu_dispatch.h>
+#include <ATen/ops/special_chebyshev_polynomial_w_cpu_dispatch.h>
+#include <ATen/ops/special_entr_cpu_dispatch.h>
+#include <ATen/ops/special_erfcx_cpu_dispatch.h>
+#include <ATen/ops/special_hermite_polynomial_h_cpu_dispatch.h>
+#include <ATen/ops/special_hermite_polynomial_he_cpu_dispatch.h>
+#include <ATen/ops/special_i0e_cpu_dispatch.h>
+#include <ATen/ops/special_i1_cpu_dispatch.h>
+#include <ATen/ops/special_i1e_cpu_dispatch.h>
+#include <ATen/ops/special_laguerre_polynomial_l_cpu_dispatch.h>
+#include <ATen/ops/special_legendre_polynomial_p_cpu_dispatch.h>
+#include <ATen/ops/special_log_ndtr_cpu_dispatch.h>
+#include <ATen/ops/special_modified_bessel_i0_cpu_dispatch.h>
+#include <ATen/ops/special_modified_bessel_i1_cpu_dispatch.h>
+#include <ATen/ops/special_modified_bessel_k0_cpu_dispatch.h>
+#include <ATen/ops/special_modified_bessel_k1_cpu_dispatch.h>
+#include <ATen/ops/special_ndtri_cpu_dispatch.h>
+#include <ATen/ops/special_scaled_modified_bessel_k0_cpu_dispatch.h>
+#include <ATen/ops/special_scaled_modified_bessel_k1_cpu_dispatch.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_t_cpu_dispatch.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_u_cpu_dispatch.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_v_cpu_dispatch.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_w_cpu_dispatch.h>
+#include <ATen/ops/special_spherical_bessel_j0_cpu_dispatch.h>
+#include <ATen/ops/special_xlog1py_cpu_dispatch.h>
+#include <ATen/ops/special_zeta_cpu_dispatch.h>
+#include <ATen/ops/sqrt_cpu_dispatch.h>
+#include <ATen/ops/sspaddmm_cpu_dispatch.h>
+#include <ATen/ops/std_cpu_dispatch.h>
+#include <ATen/ops/std_mean_cpu_dispatch.h>
+#include <ATen/ops/sub_cpu_dispatch.h>
+#include <ATen/ops/sum_cpu_dispatch.h>
+#include <ATen/ops/take_cpu_dispatch.h>
+#include <ATen/ops/tan_cpu_dispatch.h>
+#include <ATen/ops/tanh_cpu_dispatch.h>
+#include <ATen/ops/tanh_backward_cpu_dispatch.h>
+#include <ATen/ops/threshold_cpu_dispatch.h>
+#include <ATen/ops/threshold_backward_cpu_dispatch.h>
+#include <ATen/ops/to_mkldnn_cpu_dispatch.h>
+#include <ATen/ops/topk_cpu_dispatch.h>
+#include <ATen/ops/trace_cpu_dispatch.h>
+#include <ATen/ops/triangular_solve_cpu_dispatch.h>
+#include <ATen/ops/tril_cpu_dispatch.h>
+#include <ATen/ops/tril_indices_cpu_dispatch.h>
+#include <ATen/ops/triu_cpu_dispatch.h>
+#include <ATen/ops/triu_indices_cpu_dispatch.h>
+#include <ATen/ops/trunc_cpu_dispatch.h>
+#include <ATen/ops/unfold_cpu_dispatch.h>
+#include <ATen/ops/unfold_backward_cpu_dispatch.h>
+#include <ATen/ops/uniform_cpu_dispatch.h>
+#include <ATen/ops/unique_consecutive_cpu_dispatch.h>
+#include <ATen/ops/unique_dim_cpu_dispatch.h>
+#include <ATen/ops/unique_dim_consecutive_cpu_dispatch.h>
+#include <ATen/ops/upsample_bicubic2d_cpu_dispatch.h>
+#include <ATen/ops/upsample_bicubic2d_backward_cpu_dispatch.h>
+#include <ATen/ops/upsample_bilinear2d_cpu_dispatch.h>
+#include <ATen/ops/upsample_bilinear2d_backward_cpu_dispatch.h>
+#include <ATen/ops/upsample_linear1d_cpu_dispatch.h>
+#include <ATen/ops/upsample_linear1d_backward_cpu_dispatch.h>
+#include <ATen/ops/upsample_nearest1d_cpu_dispatch.h>
+#include <ATen/ops/upsample_nearest1d_backward_cpu_dispatch.h>
+#include <ATen/ops/upsample_nearest2d_cpu_dispatch.h>
+#include <ATen/ops/upsample_nearest2d_backward_cpu_dispatch.h>
+#include <ATen/ops/upsample_nearest3d_cpu_dispatch.h>
+#include <ATen/ops/upsample_nearest3d_backward_cpu_dispatch.h>
+#include <ATen/ops/upsample_trilinear3d_cpu_dispatch.h>
+#include <ATen/ops/upsample_trilinear3d_backward_cpu_dispatch.h>
+#include <ATen/ops/var_cpu_dispatch.h>
+#include <ATen/ops/var_mean_cpu_dispatch.h>
+#include <ATen/ops/vdot_cpu_dispatch.h>
+#include <ATen/ops/view_cpu_dispatch.h>
+#include <ATen/ops/view_as_complex_cpu_dispatch.h>
+#include <ATen/ops/view_as_real_cpu_dispatch.h>
+#include <ATen/ops/where_cpu_dispatch.h>
+#include <ATen/ops/xlogy_cpu_dispatch.h>
+#include <ATen/ops/zero_cpu_dispatch.h>
+
+
+

lib/python3.12/site-packages/torch/include/ATen/CPUGeneratorImpl.h

@@ -0,0 +1,49 @@
+#pragma once
+
+#include <ATen/core/Generator.h>
+#include <ATen/core/MT19937RNGEngine.h>
+#include <c10/core/GeneratorImpl.h>
+#include <optional>
+
+namespace at {
+
+struct TORCH_API CPUGeneratorImpl : public c10::GeneratorImpl {
+  // Constructors
+  CPUGeneratorImpl(uint64_t seed_in = default_rng_seed_val);
+  ~CPUGeneratorImpl() override = default;
+
+  // CPUGeneratorImpl methods
+  std::shared_ptr<CPUGeneratorImpl> clone() const;
+  void set_current_seed(uint64_t seed) override;
+  void set_offset(uint64_t offset) override;
+  uint64_t get_offset() const override;
+  uint64_t current_seed() const override;
+  uint64_t seed() override;
+  void set_state(const c10::TensorImpl& new_state) override;
+  c10::intrusive_ptr<c10::TensorImpl> get_state() const override;
+  static c10::DeviceType device_type();
+  uint32_t random();
+  uint64_t random64();
+  std::optional<float> next_float_normal_sample();
+  std::optional<double> next_double_normal_sample();
+  void set_next_float_normal_sample(std::optional<float> randn);
+  void set_next_double_normal_sample(std::optional<double> randn);
+  at::mt19937 engine();
+  void set_engine(at::mt19937 engine);
+
+ private:
+  CPUGeneratorImpl* clone_impl() const override;
+  at::mt19937 engine_;
+  std::optional<float> next_float_normal_sample_;
+  std::optional<double> next_double_normal_sample_;
+};
+
+namespace detail {
+
+TORCH_API const Generator& getDefaultCPUGenerator();
+TORCH_API Generator
+createCPUGenerator(uint64_t seed_val = default_rng_seed_val);
+
+} // namespace detail
+
+} // namespace at

lib/python3.12/site-packages/torch/include/ATen/CUDAFunctions.h

@@ -0,0 +1,29 @@
+#include <ATen/core/TensorBody.h>
+
+// TODO Undo all logic introduced for Note [Avoiding Include Cycles In Static Dispatch]
+// Code introduced to avoid cyclic dependency in static dispatch is no longer
+// needed as static dispatch logic is moved from TensorBody.h, which caused cycles in the first place,
+// to Operators.cpp for supporting multiple backends with multiple kernels.
+//
+// Note [Avoiding Include Cycles In Static Dispatch]
+// In order to avoid #include cycles in the static dispatch build, we've carefully split out
+// the static function definition files into {DispatchKey}Functions.h and {DispatchKey}Functions_inl.h.
+//
+// Without this split, the include cycle looks like TensorBody.h -> CPUFunctions.h -> TensorBody.h.
+// - TensorBody.h #includes CPUFunctions.h in the static dispatch build, because the tensor methods
+//   all need to call into the fastpath C++ API defined in CPUFunctions.h. The methods are also all
+//   directly inlined into TensorBody.h.
+// - CPUFunctions.h #includes TensorBody.h because it contains function declarations for the entire C++ API,
+//   which include functions that have defaultable std::optional<Tensor> arguments.
+//   That requires knowing the full Tensor class definition.
+//
+// We break the cycle by doing the following:
+// - Split out CPUFunction.h into two files: CPUFunctions.h and CPUFunctions_inl.h
+// - CPUFunction.h is a dummy file that just includes the Tensor class and includes CPUFunctions_inl.,
+// - CPUFunctions_inl.h includes everything else
+// - (only in the static dispatch build) TensorBody.h makes sure to finish defining the Tensor class,
+//   and then it includes CPUFunctions_inl.h.
+// - All other files that want the cpu fastpath functions can include CPUFunctions.h directly.
+// - This also means that static dispatch build, CPUFunctions.h only needs to
+//   #include TensorBody.h, and it will automatically bring in CPUFunctions_inl.h.
+#include <ATen/CUDAFunctions_inl.h>

lib/python3.12/site-packages/torch/include/ATen/CollapseDims.h

@@ -0,0 +1,94 @@
+#include <c10/util/Exception.h>
+#include <utility>
+
+namespace at {
+
+/*
+[collapse dims] Updates sizes, and strides to reflect a "collapse" of
+the info, possibly excluding the optional excludeDim. A "collapsed" version
+of the info is the fewest dims that order the tensor's elements in the same
+way as the original info. If excludeDim is specified, the collapse is the
+fewest dims that order the tensor's elements as the original and preserve the
+excluded dimension, unless the tensor collapses to a point.
+
+This function returns a pair of values.
+
+1) The (new) index of the preserved dimension if excludeDim is
+specified. 0 if the tensor is collapsed to a point. -1
+otherwise.
+
+2) The new number of dimensions.
+*/
+template <typename T>
+inline std::pair<int64_t, int64_t> collapse_dims(
+    T* sizes,
+    T* strides,
+    int64_t dims,
+    const int excludeDim = -1) {
+  TORCH_CHECK(
+      excludeDim >= -1 && excludeDim < dims,
+      "expected excluded dim between -1 and dims - 1");
+
+  int64_t stopDim = (excludeDim == -1) ? dims : excludeDim;
+  int64_t newIndex = -1;
+  int64_t oldIndex = 0;
+  int64_t remappedExcludedDim = -1;
+
+  while (oldIndex < dims) {
+    // Finds a dimension to collapse into
+    for (; oldIndex < stopDim; ++oldIndex) {
+      if (sizes[oldIndex] == 1) {
+        continue;
+      }
+
+      ++newIndex;
+      sizes[newIndex] = sizes[oldIndex];
+      strides[newIndex] = strides[oldIndex];
+      ++oldIndex;
+      break;
+    }
+
+    // Collapses dims
+    for (; oldIndex < stopDim; ++oldIndex) {
+      if (sizes[oldIndex] == 1) {
+        continue;
+      }
+
+      if (strides[newIndex] == sizes[oldIndex] * strides[oldIndex]) {
+        sizes[newIndex] *= sizes[oldIndex];
+        strides[newIndex] = strides[oldIndex];
+      } else {
+        ++newIndex;
+        sizes[newIndex] = sizes[oldIndex];
+        strides[newIndex] = strides[oldIndex];
+      }
+    }
+
+    // Handles excludeDim being set (oldIndex == excludeDim)
+    if (oldIndex != dims) {
+      // Preserves excluded dimension
+      ++newIndex;
+      sizes[newIndex] = sizes[oldIndex];
+      strides[newIndex] = strides[oldIndex];
+      remappedExcludedDim = newIndex;
+
+      // Restarts iteration after excludeDim
+      ++oldIndex;
+      stopDim = dims;
+    }
+  }
+
+  // Handles special case of all dims size 1
+  if (newIndex == -1 || (newIndex == 0 && sizes[0] == 1)) {
+    dims = 1;
+    sizes[0] = 1;
+    strides[0] = 1;
+
+    return std::pair<int64_t, int64_t>(0, 1);
+  }
+
+  dims = newIndex + 1;
+  return std::pair<int64_t, int64_t>(remappedExcludedDim, dims);
+}
+
+} // namespace at

lib/python3.12/site-packages/torch/include/ATen/CompositeExplicitAutogradFunctions.h

@@ -0,0 +1,29 @@
+#include <ATen/core/TensorBody.h>
+
+// TODO Undo all logic introduced for Note [Avoiding Include Cycles In Static Dispatch]
+// Code introduced to avoid cyclic dependency in static dispatch is no longer
+// needed as static dispatch logic is moved from TensorBody.h, which caused cycles in the first place,
+// to Operators.cpp for supporting multiple backends with multiple kernels.
+//
+// Note [Avoiding Include Cycles In Static Dispatch]
+// In order to avoid #include cycles in the static dispatch build, we've carefully split out
+// the static function definition files into {DispatchKey}Functions.h and {DispatchKey}Functions_inl.h.
+//
+// Without this split, the include cycle looks like TensorBody.h -> CPUFunctions.h -> TensorBody.h.
+// - TensorBody.h #includes CPUFunctions.h in the static dispatch build, because the tensor methods
+//   all need to call into the fastpath C++ API defined in CPUFunctions.h. The methods are also all
+//   directly inlined into TensorBody.h.
+// - CPUFunctions.h #includes TensorBody.h because it contains function declarations for the entire C++ API,
+//   which include functions that have defaultable std::optional<Tensor> arguments.
+//   That requires knowing the full Tensor class definition.
+//
+// We break the cycle by doing the following:
+// - Split out CPUFunction.h into two files: CPUFunctions.h and CPUFunctions_inl.h
+// - CPUFunction.h is a dummy file that just includes the Tensor class and includes CPUFunctions_inl.,
+// - CPUFunctions_inl.h includes everything else
+// - (only in the static dispatch build) TensorBody.h makes sure to finish defining the Tensor class,
+//   and then it includes CPUFunctions_inl.h.
+// - All other files that want the cpu fastpath functions can include CPUFunctions.h directly.
+// - This also means that static dispatch build, CPUFunctions.h only needs to
+//   #include TensorBody.h, and it will automatically bring in CPUFunctions_inl.h.
+#include <ATen/CompositeExplicitAutogradFunctions_inl.h>

lib/python3.12/site-packages/torch/include/ATen/CompositeExplicitAutogradNonFunctionalFunctions.h

@@ -0,0 +1,29 @@
+#include <ATen/core/TensorBody.h>
+
+// TODO Undo all logic introduced for Note [Avoiding Include Cycles In Static Dispatch]
+// Code introduced to avoid cyclic dependency in static dispatch is no longer
+// needed as static dispatch logic is moved from TensorBody.h, which caused cycles in the first place,
+// to Operators.cpp for supporting multiple backends with multiple kernels.
+//
+// Note [Avoiding Include Cycles In Static Dispatch]
+// In order to avoid #include cycles in the static dispatch build, we've carefully split out
+// the static function definition files into {DispatchKey}Functions.h and {DispatchKey}Functions_inl.h.
+//
+// Without this split, the include cycle looks like TensorBody.h -> CPUFunctions.h -> TensorBody.h.
+// - TensorBody.h #includes CPUFunctions.h in the static dispatch build, because the tensor methods
+//   all need to call into the fastpath C++ API defined in CPUFunctions.h. The methods are also all
+//   directly inlined into TensorBody.h.
+// - CPUFunctions.h #includes TensorBody.h because it contains function declarations for the entire C++ API,
+//   which include functions that have defaultable std::optional<Tensor> arguments.
+//   That requires knowing the full Tensor class definition.
+//
+// We break the cycle by doing the following:
+// - Split out CPUFunction.h into two files: CPUFunctions.h and CPUFunctions_inl.h
+// - CPUFunction.h is a dummy file that just includes the Tensor class and includes CPUFunctions_inl.,
+// - CPUFunctions_inl.h includes everything else
+// - (only in the static dispatch build) TensorBody.h makes sure to finish defining the Tensor class,
+//   and then it includes CPUFunctions_inl.h.
+// - All other files that want the cpu fastpath functions can include CPUFunctions.h directly.
+// - This also means that static dispatch build, CPUFunctions.h only needs to
+//   #include TensorBody.h, and it will automatically bring in CPUFunctions_inl.h.
+#include <ATen/CompositeExplicitAutogradNonFunctionalFunctions_inl.h>

lib/python3.12/site-packages/torch/include/ATen/CompositeExplicitAutogradNonFunctionalFunctions_inl.h

@@ -0,0 +1,323 @@
+#pragma once
+// @generated by torchgen/gen.py from DispatchKeyFunctions_inl.h
+
+// NB: The implementing C++ file is RegisterDispatchKey.cpp
+
+// The only #includes we need are for custom classes that have defaults in the C++ API
+#include <c10/core/MemoryFormat.h>
+#include <c10/core/Scalar.h>
+#include <ATen/core/Reduction.h>
+
+#if defined(AT_PER_OPERATOR_HEADERS) && defined(TORCH_ASSERT_ONLY_METHOD_OPERATORS)
+#error This change adds a dependency on all pytorch operators, meaning the     \
+  file will need to be re-compiled every time an operator is changed or added. \
+  Consider including a specific operator from                                  \
+  <ATen/ops/{my_operator}_compositeexplicitautogradnonfunctional_dispatch.h>.                   \
+  See NOTE [TORCH_ASSERT_ONLY_METHOD_OPERATORS].
+#endif
+
+#include <ATen/ops/_addmm_activation_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_conj_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_convert_indices_from_coo_to_csr_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_convert_indices_from_csr_to_coo_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_fw_primal_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_indices_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_linalg_det_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_linalg_eigh_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_linalg_slogdet_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_linalg_solve_ex_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_linalg_svd_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_log_softmax_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_log_softmax_backward_data_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_make_dual_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_neg_view_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_nested_get_values_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_nested_view_from_buffer_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_nested_view_from_jagged_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_reshape_alias_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_softmax_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_softmax_backward_data_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_sparse_broadcast_to_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_test_autograd_multiple_dispatch_view_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_trilinear_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_upsample_bicubic2d_aa_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_upsample_bicubic2d_aa_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_upsample_bilinear2d_aa_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_upsample_bilinear2d_aa_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact1d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact1d_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact2d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact2d_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact3d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact3d_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/_values_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/acos_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/acosh_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/adaptive_max_pool2d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/adaptive_max_pool2d_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/adaptive_max_pool3d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/adaptive_max_pool3d_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/add_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/addcdiv_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/addcmul_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/addmm_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/addmv_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/alias_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/all_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/amax_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/amin_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/aminmax_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/any_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/argmax_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/argmin_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/as_strided_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/as_strided_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/as_strided_scatter_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/asin_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/asinh_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/atan_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/atan2_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/atanh_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/avg_pool2d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/avg_pool2d_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/avg_pool3d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/avg_pool3d_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/baddbmm_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/bernoulli_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/bitwise_and_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/bitwise_left_shift_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/bitwise_not_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/bitwise_or_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/bitwise_right_shift_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/bitwise_xor_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/bmm_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/cat_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/ccol_indices_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/ceil_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/clamp_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/clamp_max_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/clamp_min_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/col_indices_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/copysign_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/cos_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/cosh_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/crow_indices_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/cumprod_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/cumsum_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/detach_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/diag_embed_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/diagonal_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/diagonal_scatter_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/digamma_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/div_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/elu_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/elu_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/eq_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/erf_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/erfc_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/erfinv_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/exp_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/exp2_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/expand_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/expm1_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/floor_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/fmax_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/fmin_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/fmod_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/frac_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/fractional_max_pool2d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/fractional_max_pool2d_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/fractional_max_pool3d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/gather_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/gcd_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/ge_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/gelu_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/gelu_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/glu_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/gt_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/hardshrink_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/hardshrink_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/hardsigmoid_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/hardsigmoid_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/heaviside_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/hypot_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/i0_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/igamma_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/igammac_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/index_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/index_add_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/index_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/index_reduce_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/indices_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/isin_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/isneginf_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/isposinf_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/lcm_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/le_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/leaky_relu_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/leaky_relu_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/lerp_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/lgamma_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/lift_fresh_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/linalg_cholesky_ex_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/linalg_cross_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/linalg_inv_ex_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/linalg_ldl_factor_ex_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/linalg_ldl_solve_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/linalg_lu_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/linalg_lu_factor_ex_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/linalg_lu_solve_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/linalg_pinv_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/linalg_qr_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/linalg_vector_norm_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/log_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/log10_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/log1p_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/log2_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/logaddexp_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/logaddexp2_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/logit_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/logsumexp_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/lt_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/lu_unpack_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/max_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/max_pool2d_with_indices_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/max_pool2d_with_indices_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/maximum_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/mean_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/min_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/minimum_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/mish_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/mm_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/mse_loss_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/mul_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/narrow_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/ne_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/neg_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/new_empty_strided_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/nextafter_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/nll_loss_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/nll_loss_forward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/norm_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/permute_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/pixel_shuffle_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/pixel_unshuffle_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/polygamma_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/pow_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/prod_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/reciprocal_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/reflection_pad1d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/reflection_pad1d_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/reflection_pad3d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/reflection_pad3d_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/remainder_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/renorm_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/replication_pad1d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/replication_pad1d_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/replication_pad2d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/replication_pad3d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/round_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/row_indices_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/rsqrt_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/scatter_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/scatter_add_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/scatter_reduce_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/select_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/select_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/select_scatter_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/sgn_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/sigmoid_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/sigmoid_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/sign_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/signbit_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/silu_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/silu_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/sin_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/sinc_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/sinh_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/slice_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/slice_scatter_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/slow_conv_transpose2d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/smooth_l1_loss_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/softplus_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/softplus_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/softshrink_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/softshrink_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/sort_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_airy_ai_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_bessel_j0_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_bessel_j1_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_bessel_y0_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_bessel_y1_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_chebyshev_polynomial_t_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_chebyshev_polynomial_u_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_chebyshev_polynomial_v_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_chebyshev_polynomial_w_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_entr_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_erfcx_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_hermite_polynomial_h_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_hermite_polynomial_he_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_i0e_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_i1_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_i1e_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_laguerre_polynomial_l_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_legendre_polynomial_p_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_log_ndtr_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_modified_bessel_i0_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_modified_bessel_i1_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_modified_bessel_k0_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_modified_bessel_k1_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_ndtri_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_scaled_modified_bessel_k0_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_scaled_modified_bessel_k1_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_t_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_u_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_v_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_w_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_spherical_bessel_j0_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_xlog1py_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/special_zeta_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/split_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/split_with_sizes_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/sqrt_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/squeeze_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/sub_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/sum_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/t_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/tan_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/tanh_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/tanh_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/threshold_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/threshold_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/topk_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/transpose_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/triangular_solve_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/tril_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/triu_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/trunc_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/unbind_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/unfold_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/unsqueeze_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/upsample_bicubic2d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/upsample_bicubic2d_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/upsample_bilinear2d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/upsample_bilinear2d_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/upsample_linear1d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/upsample_linear1d_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/upsample_nearest1d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/upsample_nearest1d_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/upsample_nearest2d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/upsample_nearest2d_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/upsample_nearest3d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/upsample_nearest3d_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/upsample_trilinear3d_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/upsample_trilinear3d_backward_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/values_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/view_as_complex_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/view_as_real_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/view_copy_compositeexplicitautogradnonfunctional_dispatch.h>
+#include <ATen/ops/xlogy_compositeexplicitautogradnonfunctional_dispatch.h>
+
+
+

lib/python3.12/site-packages/torch/include/ATen/CompositeImplicitAutogradFunctions.h

@@ -0,0 +1,29 @@
+#include <ATen/core/TensorBody.h>
+
+// TODO Undo all logic introduced for Note [Avoiding Include Cycles In Static Dispatch]
+// Code introduced to avoid cyclic dependency in static dispatch is no longer
+// needed as static dispatch logic is moved from TensorBody.h, which caused cycles in the first place,
+// to Operators.cpp for supporting multiple backends with multiple kernels.
+//
+// Note [Avoiding Include Cycles In Static Dispatch]
+// In order to avoid #include cycles in the static dispatch build, we've carefully split out
+// the static function definition files into {DispatchKey}Functions.h and {DispatchKey}Functions_inl.h.
+//
+// Without this split, the include cycle looks like TensorBody.h -> CPUFunctions.h -> TensorBody.h.
+// - TensorBody.h #includes CPUFunctions.h in the static dispatch build, because the tensor methods
+//   all need to call into the fastpath C++ API defined in CPUFunctions.h. The methods are also all
+//   directly inlined into TensorBody.h.
+// - CPUFunctions.h #includes TensorBody.h because it contains function declarations for the entire C++ API,
+//   which include functions that have defaultable std::optional<Tensor> arguments.
+//   That requires knowing the full Tensor class definition.
+//
+// We break the cycle by doing the following:
+// - Split out CPUFunction.h into two files: CPUFunctions.h and CPUFunctions_inl.h
+// - CPUFunction.h is a dummy file that just includes the Tensor class and includes CPUFunctions_inl.,
+// - CPUFunctions_inl.h includes everything else
+// - (only in the static dispatch build) TensorBody.h makes sure to finish defining the Tensor class,
+//   and then it includes CPUFunctions_inl.h.
+// - All other files that want the cpu fastpath functions can include CPUFunctions.h directly.
+// - This also means that static dispatch build, CPUFunctions.h only needs to
+//   #include TensorBody.h, and it will automatically bring in CPUFunctions_inl.h.
+#include <ATen/CompositeImplicitAutogradFunctions_inl.h>

lib/python3.12/site-packages/torch/include/ATen/CompositeImplicitAutogradNestedTensorFunctions.h

@@ -0,0 +1,29 @@
+#include <ATen/core/TensorBody.h>
+
+// TODO Undo all logic introduced for Note [Avoiding Include Cycles In Static Dispatch]
+// Code introduced to avoid cyclic dependency in static dispatch is no longer
+// needed as static dispatch logic is moved from TensorBody.h, which caused cycles in the first place,
+// to Operators.cpp for supporting multiple backends with multiple kernels.
+//
+// Note [Avoiding Include Cycles In Static Dispatch]
+// In order to avoid #include cycles in the static dispatch build, we've carefully split out
+// the static function definition files into {DispatchKey}Functions.h and {DispatchKey}Functions_inl.h.
+//
+// Without this split, the include cycle looks like TensorBody.h -> CPUFunctions.h -> TensorBody.h.
+// - TensorBody.h #includes CPUFunctions.h in the static dispatch build, because the tensor methods
+//   all need to call into the fastpath C++ API defined in CPUFunctions.h. The methods are also all
+//   directly inlined into TensorBody.h.
+// - CPUFunctions.h #includes TensorBody.h because it contains function declarations for the entire C++ API,
+//   which include functions that have defaultable std::optional<Tensor> arguments.
+//   That requires knowing the full Tensor class definition.
+//
+// We break the cycle by doing the following:
+// - Split out CPUFunction.h into two files: CPUFunctions.h and CPUFunctions_inl.h
+// - CPUFunction.h is a dummy file that just includes the Tensor class and includes CPUFunctions_inl.,
+// - CPUFunctions_inl.h includes everything else
+// - (only in the static dispatch build) TensorBody.h makes sure to finish defining the Tensor class,
+//   and then it includes CPUFunctions_inl.h.
+// - All other files that want the cpu fastpath functions can include CPUFunctions.h directly.
+// - This also means that static dispatch build, CPUFunctions.h only needs to
+//   #include TensorBody.h, and it will automatically bring in CPUFunctions_inl.h.
+#include <ATen/CompositeImplicitAutogradNestedTensorFunctions_inl.h>

lib/python3.12/site-packages/torch/include/ATen/CompositeImplicitAutogradNestedTensorFunctions_inl.h

@@ -0,0 +1,25 @@
+#pragma once
+// @generated by torchgen/gen.py from DispatchKeyFunctions_inl.h
+
+// NB: The implementing C++ file is RegisterDispatchKey.cpp
+
+// The only #includes we need are for custom classes that have defaults in the C++ API
+#include <c10/core/MemoryFormat.h>
+#include <c10/core/Scalar.h>
+#include <ATen/core/Reduction.h>
+
+#if defined(AT_PER_OPERATOR_HEADERS) && defined(TORCH_ASSERT_ONLY_METHOD_OPERATORS)
+#error This change adds a dependency on all pytorch operators, meaning the     \
+  file will need to be re-compiled every time an operator is changed or added. \
+  Consider including a specific operator from                                  \
+  <ATen/ops/{my_operator}_compositeimplicitautogradnestedtensor_dispatch.h>.                   \
+  See NOTE [TORCH_ASSERT_ONLY_METHOD_OPERATORS].
+#endif
+
+#include <ATen/ops/randn_like_compositeimplicitautogradnestedtensor_dispatch.h>
+#include <ATen/ops/reshape_compositeimplicitautogradnestedtensor_dispatch.h>
+#include <ATen/ops/reshape_as_compositeimplicitautogradnestedtensor_dispatch.h>
+#include <ATen/ops/zeros_like_compositeimplicitautogradnestedtensor_dispatch.h>
+
+
+

lib/python3.12/site-packages/torch/include/ATen/Config.h

@@ -0,0 +1,21 @@
+#pragma once
+
+// Test these using #if AT_MKL_ENABLED(), not #ifdef, so that it's
+// obvious if you forgot to include Config.h
+//    c.f. https://stackoverflow.com/questions/33759787/generating-an-error-if-checked-boolean-macro-is-not-defined
+//
+// DO NOT put the macros for CUDA libraries in this file; they belong in cuda/CUDAConfig.h
+
+#define AT_MKLDNN_ENABLED() 1
+#define AT_MKLDNN_ACL_ENABLED() 0
+#define AT_MKL_ENABLED() 1
+#define AT_MKL_SEQUENTIAL() 0
+#define AT_POCKETFFT_ENABLED() 0
+#define AT_NNPACK_ENABLED() 1
+#define CAFFE2_STATIC_LINK_CUDA() 0
+#define AT_BUILD_WITH_BLAS() 1
+#define AT_BUILD_WITH_LAPACK() 1
+#define AT_PARALLEL_OPENMP 1
+#define AT_PARALLEL_NATIVE 0
+#define AT_BLAS_F2C() 0
+#define AT_BLAS_USE_CBLAS_DOT() 0

lib/python3.12/site-packages/torch/include/ATen/Context.h

@@ -0,0 +1,627 @@
+#pragma once
+
+#include <ATen/BlasBackend.h>
+#include <ATen/CPUGeneratorImpl.h>
+#include <ATen/DeviceAccelerator.h>
+#include <ATen/LinalgBackend.h>
+#include <ATen/SDPBackend.h>
+#include <ATen/core/ATenGeneral.h>
+#include <ATen/core/DeprecatedTypeProperties.h>
+#include <ATen/core/Generator.h>
+#include <ATen/core/LegacyTypeDispatch.h>
+#include <ATen/detail/AcceleratorHooksInterface.h>
+#include <ATen/detail/CUDAHooksInterface.h>
+#include <ATen/detail/HIPHooksInterface.h>
+#include <ATen/detail/HPUHooksInterface.h>
+#include <ATen/detail/IPUHooksInterface.h>
+#include <ATen/detail/MAIAHooksInterface.h>
+#include <ATen/detail/MPSHooksInterface.h>
+#include <ATen/detail/MTIAHooksInterface.h>
+#include <ATen/detail/PrivateUse1HooksInterface.h>
+#include <ATen/detail/XPUHooksInterface.h>
+#include <c10/core/QEngine.h>
+#include <c10/core/impl/DeviceGuardImplInterface.h>
+#include <c10/util/CallOnce.h>
+#include <c10/util/Exception.h>
+#include <c10/util/env.h>
+#include <c10/util/irange.h>
+
+#include <cstdint>
+#include <mutex>
+
+namespace at {
+
+class Tensor;
+
+enum class TORCH_API Float32MatmulPrecision { HIGHEST, HIGH, MEDIUM };
+
+class TORCH_API Context {
+ public:
+  Context();
+
+  const Generator& defaultGenerator(Device device) {
+    c10::DeviceType device_type = device.type();
+    lazyInitDevice(device_type);
+
+    if (device_type == at::kCPU) {
+      return at::detail::getDefaultCPUGenerator();
+    } else {
+      return getAcceleratorHooksInterface(device_type)
+          .getDefaultGenerator(device.index());
+    }
+  }
+
+  const AcceleratorHooksInterface& getAcceleratorHooksInterface(
+      std::optional<c10::DeviceType> opt_device_type = std::nullopt) {
+    if (!opt_device_type.has_value()) {
+      opt_device_type = at::getAccelerator(true);
+    }
+    if (opt_device_type == at::kCUDA) {
+      return at::detail::getCUDAHooks();
+    } else if (opt_device_type == at::kXPU) {
+      return at::detail::getXPUHooks();
+    } else if (opt_device_type == at::kMPS) {
+      return at::detail::getMPSHooks();
+    } else if (opt_device_type == at::kPrivateUse1) {
+      return at::detail::getPrivateUse1Hooks();
+    } else if (opt_device_type == at::kMTIA) {
+      return at::detail::getMTIAHooks();
+    } else if (opt_device_type == at::kHIP) {
+      return at::detail::getHIPHooks();
+    } else if (opt_device_type == at::kHPU) {
+      return at::detail::getHPUHooks();
+    } else {
+      TORCH_CHECK(
+          false,
+          opt_device_type.has_value()
+              ? c10::DeviceTypeName(opt_device_type.value())
+              : "None",
+          " device type not an accelerator.");
+    }
+  }
+
+  Device getDeviceFromPtr(void* data, c10::DeviceType device_type) {
+    lazyInitDevice(device_type);
+
+    if (device_type == at::kCPU) {
+      return c10::DeviceType::CPU;
+    } else {
+      return getAcceleratorHooksInterface(device_type).getDeviceFromPtr(data);
+    }
+  }
+
+  bool isPinnedPtr(
+      const void* data,
+      std::optional<c10::DeviceType> device_type = std::nullopt) {
+    auto opt_device_type =
+        device_type.has_value() ? device_type : at::getAccelerator();
+    if (!opt_device_type.has_value() || // there is no accelerator
+        !at::isAccelerator(
+            opt_device_type.value())) { // passed device not an accelerator
+      return false;
+    }
+    return getAcceleratorHooksInterface(opt_device_type).isPinnedPtr(data);
+  }
+
+  Allocator* getPinnedMemoryAllocator(
+      std::optional<c10::DeviceType> device_type = std::nullopt) {
+    return getAcceleratorHooksInterface(device_type).getPinnedMemoryAllocator();
+  }
+
+  void lazyInitDevice(c10::DeviceType device_type) {
+    if (device_type != at::kCPU) {
+      c10::call_once(init_[static_cast<int8_t>(device_type)], [&] {
+        getAcceleratorHooksInterface(device_type).init();
+      });
+    }
+  }
+
+  static bool hasOpenMP();
+  static bool hasMKL();
+  static bool hasLAPACK();
+  static bool hasMKLDNN();
+  static bool hasMAGMA() {
+    return detail::getCUDAHooks().hasMAGMA();
+  }
+  static bool hasCUDA() {
+    return detail::getCUDAHooks().hasCUDA();
+  }
+  static bool hasMTIA() {
+    return detail::getMTIAHooks().hasMTIA();
+  }
+  static bool hasCUDART() {
+    return detail::getCUDAHooks().hasCUDART();
+  }
+  static long versionCUDART() {
+    return detail::getCUDAHooks().versionCUDART();
+  }
+  static bool hasCuDNN() {
+    return detail::getCUDAHooks().hasCuDNN();
+  }
+  static long versionCuDNN() {
+    return detail::getCUDAHooks().versionCuDNN();
+  }
+  static bool hasCuSOLVER() {
+    return detail::getCUDAHooks().hasCuSOLVER();
+  }
+  static bool hasCuBLASLt() {
+    return detail::getCUDAHooks().hasCuBLASLt();
+  }
+  static bool hasROCM() {
+    return detail::getCUDAHooks().hasROCM();
+  }
+  static bool hasHIP() {
+    return detail::getHIPHooks().hasHIP();
+  }
+  static bool hasMPS() {
+    return detail::getMPSHooks().hasMPS();
+  }
+  static bool hasIPU() {
+    return c10::impl::hasDeviceGuardImpl(c10::DeviceType::IPU);
+  }
+  static bool hasXLA() {
+    return c10::impl::hasDeviceGuardImpl(c10::DeviceType::XLA);
+  }
+  static bool hasXPU() {
+    return detail::getXPUHooks().hasXPU();
+  }
+  static bool hasLazy() {
+    return c10::impl::hasDeviceGuardImpl(c10::DeviceType::Lazy);
+  }
+  static bool hasMAIA() {
+    return c10::impl::hasDeviceGuardImpl(c10::DeviceType::MAIA);
+  }
+  static bool hasHPU() {
+    return detail::getHPUHooks().hasHPU();
+  }
+
+  static const at::cuda::NVRTC& getNVRTC() {
+    return detail::getCUDAHooks().nvrtc();
+  }
+
+  static bool setFlushDenormal(bool on);
+
+  // NB: This method is *purely* whether or not a user requested
+  // that CuDNN was enabled, it doesn't actually say anything about
+  // whether or not CuDNN is actually usable.  Use cudnn_is_acceptable
+  // to test this instead
+  bool userEnabledCuDNN() const;
+  void setUserEnabledCuDNN(bool e);
+  bool userEnabledMkldnn() const;
+  void setUserEnabledMkldnn(bool e);
+  bool benchmarkCuDNN() const;
+  void setBenchmarkCuDNN(bool);
+  int benchmarkLimitCuDNN() const;
+  void setBenchmarkLimitCuDNN(int);
+  bool deterministicCuDNN() const;
+  void setDeterministicCuDNN(bool);
+  bool deterministicMkldnn() const;
+  void setDeterministicMkldnn(bool);
+  bool userEnabledNNPACK() const;
+  void setUserEnabledNNPACK(bool e);
+
+  // Note [Disabling Fused SDP Kernels]
+  // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+  // Flash and Memory Efficient SDP kernels are enabled by default.
+  // However, they can be disabled by setting
+  // at::globalContext().setUserEnabledFlashSDP(false) flag.
+  // This is useful for debugging purposes. For example, if you want to
+  // compare the performance of the flash SDP kernels with the unfused
+  // kernel, you can disable the flash SDP kernels. By disabling
+  // the math SDP kernel, you can force your code to use flash kernels.
+  // The math SDP kernel can be disabled by setting
+  // at::globalContext().setUserEnabledMathSDP(false) flag.
+  void setSDPPriorityOrder(const std::vector<int64_t>& order);
+  std::array<at::SDPBackend, at::num_sdp_backends> sDPPriorityOrder();
+
+  void setSDPUseFlash(bool);
+  bool userEnabledFlashSDP() const;
+
+  void setSDPUseMemEfficient(bool);
+  bool userEnabledMemEfficientSDP() const;
+
+  void setSDPUseMath(bool);
+  bool userEnabledMathSDP() const;
+
+  void setSDPUseCuDNN(bool);
+  bool userEnabledCuDNNSDP() const;
+
+  void setAllowFP16BF16ReductionMathSDP(bool);
+  bool allowFP16BF16ReductionMathSDP() const;
+
+  void setSDPUseOverrideable(bool);
+  bool userEnabledOverrideableSDP() const;
+
+  at::LinalgBackend linalgPreferredBackend() const;
+  void setLinalgPreferredBackend(at::LinalgBackend);
+
+  at::BlasBackend blasPreferredBackend();
+  void setBlasPreferredBackend(at::BlasBackend);
+
+  // Note [Enabling Deterministic Operations]
+  // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+  // Operations in PyTorch that normally act nondeterministically, but have an
+  // alternate deterministic implementation, should satisfy the following
+  // requirements:
+  //
+  // * Include this comment: "See Note [Enabling Deterministic Operations]"
+  //
+  // * Check the value of `at::globalContext().deterministicAlgorithms()` to
+  // toggle
+  //   between nondeterministic and deterministic implementations.
+  //
+  // * Have an entry in the list of PyTorch operations that toggle between
+  // nondeterministic
+  //   and deterministic implementations, in the docstring of
+  //   `use_deterministic_algorithms()` in torch/__init__.py
+  //
+  // `example_func()` below shows an example of toggling between
+  // nondeterministic and deterministic implementations:
+  //
+  //    void example_func() {
+  //      // See Note [Enabling Deterministic Operations]
+  //      if (at::globalContext().deterministicAlgorithms()) {
+  //        example_func_deterministic();
+  //      } else {
+  //        example_func_nondeterministic();
+  //      }
+  //    }
+
+  bool deterministicAlgorithms() const;
+  bool deterministicAlgorithmsWarnOnly() const;
+  void setDeterministicAlgorithms(bool, bool);
+  bool deterministicFillUninitializedMemory() const;
+  void setDeterministicFillUninitializedMemory(bool);
+
+  // Note [Writing Nondeterministic Operations]
+  // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+  // Operations in PyTorch that act nondeterministically and do not have an
+  // alternate deterministic implementation should satisfy the following
+  // requirements:
+  //
+  // * Include this comment: "See Note [Writing Nondeterministic Operations]"
+  //
+  // * Include a comment explaining why the operation is nondeterministic.
+  //
+  // * Throw an error when `Context::deterministicAlgorithms()` is true. Most
+  //   of the time, this should be accomplished by calling
+  //   `at::globalContext().alertNotDeterminstic()`.  However, if the
+  //   nondeterministic behavior is caused by the CuBLAS workspace
+  //   configuration in CUDA >= 10.2,
+  //   `at::globalContext().alertCuBLASConfigNotDeterministic()` should be
+  //   called instead (in this case, a comment explaining why the operation is
+  //   nondeterministic is not necessary). See below for details on these
+  //   methods.
+  //
+  // * Have an entry in the list of nondeterministic PyTorch operations in the
+  //   docstring of `use_deterministic_algorithms()` in torch/__init__.py
+  //
+  // * Have a test function in `test/test_torch.py` whose name begins with
+  //   `test_nondeterministic_alert_`. Alternatively, if CuBLAS workspace
+  //   configuration is the reason for nondeterminism, the operation should be
+  //   included in the `test_cublas_config_nondeterministic_alert` test. Any new
+  //   tests should ideally follow a pattern similar to the existing ones.
+  //
+  // `example_func()` below shows an example of the comments and error-throwing
+  // code for a nondeterministic operation:
+  //
+  //    void example_func() {
+  //      // See Note [Writing Nondeterministic Operations]
+  //      // Nondeterministic because <reason>
+  //      at::globalContext().alertNondeterministic("example_func");
+  //      ...
+  //    }
+
+  // Throws an error if `Context::deterministicAlgorithms()` is true
+  static void alertNotDeterministic(std::string_view const& caller);
+
+  // Throws an error if `Context::deterministicAlgorithms()` is true, CUDA
+  // >= 10.2, and CUBLAS_WORKSPACE_CONFIG is not set to either ":16:8" or
+  // ":4096:8". For more details:
+  // https://docs.nvidia.com/cuda/cublas/index.html#results-reproducibility
+  void alertCuBLASConfigNotDeterministic() const;
+
+  void setFloat32MatmulPrecision(const std::string& s);
+  bool allowTF32CuDNN() const;
+  void setAllowTF32CuDNN(bool);
+  bool allowTF32CuBLAS() const;
+  void setAllowTF32CuBLAS(bool);
+  Float32MatmulPrecision float32MatmulPrecision() const;
+  void setFloat32MatmulPrecision(Float32MatmulPrecision p);
+  bool allowFP16ReductionCuBLAS() const;
+  void setAllowFP16ReductionCuBLAS(bool);
+  bool allowBF16ReductionCuBLAS() const;
+  void setAllowBF16ReductionCuBLAS(bool);
+  at::QEngine qEngine() const;
+  void setQEngine(at::QEngine e);
+  static const std::vector<at::QEngine>& supportedQEngines();
+  static bool isXNNPACKAvailable();
+  void setCheckSparseTensorInvariants(bool e);
+  bool checkSparseTensorInvariants() const;
+  // This method is used to release the original weight after pre-packing.
+  // It should be called once before loading/running the model.
+  // NB: By default it is set to true for mobile builds.
+  void setReleaseWeightsWhenPrepacking(bool e);
+  bool releaseWeightsWhenPrepacking() const;
+
+  void setDisplayVmapFallbackWarnings(bool enabled);
+  bool areVmapFallbackWarningsEnabled() const;
+
+  void setDefaultMobileCPUAllocator();
+  void unsetDefaultMobileCPUAllocator();
+  bool allowFP16ReductionCPU() const;
+  void setAllowFP16ReductionCPU(bool);
+
+  // Preserved for BC
+  void lazyInitCUDA() {
+    TORCH_WARN_DEPRECATION(
+        "lazyInitCUDA is deprecated. Please use lazyInitDevice(at::kCUDA) instead.")
+    lazyInitDevice(at::kCUDA);
+  }
+  void lazyInitHIP() {
+    TORCH_WARN_DEPRECATION(
+        "lazyInitHIP is deprecated. Please use lazyInitDevice(at::kHIP) instead.")
+    lazyInitDevice(at::kHIP);
+  }
+  void lazyInitXPU() {
+    TORCH_WARN_DEPRECATION(
+        "lazyInitXPU is deprecated. Please use lazyInitDevice(at::kXPU) instead.")
+    lazyInitDevice(at::kXPU);
+  }
+  void lazyInitMTIA() {
+    TORCH_WARN_DEPRECATION(
+        "lazyInitMTIA is deprecated. Please use lazyInitDevice(at::kMTIA) instead.")
+    lazyInitDevice(at::kMTIA);
+  }
+  void lazyInitPrivateUse1() {
+    TORCH_WARN_DEPRECATION(
+        "lazyInitPrivateUse1 is deprecated. Please use lazyInitDevice(at::kPrivateUse1) instead.")
+    lazyInitDevice(at::kPrivateUse1);
+  }
+
+ private:
+  static bool checkCuBLASConfigDeterministic();
+  std::array<c10::once_flag, at::COMPILE_TIME_MAX_DEVICE_TYPES> init_;
+  bool enabled_cudnn = true;
+  bool deterministic_cudnn = false;
+  bool deterministic_mkldnn = false;
+  bool _deterministic_algorithms = false;
+  bool _deterministic_algorithms_warn_only = false;
+  bool _deterministic_fill_uninitialized_memory = true;
+  std::array<at::SDPBackend, at::num_sdp_backends> sdp_priority_order = {
+      at::SDPBackend::flash_attention,
+      at::SDPBackend::efficient_attention,
+      at::SDPBackend::math,
+      at::SDPBackend::cudnn_attention};
+  bool enabled_flashSDP = true;
+  bool enabled_mem_efficientSDP = true;
+  bool enabled_mathSDP = true;
+  bool enabled_cudnnSDP = true;
+  bool enabled_overrideable = true;
+  bool allow_fp16_bf16_reduction_mathSDP = false;
+#ifdef USE_ROCM
+  bool benchmark_cudnn = true;
+#else
+  bool benchmark_cudnn = false;
+#endif
+  Float32MatmulPrecision float32_matmul_precision =
+      c10::utils::check_env("TORCH_ALLOW_TF32_CUBLAS_OVERRIDE") == true
+      ? at::Float32MatmulPrecision::HIGH
+      : at::Float32MatmulPrecision::HIGHEST;
+  int benchmark_limit_cudnn = 10;
+  bool allow_tf32_cudnn = true;
+  bool allow_fp16_reduction_cublas = true;
+  bool allow_bf16_reduction_cublas = true;
+  bool enabled_mkldnn = true;
+  bool enabled_nnpack = true;
+  at::LinalgBackend linalg_preferred_backend =
+      c10::utils::check_env("TORCH_LINALG_PREFER_CUSOLVER") == true
+      ? at::LinalgBackend::Cusolver
+      : at::LinalgBackend::Default;
+  at::BlasBackend blas_preferred_backend =
+#ifdef USE_ROCM
+      (c10::utils::check_env("TORCH_BLAS_PREFER_HIPBLASLT") != false)
+#else
+      (c10::utils::check_env("TORCH_BLAS_PREFER_CUBLASLT") == true)
+#endif
+      ? at::BlasBackend::Cublaslt
+      : at::BlasBackend::Cublas;
+#ifdef C10_MOBILE
+  bool release_original_weights = true;
+#else
+  bool release_original_weights = false;
+#endif
+  bool display_vmap_fallback_warnings_ = false;
+  std::optional<at::QEngine> quantized_engine = std::nullopt;
+  bool enable_sparse_tensor_invariant_checks = false;
+  bool allow_fp16_reduction_cpu = false;
+
+  Allocator* prev_allocator_ptr_{nullptr};
+};
+
+TORCH_API Context& globalContext();
+
+inline void init() {
+  globalContext();
+}
+
+TORCH_API Allocator* getCPUAllocator();
+
+inline DeprecatedTypeProperties& getDeprecatedTypeProperties(
+    Backend p,
+    ScalarType s) {
+  return globalDeprecatedTypePropertiesRegistry().getDeprecatedTypeProperties(
+      p, s);
+}
+
+inline DeprecatedTypeProperties& CPU(ScalarType s) {
+  return globalDeprecatedTypePropertiesRegistry().getDeprecatedTypeProperties(
+      Backend::CPU, s);
+}
+
+inline DeprecatedTypeProperties& CUDA(ScalarType s) {
+  return globalDeprecatedTypePropertiesRegistry().getDeprecatedTypeProperties(
+      Backend::CUDA, s);
+}
+
+inline DeprecatedTypeProperties& HIP(ScalarType s) {
+  return globalDeprecatedTypePropertiesRegistry().getDeprecatedTypeProperties(
+      Backend::HIP, s);
+}
+
+inline DeprecatedTypeProperties& MPS(ScalarType s) {
+  return globalDeprecatedTypePropertiesRegistry().getDeprecatedTypeProperties(
+      Backend::MPS, s);
+}
+
+inline bool hasCUDA() {
+  return globalContext().hasCUDA();
+}
+
+inline bool hasMTIA() {
+  return globalContext().hasMTIA();
+}
+
+inline bool hasHIP() {
+  return globalContext().hasHIP();
+}
+
+inline bool hasIPU() {
+  return globalContext().hasIPU();
+}
+
+inline bool hasXLA() {
+  return globalContext().hasXLA();
+}
+
+inline bool hasMPS() {
+  return globalContext().hasMPS();
+}
+
+inline bool hasMAIA() {
+  return globalContext().hasMAIA();
+}
+
+inline bool hasXPU() {
+  return globalContext().hasXPU();
+}
+
+inline bool hasHPU() {
+  return globalContext().hasHPU();
+}
+
+// Despite its name, this function returns the number of *CUDA* GPUs.
+inline size_t getNumGPUs() {
+  // WARNING: DO NOT ADD LOGIC TO HANDLE OTHER DEVICE TYPES TO THIS
+  // FUNCTION.  If you are interested in interrogating the number of
+  // devices for a specific device type, add that function to the
+  // relevant library (e.g., similar to at::cuda::device_count())
+  if (hasCUDA() && hasHIP()) {
+    throw std::runtime_error(
+        "Enabling both CUDA and HIP in ATen is not supported, as HIP masquerades "
+        "to be CUDA (e.g., when you say CUDA, on a HIP build of ATen, this actually "
+        "means HIP.  Rebuild PyTorch with one or the other disabled.");
+  } else if (hasCUDA()) {
+    return detail::getCUDAHooks().deviceCount();
+  } else if (hasHIP()) {
+    return detail::getHIPHooks().getNumGPUs();
+  } else {
+    return 0;
+  }
+}
+
+inline bool hasOpenMP() {
+  return globalContext().hasOpenMP();
+}
+
+inline bool hasMKL() {
+  return globalContext().hasMKL();
+}
+
+inline bool hasLAPACK() {
+  return globalContext().hasLAPACK();
+}
+
+inline bool hasMAGMA() {
+  return globalContext().hasMAGMA();
+}
+
+inline bool hasMKLDNN() {
+  return globalContext().hasMKLDNN();
+}
+
+inline void manual_seed(uint64_t seed) {
+  auto gen = globalContext().defaultGenerator(c10::DeviceType::CPU);
+  {
+    // See Note [Acquire lock when using random generators]
+    std::lock_guard<std::mutex> lock(gen.mutex());
+    gen.set_current_seed(seed);
+  }
+  // NB: Sometimes we build with CUDA, but we don't have any GPUs
+  // available. In that case, we must not seed CUDA; it will fail!
+  const auto cuda_num_gpus = detail::getCUDAHooks().deviceCount();
+  if (hasCUDA() && cuda_num_gpus > 0) {
+    for (const auto i : c10::irange(cuda_num_gpus)) {
+      auto cuda_gen = globalContext().defaultGenerator(
+          Device(at::kCUDA, static_cast<c10::DeviceIndex>(i)));
+      {
+        // See Note [Acquire lock when using random generators]
+        std::lock_guard<std::mutex> lock(cuda_gen.mutex());
+        cuda_gen.set_current_seed(seed);
+      }
+    }
+  }
+
+  const auto xpu_num_gpus = detail::getXPUHooks().deviceCount();
+  if (hasXPU() && xpu_num_gpus) {
+    for (const auto i : c10::irange(xpu_num_gpus)) {
+      auto xpu_gen = globalContext().defaultGenerator(
+          Device(at::kXPU, static_cast<c10::DeviceIndex>(i)));
+      {
+        // See Note [Acquire lock when using random generators]
+        std::lock_guard<std::mutex> lock(xpu_gen.mutex());
+        xpu_gen.set_current_seed(seed);
+      }
+    }
+  }
+
+  if (hasMPS()) {
+    auto mps_gen = globalContext().defaultGenerator(c10::DeviceType::MPS);
+    // See Note [Acquire lock when using random generators]
+    std::lock_guard<std::mutex> lock(mps_gen.mutex());
+    mps_gen.set_current_seed(seed);
+  }
+}
+
+// When the global flag `allow_tf32` is set to true, cuBLAS handles are
+// automatically configured to use math mode CUBLAS_TF32_TENSOR_OP_MATH.
+// For some operators, such as addmv, TF32 offers no performance improvement
+// but causes precision loss. To help this case, this class implements
+// a RAII guard that can be used to quickly disable TF32 within its scope.
+//
+// Usage:
+//     NoTF32Guard disable_tf32;
+struct TORCH_API NoTF32Guard {
+  NoTF32Guard();
+  NoTF32Guard(NoTF32Guard&& other) = delete;
+  NoTF32Guard(const NoTF32Guard&) = delete;
+  NoTF32Guard& operator=(const NoTF32Guard&) = delete;
+  NoTF32Guard& operator=(NoTF32Guard&&) = delete;
+  ~NoTF32Guard();
+  static bool should_disable_tf32();
+
+ private:
+  bool changed = false;
+};
+
+struct TORCH_API ROCmBackwardPassGuard {
+  ROCmBackwardPassGuard();
+  ROCmBackwardPassGuard(ROCmBackwardPassGuard&& other) = delete;
+  ROCmBackwardPassGuard(const ROCmBackwardPassGuard&) = delete;
+  ROCmBackwardPassGuard& operator=(const ROCmBackwardPassGuard&) = delete;
+  ROCmBackwardPassGuard& operator=(ROCmBackwardPassGuard&&) = delete;
+  ~ROCmBackwardPassGuard();
+  static bool is_backward_pass();
+};
+
+} // namespace at

lib/python3.12/site-packages/torch/include/ATen/DLConvertor.h

@@ -0,0 +1,21 @@
+#pragma once
+
+#include <ATen/ATen.h>
+#include <ATen/Tensor.h>
+#include <ATen/dlpack.h>
+
+// this convertor will:
+// 1) take a Tensor object and wrap it in the DLPack tensor
+// 2) take a dlpack tensor and convert it to the ATen Tensor
+
+namespace at {
+
+TORCH_API ScalarType toScalarType(const DLDataType& dtype);
+TORCH_API DLManagedTensor* toDLPack(const Tensor& src);
+TORCH_API Tensor fromDLPack(DLManagedTensor* src);
+TORCH_API Tensor
+fromDLPack(DLManagedTensor* src, std::function<void(void*)> deleter);
+TORCH_API DLDataType getDLDataType(const Tensor& t);
+TORCH_API DLDevice getDLContext(const Tensor& tensor, const int64_t& device_id);
+
+} // namespace at

lib/python3.12/site-packages/torch/include/ATen/Device.h

@@ -0,0 +1,2 @@
+#pragma once
+#include <c10/core/Device.h>

lib/python3.12/site-packages/torch/include/ATen/DeviceAccelerator.h

@@ -0,0 +1,27 @@
+#pragma once
+
+#include <c10/core/DeviceType.h>
+#include <c10/macros/Macros.h>
+
+#include <ATen/detail/MTIAHooksInterface.h>
+#include <optional>
+
+// This file defines the top level Accelerator concept for PyTorch.
+// A device is an accelerator per the definition here if:
+// - It is mutually exclusive with all other accelerators
+// - It performs asynchronous compute via a Stream/Event system
+// - It provides a set of common APIs as defined by AcceleratorHooksInterface
+//
+// As of today, accelerator devices are (in no particular order):
+// CUDA, MTIA, XPU, HIP, MPS, PrivateUse1
+
+namespace at {
+
+// Ensures that only one accelerator is available (at
+// compile time if possible) and return it.
+// When checked is true, the returned optional always has a value.
+TORCH_API std::optional<c10::DeviceType> getAccelerator(bool checked = false);
+
+TORCH_API bool isAccelerator(c10::DeviceType d);
+
+} // namespace at

lib/python3.12/site-packages/torch/include/ATen/DeviceGuard.h

@@ -0,0 +1,41 @@
+#pragma once
+
+#include <ATen/core/IListRef.h>
+#include <ATen/core/Tensor.h>
+#include <c10/core/DeviceGuard.h>
+#include <c10/core/ScalarType.h> // TensorList whyyyyy
+
+namespace at {
+
+// Are you here because you're wondering why DeviceGuard(tensor) no
+// longer works?  For code organization reasons, we have temporarily(?)
+// removed this constructor from DeviceGuard.  The new way to
+// spell it is:
+//
+//    OptionalDeviceGuard guard(device_of(tensor));
+
+/// Return the Device of a Tensor, if the Tensor is defined.
+inline std::optional<Device> device_of(const Tensor& t) {
+  if (t.defined()) {
+    return t.device();
+  } else {
+    return std::nullopt;
+  }
+}
+
+inline std::optional<Device> device_of(const std::optional<Tensor>& t) {
+  return t.has_value() ? device_of(t.value()) : std::nullopt;
+}
+
+/// Return the Device of a TensorList, if the list is non-empty and
+/// the first Tensor is defined.  (This function implicitly assumes
+/// that all tensors in the list have the same device.)
+inline std::optional<Device> device_of(ITensorListRef t) {
+  if (!t.empty()) {
+    return device_of(t.front());
+  } else {
+    return std::nullopt;
+  }
+}
+
+} // namespace at

lib/python3.12/site-packages/torch/include/ATen/DimVector.h

@@ -0,0 +1,2 @@
+#pragma once
+#include <ATen/core/DimVector.h>

lib/python3.12/site-packages/torch/include/ATen/Dimname.h

@@ -0,0 +1 @@
+#include <ATen/core/Dimname.h>

lib/python3.12/site-packages/torch/include/ATen/Dispatch_v2.h

@@ -0,0 +1,202 @@
+#include <ATen/Dispatch.h>
+
+// This is a new implementation of the AT_DISPATCH macro family from
+// ATen/Dispatch.h
+//
+// The intended usage is:
+//
+//  ScalarType scalar_type;
+//
+//  AT_DISPATCH_V2(
+//    scalar_type,
+//    "debug string",
+//    AT_WRAP([&] {
+//      ... code to specialize with scalar_t ...
+//    }),
+//    kHalf,
+//    AT_EXPAND(AT_ALL_TYPES),
+//    ... as many types arguments as needed ...
+//  )
+//
+// For example, given an old style:
+//
+//  AT_DISPATCH_ALL_TYPES_AND_COMPLEX_AND2(
+//    kComplexHalf,
+//    kHalf,
+//    self.scalar_type(),
+//    "_local_scalar_dense_cpu",
+//    [&] {
+//      scalar_t value = *self.data_ptr<scalar_t>();
+//      r = Scalar(value);
+//    }
+//  )
+//
+// You now write:
+//
+//  AT_DISPATCH_V2(
+//    self.scalar_type(),
+//    "_local_scalar_dense_cpu",
+//    AT_WRAP([&] {
+//      scalar_t value = *self.data_ptr<scalar_t>();
+//      r = Scalar(value);
+//    }),
+//    AT_EXPAND(AT_ALL_TYPES),
+//    AT_EXPAND(AT_COMPLEX_TYPES),
+//    kComplexHalf,
+//    kHalf,
+//  )
+//
+// Notably, it sports the following improvements:
+//
+//  - It is not necessary to specify the arity (e.g.,
+//    AT_DISPATCH_FLOATING_AND_COMPLEX_TYPES_AND{2,3,4,...})
+//    when using the macro
+//
+//  - It is not necessary to specify each dtype individually; if
+//    there is a set of related dtypes and you want to dispatch
+//    over all of them, you can simply say, e.g., AT_EXPAND(AT_INTEGRAL_TYPES)
+//    in your argument list.
+//
+// However, you must remember to wrap the payload body in AT_WRAP, or commas
+// inside your lambda will be improperly handled.  Furthermore, if you more
+// entries to ScalarType than can be supported by this macro, it will fail
+// with an obscure error (due to attempting to concatenate AT_AP with
+// something that is not a number).
+//
+// The implementation strategy is to use the count arguments trick
+// (e.g., as described in https://stackoverflow.com/a/2124385/23845)
+// to discover how many dtypes have been passed, and then dispatch to a
+// hand-written macro for each arity that applies as many DISPATCH_CASE as
+// necessary.  The hand-written macros can be regenerated for other arities
+// with the script below.
+//
+// There is some delicacy in the implementation in controlling when
+// macro expansion occurs, mediated with AT_EXPAND and AT_GUARD.  I mostly
+// relied on GPT4 to help me get it right.
+
+// Public API macros
+
+// See documentation above
+#define AT_DISPATCH_V2(TYPE, NAME, BODY, ...) \
+  AT_DISPATCH_SWITCH(TYPE, NAME, AT_AP_VAR(AT_WRAP(BODY), TYPE, __VA_ARGS__))
+
+// This macro lets you pass an arbitrary expression that may contain internal
+// commas to another macro without having the commas causing the expression
+// to be interpreted as being multiple arguments
+#define AT_WRAP(...) __VA_ARGS__
+
+#define AT_FLOAT8_TYPES                                          \
+  c10::kFloat8_e5m2, c10::kFloat8_e5m2fnuz, c10::kFloat8_e4m3fn, \
+      c10::kFloat8_e4m3fnuz
+
+#define AT_INTEGRAL_TYPES \
+  c10::kByte, c10::kChar, c10::kInt, c10::kLong, c10::kShort
+#define AT_FLOATING_TYPES c10::kDouble, c10::kFloat
+#define AT_BAREBONES_UNSIGNED_TYPES c10::kUInt16, c10::kUInt32, c10::kUInt64
+#define AT_INTEGRAL_TYPES_V2 \
+  AT_EXPAND(AT_INTEGRAL_TYPES), AT_EXPAND(AT_BAREBONES_UNSIGNED_TYPES)
+#define AT_COMPLEX_TYPES c10::kComplexDouble, c10::kComplexFloat
+#define AT_QINT_TYPES c10::kQInt8, c10::kQUInt8, c10::kQInt32
+// NB: not *actually* all types
+#define AT_ALL_TYPES AT_EXPAND(AT_INTEGRAL_TYPES), AT_EXPAND(AT_FLOATING_TYPES)
+#define AT_ALL_TYPES_AND_COMPLEX \
+  AT_EXPAND(AT_ALL_TYPES), AT_EXPAND(AT_COMPLEX_TYPES)
+
+// Helper macros
+
+#define AT_AP_VAR(N, T, ...) \
+  AT_EXPAND(AT_CONCAT(AT_AP, AT_NUM_ARGS(__VA_ARGS__))(AT_WRAP(N), __VA_ARGS__))
+#define AT_CONCAT(a, b) AT_CONCAT_AUX(a, b)
+#define AT_CONCAT_AUX(a, b) a##b
+#define AT_EXPAND(X) X
+
+// Ensure we never have too many scalar types for the expansion here to
+// support.  To bump this, you must regenerate the macros below.
+static_assert(static_cast<int>(c10::ScalarType::NumOptions) < 60);
+
+// Python code to regenerate generate code below:
+#if 0
+
+num_args = 60
+
+nums = ', '.join(str(i) for i in reversed(range(num_args+1)))
+args = ', '.join(f'_{i}' for i in range(1, num_args+1))
+
+print(f'#define AT_NUM_ARGS(...) AT_EXPAND(AT_NUM_ARGS_AUX(__VA_ARGS__, {nums}))')
+print(f'#define AT_NUM_ARGS_AUX({args}, N, ...) N')
+
+for i in range(1, num_args+1):
+    args = ', '.join(f'_{i}' for i in range(1, i+1))
+    cases = ' '.join([f'AT_DISPATCH_CASE(_{j}, N)' for j in range(1, i+1)])
+    print(f'#define AT_AP{i}(N, {args}) {cases}')
+
+#endif
+
+// Begin generated code
+// clang-format off
+
+#define AT_NUM_ARGS(...) AT_EXPAND(AT_NUM_ARGS_AUX(__VA_ARGS__, 60, 59, 58, 57, 56, 55, 54, 53, 52, 51, 50, 49, 48, 47, 46, 45, 44, 43, 42, 41, 40, 39, 38, 37, 36, 35, 34, 33, 32, 31, 30, 29, 28, 27, 26, 25, 24, 23, 22, 21, 20, 19, 18, 17, 16, 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0))
+#define AT_NUM_ARGS_AUX(_1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43, _44, _45, _46, _47, _48, _49, _50, _51, _52, _53, _54, _55, _56, _57, _58, _59, _60, N, ...) N
+#define AT_AP1(N, _1) AT_DISPATCH_CASE(_1, N)
+#define AT_AP2(N, _1, _2) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N)
+#define AT_AP3(N, _1, _2, _3) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N)
+#define AT_AP4(N, _1, _2, _3, _4) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N)
+#define AT_AP5(N, _1, _2, _3, _4, _5) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N)
+#define AT_AP6(N, _1, _2, _3, _4, _5, _6) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N)
+#define AT_AP7(N, _1, _2, _3, _4, _5, _6, _7) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N)
+#define AT_AP8(N, _1, _2, _3, _4, _5, _6, _7, _8) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N)
+#define AT_AP9(N, _1, _2, _3, _4, _5, _6, _7, _8, _9) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N)
+#define AT_AP10(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N)
+#define AT_AP11(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N)
+#define AT_AP12(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N)
+#define AT_AP13(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N)
+#define AT_AP14(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N)
+#define AT_AP15(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N)
+#define AT_AP16(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N)
+#define AT_AP17(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N)
+#define AT_AP18(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N)
+#define AT_AP19(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N)
+#define AT_AP20(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N)
+#define AT_AP21(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N)
+#define AT_AP22(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N)
+#define AT_AP23(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N)
+#define AT_AP24(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N)
+#define AT_AP25(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N)
+#define AT_AP26(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N)
+#define AT_AP27(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N)
+#define AT_AP28(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N)
+#define AT_AP29(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N)
+#define AT_AP30(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N)
+#define AT_AP31(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N)
+#define AT_AP32(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N)
+#define AT_AP33(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N)
+#define AT_AP34(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N)
+#define AT_AP35(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N)
+#define AT_AP36(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N)
+#define AT_AP37(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N)
+#define AT_AP38(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N)
+#define AT_AP39(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N)
+#define AT_AP40(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N)
+#define AT_AP41(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N) AT_DISPATCH_CASE(_41, N)
+#define AT_AP42(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N) AT_DISPATCH_CASE(_41, N) AT_DISPATCH_CASE(_42, N)
+#define AT_AP43(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N) AT_DISPATCH_CASE(_41, N) AT_DISPATCH_CASE(_42, N) AT_DISPATCH_CASE(_43, N)
+#define AT_AP44(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43, _44) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N) AT_DISPATCH_CASE(_41, N) AT_DISPATCH_CASE(_42, N) AT_DISPATCH_CASE(_43, N) AT_DISPATCH_CASE(_44, N)
+#define AT_AP45(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43, _44, _45) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N) AT_DISPATCH_CASE(_41, N) AT_DISPATCH_CASE(_42, N) AT_DISPATCH_CASE(_43, N) AT_DISPATCH_CASE(_44, N) AT_DISPATCH_CASE(_45, N)
+#define AT_AP46(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43, _44, _45, _46) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N) AT_DISPATCH_CASE(_41, N) AT_DISPATCH_CASE(_42, N) AT_DISPATCH_CASE(_43, N) AT_DISPATCH_CASE(_44, N) AT_DISPATCH_CASE(_45, N) AT_DISPATCH_CASE(_46, N)
+#define AT_AP47(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43, _44, _45, _46, _47) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N) AT_DISPATCH_CASE(_41, N) AT_DISPATCH_CASE(_42, N) AT_DISPATCH_CASE(_43, N) AT_DISPATCH_CASE(_44, N) AT_DISPATCH_CASE(_45, N) AT_DISPATCH_CASE(_46, N) AT_DISPATCH_CASE(_47, N)
+#define AT_AP48(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43, _44, _45, _46, _47, _48) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N) AT_DISPATCH_CASE(_41, N) AT_DISPATCH_CASE(_42, N) AT_DISPATCH_CASE(_43, N) AT_DISPATCH_CASE(_44, N) AT_DISPATCH_CASE(_45, N) AT_DISPATCH_CASE(_46, N) AT_DISPATCH_CASE(_47, N) AT_DISPATCH_CASE(_48, N)
+#define AT_AP49(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43, _44, _45, _46, _47, _48, _49) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N) AT_DISPATCH_CASE(_41, N) AT_DISPATCH_CASE(_42, N) AT_DISPATCH_CASE(_43, N) AT_DISPATCH_CASE(_44, N) AT_DISPATCH_CASE(_45, N) AT_DISPATCH_CASE(_46, N) AT_DISPATCH_CASE(_47, N) AT_DISPATCH_CASE(_48, N) AT_DISPATCH_CASE(_49, N)
+#define AT_AP50(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43, _44, _45, _46, _47, _48, _49, _50) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N) AT_DISPATCH_CASE(_41, N) AT_DISPATCH_CASE(_42, N) AT_DISPATCH_CASE(_43, N) AT_DISPATCH_CASE(_44, N) AT_DISPATCH_CASE(_45, N) AT_DISPATCH_CASE(_46, N) AT_DISPATCH_CASE(_47, N) AT_DISPATCH_CASE(_48, N) AT_DISPATCH_CASE(_49, N) AT_DISPATCH_CASE(_50, N)
+#define AT_AP51(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43, _44, _45, _46, _47, _48, _49, _50, _51) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N) AT_DISPATCH_CASE(_41, N) AT_DISPATCH_CASE(_42, N) AT_DISPATCH_CASE(_43, N) AT_DISPATCH_CASE(_44, N) AT_DISPATCH_CASE(_45, N) AT_DISPATCH_CASE(_46, N) AT_DISPATCH_CASE(_47, N) AT_DISPATCH_CASE(_48, N) AT_DISPATCH_CASE(_49, N) AT_DISPATCH_CASE(_50, N) AT_DISPATCH_CASE(_51, N)
+#define AT_AP52(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43, _44, _45, _46, _47, _48, _49, _50, _51, _52) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N) AT_DISPATCH_CASE(_41, N) AT_DISPATCH_CASE(_42, N) AT_DISPATCH_CASE(_43, N) AT_DISPATCH_CASE(_44, N) AT_DISPATCH_CASE(_45, N) AT_DISPATCH_CASE(_46, N) AT_DISPATCH_CASE(_47, N) AT_DISPATCH_CASE(_48, N) AT_DISPATCH_CASE(_49, N) AT_DISPATCH_CASE(_50, N) AT_DISPATCH_CASE(_51, N) AT_DISPATCH_CASE(_52, N)
+#define AT_AP53(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43, _44, _45, _46, _47, _48, _49, _50, _51, _52, _53) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N) AT_DISPATCH_CASE(_41, N) AT_DISPATCH_CASE(_42, N) AT_DISPATCH_CASE(_43, N) AT_DISPATCH_CASE(_44, N) AT_DISPATCH_CASE(_45, N) AT_DISPATCH_CASE(_46, N) AT_DISPATCH_CASE(_47, N) AT_DISPATCH_CASE(_48, N) AT_DISPATCH_CASE(_49, N) AT_DISPATCH_CASE(_50, N) AT_DISPATCH_CASE(_51, N) AT_DISPATCH_CASE(_52, N) AT_DISPATCH_CASE(_53, N)
+#define AT_AP54(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43, _44, _45, _46, _47, _48, _49, _50, _51, _52, _53, _54) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N) AT_DISPATCH_CASE(_41, N) AT_DISPATCH_CASE(_42, N) AT_DISPATCH_CASE(_43, N) AT_DISPATCH_CASE(_44, N) AT_DISPATCH_CASE(_45, N) AT_DISPATCH_CASE(_46, N) AT_DISPATCH_CASE(_47, N) AT_DISPATCH_CASE(_48, N) AT_DISPATCH_CASE(_49, N) AT_DISPATCH_CASE(_50, N) AT_DISPATCH_CASE(_51, N) AT_DISPATCH_CASE(_52, N) AT_DISPATCH_CASE(_53, N) AT_DISPATCH_CASE(_54, N)
+#define AT_AP55(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43, _44, _45, _46, _47, _48, _49, _50, _51, _52, _53, _54, _55) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N) AT_DISPATCH_CASE(_41, N) AT_DISPATCH_CASE(_42, N) AT_DISPATCH_CASE(_43, N) AT_DISPATCH_CASE(_44, N) AT_DISPATCH_CASE(_45, N) AT_DISPATCH_CASE(_46, N) AT_DISPATCH_CASE(_47, N) AT_DISPATCH_CASE(_48, N) AT_DISPATCH_CASE(_49, N) AT_DISPATCH_CASE(_50, N) AT_DISPATCH_CASE(_51, N) AT_DISPATCH_CASE(_52, N) AT_DISPATCH_CASE(_53, N) AT_DISPATCH_CASE(_54, N) AT_DISPATCH_CASE(_55, N)
+#define AT_AP56(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43, _44, _45, _46, _47, _48, _49, _50, _51, _52, _53, _54, _55, _56) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N) AT_DISPATCH_CASE(_41, N) AT_DISPATCH_CASE(_42, N) AT_DISPATCH_CASE(_43, N) AT_DISPATCH_CASE(_44, N) AT_DISPATCH_CASE(_45, N) AT_DISPATCH_CASE(_46, N) AT_DISPATCH_CASE(_47, N) AT_DISPATCH_CASE(_48, N) AT_DISPATCH_CASE(_49, N) AT_DISPATCH_CASE(_50, N) AT_DISPATCH_CASE(_51, N) AT_DISPATCH_CASE(_52, N) AT_DISPATCH_CASE(_53, N) AT_DISPATCH_CASE(_54, N) AT_DISPATCH_CASE(_55, N) AT_DISPATCH_CASE(_56, N)
+#define AT_AP57(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43, _44, _45, _46, _47, _48, _49, _50, _51, _52, _53, _54, _55, _56, _57) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N) AT_DISPATCH_CASE(_41, N) AT_DISPATCH_CASE(_42, N) AT_DISPATCH_CASE(_43, N) AT_DISPATCH_CASE(_44, N) AT_DISPATCH_CASE(_45, N) AT_DISPATCH_CASE(_46, N) AT_DISPATCH_CASE(_47, N) AT_DISPATCH_CASE(_48, N) AT_DISPATCH_CASE(_49, N) AT_DISPATCH_CASE(_50, N) AT_DISPATCH_CASE(_51, N) AT_DISPATCH_CASE(_52, N) AT_DISPATCH_CASE(_53, N) AT_DISPATCH_CASE(_54, N) AT_DISPATCH_CASE(_55, N) AT_DISPATCH_CASE(_56, N) AT_DISPATCH_CASE(_57, N)
+#define AT_AP58(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43, _44, _45, _46, _47, _48, _49, _50, _51, _52, _53, _54, _55, _56, _57, _58) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N) AT_DISPATCH_CASE(_41, N) AT_DISPATCH_CASE(_42, N) AT_DISPATCH_CASE(_43, N) AT_DISPATCH_CASE(_44, N) AT_DISPATCH_CASE(_45, N) AT_DISPATCH_CASE(_46, N) AT_DISPATCH_CASE(_47, N) AT_DISPATCH_CASE(_48, N) AT_DISPATCH_CASE(_49, N) AT_DISPATCH_CASE(_50, N) AT_DISPATCH_CASE(_51, N) AT_DISPATCH_CASE(_52, N) AT_DISPATCH_CASE(_53, N) AT_DISPATCH_CASE(_54, N) AT_DISPATCH_CASE(_55, N) AT_DISPATCH_CASE(_56, N) AT_DISPATCH_CASE(_57, N) AT_DISPATCH_CASE(_58, N)
+#define AT_AP59(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43, _44, _45, _46, _47, _48, _49, _50, _51, _52, _53, _54, _55, _56, _57, _58, _59) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N) AT_DISPATCH_CASE(_41, N) AT_DISPATCH_CASE(_42, N) AT_DISPATCH_CASE(_43, N) AT_DISPATCH_CASE(_44, N) AT_DISPATCH_CASE(_45, N) AT_DISPATCH_CASE(_46, N) AT_DISPATCH_CASE(_47, N) AT_DISPATCH_CASE(_48, N) AT_DISPATCH_CASE(_49, N) AT_DISPATCH_CASE(_50, N) AT_DISPATCH_CASE(_51, N) AT_DISPATCH_CASE(_52, N) AT_DISPATCH_CASE(_53, N) AT_DISPATCH_CASE(_54, N) AT_DISPATCH_CASE(_55, N) AT_DISPATCH_CASE(_56, N) AT_DISPATCH_CASE(_57, N) AT_DISPATCH_CASE(_58, N) AT_DISPATCH_CASE(_59, N)
+#define AT_AP60(N, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29, _30, _31, _32, _33, _34, _35, _36, _37, _38, _39, _40, _41, _42, _43, _44, _45, _46, _47, _48, _49, _50, _51, _52, _53, _54, _55, _56, _57, _58, _59, _60) AT_DISPATCH_CASE(_1, N) AT_DISPATCH_CASE(_2, N) AT_DISPATCH_CASE(_3, N) AT_DISPATCH_CASE(_4, N) AT_DISPATCH_CASE(_5, N) AT_DISPATCH_CASE(_6, N) AT_DISPATCH_CASE(_7, N) AT_DISPATCH_CASE(_8, N) AT_DISPATCH_CASE(_9, N) AT_DISPATCH_CASE(_10, N) AT_DISPATCH_CASE(_11, N) AT_DISPATCH_CASE(_12, N) AT_DISPATCH_CASE(_13, N) AT_DISPATCH_CASE(_14, N) AT_DISPATCH_CASE(_15, N) AT_DISPATCH_CASE(_16, N) AT_DISPATCH_CASE(_17, N) AT_DISPATCH_CASE(_18, N) AT_DISPATCH_CASE(_19, N) AT_DISPATCH_CASE(_20, N) AT_DISPATCH_CASE(_21, N) AT_DISPATCH_CASE(_22, N) AT_DISPATCH_CASE(_23, N) AT_DISPATCH_CASE(_24, N) AT_DISPATCH_CASE(_25, N) AT_DISPATCH_CASE(_26, N) AT_DISPATCH_CASE(_27, N) AT_DISPATCH_CASE(_28, N) AT_DISPATCH_CASE(_29, N) AT_DISPATCH_CASE(_30, N) AT_DISPATCH_CASE(_31, N) AT_DISPATCH_CASE(_32, N) AT_DISPATCH_CASE(_33, N) AT_DISPATCH_CASE(_34, N) AT_DISPATCH_CASE(_35, N) AT_DISPATCH_CASE(_36, N) AT_DISPATCH_CASE(_37, N) AT_DISPATCH_CASE(_38, N) AT_DISPATCH_CASE(_39, N) AT_DISPATCH_CASE(_40, N) AT_DISPATCH_CASE(_41, N) AT_DISPATCH_CASE(_42, N) AT_DISPATCH_CASE(_43, N) AT_DISPATCH_CASE(_44, N) AT_DISPATCH_CASE(_45, N) AT_DISPATCH_CASE(_46, N) AT_DISPATCH_CASE(_47, N) AT_DISPATCH_CASE(_48, N) AT_DISPATCH_CASE(_49, N) AT_DISPATCH_CASE(_50, N) AT_DISPATCH_CASE(_51, N) AT_DISPATCH_CASE(_52, N) AT_DISPATCH_CASE(_53, N) AT_DISPATCH_CASE(_54, N) AT_DISPATCH_CASE(_55, N) AT_DISPATCH_CASE(_56, N) AT_DISPATCH_CASE(_57, N) AT_DISPATCH_CASE(_58, N) AT_DISPATCH_CASE(_59, N) AT_DISPATCH_CASE(_60, N)
+
+// End generated code
+// clang-format on

lib/python3.12/site-packages/torch/include/ATen/DynamicLibrary.h

@@ -0,0 +1,36 @@
+#pragma once
+
+#include <ATen/Utils.h>
+#include <c10/macros/Export.h>
+#include <c10/util/Exception.h>
+
+namespace c10 {
+
+class DynamicLibraryError : public Error {
+  using Error::Error;
+};
+
+} // namespace c10
+
+namespace at {
+
+struct DynamicLibrary {
+  AT_DISALLOW_COPY_AND_ASSIGN(DynamicLibrary);
+  DynamicLibrary(DynamicLibrary&& other) = delete;
+  DynamicLibrary& operator=(DynamicLibrary&&) = delete;
+
+  TORCH_API DynamicLibrary(
+      const char* name,
+      const char* alt_name = nullptr,
+      bool leak_handle = false);
+
+  TORCH_API void* sym(const char* name);
+
+  TORCH_API ~DynamicLibrary();
+
+ private:
+  bool leak_handle;
+  void* handle = nullptr;
+};
+
+} // namespace at

lib/python3.12/site-packages/torch/include/ATen/EmptyTensor.h

@@ -0,0 +1,166 @@
+#pragma once
+#include <ATen/core/TensorBase.h>
+
+namespace at::detail {
+
+inline void check_size_nonnegative(ArrayRef<int64_t> size) {
+  for (const auto& x : size) {
+    TORCH_CHECK(
+        x >= 0,
+        "Trying to create tensor with negative dimension ",
+        x,
+        ": ",
+        size);
+  }
+}
+
+inline void check_size_nonnegative(ArrayRef<c10::SymInt> size) {
+  for (const auto& x : size) {
+    TORCH_CHECK(
+        x.expect_size(__FILE__, __LINE__),
+        "Trying to create tensor with negative dimension ",
+        x,
+        ": ",
+        size);
+  }
+}
+
+TORCH_API size_t computeStorageNbytesContiguous(
+    IntArrayRef sizes,
+    size_t itemsize,
+    size_t storage_offset = 0);
+TORCH_API SymInt computeStorageNbytesContiguous(
+    SymIntArrayRef sizes,
+    const SymInt& itemsize,
+    const SymInt& storage_offset = 0);
+TORCH_API size_t computeStorageNbytes(
+    IntArrayRef sizes,
+    IntArrayRef strides,
+    size_t itemsize,
+    size_t storage_offset = 0);
+TORCH_API SymInt computeStorageNbytes(
+    SymIntArrayRef sizes,
+    SymIntArrayRef strides,
+    const SymInt& itemsize,
+    const SymInt& storage_offset = 0);
+
+TORCH_API TensorBase empty_generic(
+    IntArrayRef size,
+    c10::Allocator* allocator,
+    c10::DispatchKeySet ks,
+    ScalarType scalar_type,
+    std::optional<c10::MemoryFormat> memory_format_opt);
+
+TORCH_API TensorBase empty_generic_symint(
+    SymIntArrayRef size,
+    c10::Allocator* allocator,
+    c10::DispatchKeySet ks,
+    ScalarType scalar_type,
+    std::optional<c10::MemoryFormat> memory_format_opt);
+
+TORCH_API TensorBase empty_strided_generic(
+    IntArrayRef size,
+    IntArrayRef stride,
+    c10::Allocator* allocator,
+    c10::DispatchKeySet ks,
+    ScalarType scalar_type);
+
+TORCH_API TensorBase empty_strided_symint_generic(
+    SymIntArrayRef size,
+    SymIntArrayRef stride,
+    c10::Allocator* allocator,
+    c10::DispatchKeySet ks,
+    ScalarType scalar_type);
+
+TORCH_API TensorBase empty_cpu(
+    IntArrayRef size,
+    ScalarType dtype,
+    bool pin_memory = false,
+    std::optional<c10::MemoryFormat> memory_format_opt = std::nullopt);
+
+TORCH_API TensorBase empty_cpu(
+    IntArrayRef size,
+    std::optional<ScalarType> dtype_opt,
+    std::optional<Layout> layout_opt,
+    std::optional<Device> device_opt,
+    std::optional<bool> pin_memory_opt,
+    std::optional<c10::MemoryFormat> memory_format_opt);
+
+TORCH_API TensorBase empty_cpu(IntArrayRef size, const TensorOptions& options);
+
+TORCH_API TensorBase empty_strided_cpu(
+    IntArrayRef size,
+    IntArrayRef stride,
+    ScalarType dtype,
+    bool pin_memory = false);
+
+TORCH_API TensorBase empty_strided_cpu(
+    IntArrayRef size,
+    IntArrayRef stride,
+    std::optional<ScalarType> dtype_opt,
+    std::optional<Layout> layout_opt,
+    std::optional<Device> device_opt,
+    std::optional<bool> pin_memory_opt);
+
+TORCH_API TensorBase empty_strided_cpu(
+    IntArrayRef size,
+    IntArrayRef stride,
+    const TensorOptions& options);
+
+TORCH_API TensorBase empty_meta(
+    IntArrayRef size,
+    ScalarType dtype,
+    std::optional<c10::MemoryFormat> memory_format_opt = std::nullopt);
+
+TORCH_API TensorBase empty_meta(
+    IntArrayRef size,
+    std::optional<ScalarType> dtype_opt,
+    std::optional<Layout> layout_opt,
+    std::optional<Device> device_opt,
+    std::optional<bool> pin_memory_opt,
+    std::optional<c10::MemoryFormat> memory_format_opt);
+
+TORCH_API TensorBase empty_symint_meta(
+    SymIntArrayRef size,
+    std::optional<ScalarType> dtype_opt,
+    std::optional<Layout> layout_opt,
+    std::optional<Device> device_opt,
+    std::optional<bool> pin_memory_opt,
+    std::optional<c10::MemoryFormat> memory_format_opt);
+
+TORCH_API TensorBase empty_meta(IntArrayRef size, const TensorOptions& options);
+
+TORCH_API TensorBase
+empty_strided_meta(IntArrayRef size, IntArrayRef stride, ScalarType dtype);
+
+TORCH_API TensorBase empty_strided_meta(
+    IntArrayRef size,
+    IntArrayRef stride,
+    std::optional<ScalarType> dtype_opt,
+    std::optional<Layout> layout_opt,
+    std::optional<Device> device_opt,
+    std::optional<bool> pin_memory_opt);
+
+TORCH_API TensorBase empty_strided_meta(
+    IntArrayRef size,
+    IntArrayRef stride,
+    const TensorOptions& options);
+
+TORCH_API TensorBase empty_strided_symint_meta(
+    SymIntArrayRef size,
+    SymIntArrayRef stride,
+    ScalarType dtype);
+
+TORCH_API TensorBase empty_strided_symint_meta(
+    SymIntArrayRef size,
+    SymIntArrayRef stride,
+    std::optional<ScalarType> dtype_opt,
+    std::optional<Layout> layout_opt,
+    std::optional<Device> device_opt);
+
+TORCH_API TensorBase empty_strided_symint_meta(
+    SymIntArrayRef size,
+    SymIntArrayRef stride,
+    const TensorOptions& options);
+
+} // namespace at::detail

lib/python3.12/site-packages/torch/include/ATen/ExpandUtils.h

@@ -0,0 +1,527 @@
+#pragma once
+
+#ifndef AT_PER_OPERATOR_HEADERS
+#include <ATen/Functions.h>
+#else
+#include <ATen/ops/view.h>
+#include <ATen/ops/view_copy.h>
+#endif
+
+#include <ATen/Tensor.h>
+#include <ATen/core/DimVector.h>
+#include <c10/util/Exception.h>
+#include <c10/util/MaybeOwned.h>
+#include <c10/util/irange.h>
+
+#include <functional>
+#include <tuple>
+#include <utility>
+
+namespace at {
+
+TORCH_API std::vector<int64_t> infer_size(IntArrayRef a, IntArrayRef b);
+TORCH_API std::vector<SymInt> infer_size_symint(
+    SymIntArrayRef a,
+    SymIntArrayRef b);
+TORCH_API DimVector infer_size_dimvector(IntArrayRef a, IntArrayRef b);
+TORCH_API SymDimVector
+infer_size_symdimvector(SymIntArrayRef a, SymIntArrayRef b);
+
+// Named type instead of a pair/tuple so that we can be sure to
+// construct the vectors in place and get NRVO.
+template <typename Container>
+struct InferExpandGeometryResult {
+  Container sizes;
+  Container strides;
+  explicit InferExpandGeometryResult(size_t ndim)
+      : sizes(ndim), strides(ndim) {}
+  explicit InferExpandGeometryResult(IntArrayRef sizes_, size_t ndim)
+      : sizes(sizes_.begin(), sizes_.end()), strides(ndim) {}
+};
+
+TORCH_API std::tuple<std::vector<int64_t>, std::vector<int64_t>>
+inferExpandGeometry(
+    IntArrayRef tensor_sizes,
+    IntArrayRef tensor_strides,
+    IntArrayRef sizes);
+
+TORCH_API InferExpandGeometryResult<DimVector> inferExpandGeometry_dimvector(
+    IntArrayRef tensor_sizes,
+    IntArrayRef tensor_strides,
+    IntArrayRef sizes);
+
+TORCH_API std::vector<int64_t> infer_dense_strides(
+    IntArrayRef tensor_sizes,
+    IntArrayRef tensor_strides);
+
+// True if input shapes are expandable
+// NOTE: infer_size did a similar check, please keep them sync if change is
+// needed
+inline bool are_expandable(IntArrayRef shape1, IntArrayRef shape2) {
+  size_t ndim1 = shape1.size();
+  size_t ndim2 = shape2.size();
+  size_t ndim = ndim1 < ndim2 ? ndim1 : ndim2;
+
+  for (int64_t i = static_cast<int64_t>(ndim) - 1; i >= 0; --i) {
+    if (shape1[--ndim1] == shape2[--ndim2] || shape1[ndim1] == 1 ||
+        shape2[ndim2] == 1) {
+      continue;
+    }
+    return false;
+  }
+  return true;
+}
+
+// avoid copy-construction of Tensor by using a reference_wrapper.
+inline void check_defined(
+    std::initializer_list<std::reference_wrapper<const Tensor>> tensors,
+    const char* api_name) {
+  for (auto& t : tensors) {
+    if (!t.get().defined()) {
+      TORCH_CHECK(false, api_name, "(...) called with an undefined Tensor");
+    }
+  }
+}
+
+// NOTE [ ExpandUtils Borrowing ]
+//
+// Functions in ExpandUtils return `c10::MaybeOwned<Tensor>` because
+// expansion may not actually be needed, in which case we can improve
+// efficiency by returning
+// `c10::MaybeOwned<Tensor>::borrowed(to_expand)`. However, this means
+// that you need to be careful: the returned `c10::MaybeOwned<Tensor>`
+// must not outlive the original `Tensor` object that `to_expand`
+// referred to! The deleted rvalue reference overloads of these
+// functions help with this by preventing trivial use of a temporary
+// resulting from a function call, but it is still possible to make a
+// mistake.
+
+inline c10::MaybeOwned<Tensor> expand_inplace(
+    const Tensor& tensor,
+    const Tensor& to_expand) {
+  if (tensor.sym_sizes().equals(to_expand.sym_sizes())) {
+    return c10::MaybeOwned<Tensor>::borrowed(to_expand);
+  }
+  return c10::MaybeOwned<Tensor>::owned(
+      to_expand.expand_symint(tensor.sym_sizes()));
+}
+
+inline c10::MaybeOwned<Tensor> expand_inplace(
+    const Tensor& tensor,
+    Tensor&& to_expand) = delete;
+
+inline c10::MaybeOwned<Tensor> expand_inplace(
+    const Tensor& tensor,
+    const Tensor& to_expand,
+    const char* api_name) {
+  check_defined({tensor, to_expand}, api_name);
+  return expand_inplace(tensor, to_expand);
+}
+
+inline c10::MaybeOwned<Tensor> expand_inplace(
+    const Tensor& tensor,
+    Tensor&& to_expand,
+    const char* api_name) = delete;
+
+inline std::tuple<c10::MaybeOwned<Tensor>, c10::MaybeOwned<Tensor>>
+expand_inplace(
+    const Tensor& tensor,
+    const Tensor& to_expand1,
+    const Tensor& to_expand2) {
+  if (tensor.sizes().equals(to_expand1.sizes()) &&
+      tensor.sizes().equals((to_expand2.sizes()))) {
+    return std::make_tuple(
+        c10::MaybeOwned<Tensor>::borrowed(to_expand1),
+        c10::MaybeOwned<Tensor>::borrowed(to_expand2));
+  }
+
+  return std::make_tuple(
+      c10::MaybeOwned<Tensor>::owned(to_expand1.expand(tensor.sizes())),
+      c10::MaybeOwned<Tensor>::owned(to_expand2.expand(tensor.sizes())));
+}
+
+inline std::tuple<c10::MaybeOwned<Tensor>, c10::MaybeOwned<Tensor>>
+expand_inplace(
+    const Tensor& tensor,
+    Tensor&& to_expand1,
+    const Tensor& to_expand2) = delete;
+inline std::tuple<c10::MaybeOwned<Tensor>, c10::MaybeOwned<Tensor>>
+expand_inplace(
+    const Tensor& tensor,
+    const Tensor& to_expand1,
+    Tensor&& to_expand2) = delete;
+inline std::tuple<c10::MaybeOwned<Tensor>, c10::MaybeOwned<Tensor>>
+expand_inplace(const Tensor& tensor, Tensor&& to_expand1, Tensor&& to_expand2) =
+    delete;
+
+inline std::tuple<c10::MaybeOwned<Tensor>, c10::MaybeOwned<Tensor>>
+expand_inplace(
+    const Tensor& tensor,
+    const Tensor& to_expand1,
+    const Tensor& to_expand2,
+    const char* api_name) {
+  check_defined({tensor, to_expand1, to_expand2}, api_name);
+  return expand_inplace(tensor, to_expand1, to_expand2);
+}
+
+inline std::tuple<c10::MaybeOwned<Tensor>, c10::MaybeOwned<Tensor>>
+expand_inplace(
+    const Tensor& tensor,
+    Tensor&& to_expand1,
+    const Tensor& to_expand2,
+    const char* api_name) = delete;
+inline std::tuple<c10::MaybeOwned<Tensor>, c10::MaybeOwned<Tensor>>
+expand_inplace(
+    const Tensor& tensor,
+    const Tensor& to_expand1,
+    Tensor&& to_expand2,
+    const char* api_name) = delete;
+inline std::tuple<c10::MaybeOwned<Tensor>, c10::MaybeOwned<Tensor>>
+expand_inplace(
+    const Tensor& tensor,
+    Tensor&& to_expand1,
+    Tensor&& to_expand2,
+    const char* api_name) = delete;
+
+// See NOTE [ ExpandUtils Borrowing ] above for `MaybeOwned` explanation.
+inline std::tuple<c10::MaybeOwned<Tensor>, c10::MaybeOwned<Tensor>>
+expand_outplace(const Tensor& to_expand1, const Tensor& to_expand2) {
+  auto s1 = to_expand1.sym_sizes();
+  auto s2 = to_expand2.sym_sizes();
+  if (s1.equals(s2)) {
+    return std::make_tuple(
+        c10::MaybeOwned<Tensor>::borrowed(to_expand1),
+        c10::MaybeOwned<Tensor>::borrowed(to_expand2));
+  }
+
+  auto expanded_size = infer_size_symdimvector(s1, s2);
+  return std::make_tuple(
+      c10::MaybeOwned<Tensor>::owned(to_expand1.expand_symint(expanded_size)),
+      c10::MaybeOwned<Tensor>::owned(to_expand2.expand_symint(expanded_size)));
+}
+
+inline std::tuple<c10::MaybeOwned<Tensor>, c10::MaybeOwned<Tensor>>
+expand_outplace(Tensor&& to_expand1, const Tensor& to_expand2) = delete;
+inline std::tuple<c10::MaybeOwned<Tensor>, c10::MaybeOwned<Tensor>>
+expand_outplace(const Tensor& to_expand1, Tensor&& to_expand2) = delete;
+inline std::tuple<c10::MaybeOwned<Tensor>, c10::MaybeOwned<Tensor>>
+expand_outplace(Tensor&& to_expand1, Tensor&& to_expand2) = delete;
+
+inline std::tuple<c10::MaybeOwned<Tensor>, c10::MaybeOwned<Tensor>>
+expand_outplace(
+    const Tensor& to_expand1,
+    const Tensor& to_expand2,
+    const char* api_name) {
+  check_defined({to_expand1, to_expand2}, api_name);
+  return expand_outplace(to_expand1, to_expand2);
+}
+
+inline std::tuple<c10::MaybeOwned<Tensor>, c10::MaybeOwned<Tensor>>
+expand_outplace(
+    Tensor&& to_expand1,
+    const Tensor& to_expand2,
+    const char* api_name) = delete;
+inline std::tuple<c10::MaybeOwned<Tensor>, c10::MaybeOwned<Tensor>>
+expand_outplace(
+    const Tensor& to_expand1,
+    Tensor&& to_expand2,
+    const char* api_name) = delete;
+inline std::tuple<c10::MaybeOwned<Tensor>, c10::MaybeOwned<Tensor>>
+expand_outplace(
+    Tensor&& to_expand1,
+    Tensor&& to_expand2,
+    const char* api_name) = delete;
+
+inline std::tuple<
+    c10::MaybeOwned<Tensor>,
+    c10::MaybeOwned<Tensor>,
+    c10::MaybeOwned<Tensor>>
+expand_outplace(
+    const Tensor& to_expand1,
+    const Tensor& to_expand2,
+    const Tensor& to_expand3) {
+  if (to_expand1.sizes().equals(to_expand2.sizes()) &&
+      to_expand1.sizes().equals(to_expand3.sizes())) {
+    return std::make_tuple(
+        c10::MaybeOwned<Tensor>::borrowed(to_expand1),
+        c10::MaybeOwned<Tensor>::borrowed(to_expand2),
+        c10::MaybeOwned<Tensor>::borrowed(to_expand3));
+  }
+
+  auto expanded_size12 =
+      infer_size_dimvector(to_expand1.sizes(), to_expand2.sizes());
+  auto expanded_size =
+      infer_size_dimvector(expanded_size12, to_expand3.sizes());
+  return std::make_tuple(
+      c10::MaybeOwned<Tensor>::owned(to_expand1.expand(expanded_size)),
+      c10::MaybeOwned<Tensor>::owned(to_expand2.expand(expanded_size)),
+      c10::MaybeOwned<Tensor>::owned(to_expand3.expand(expanded_size)));
+}
+
+inline std::tuple<
+    c10::MaybeOwned<Tensor>,
+    c10::MaybeOwned<Tensor>,
+    c10::MaybeOwned<Tensor>>
+expand_outplace(
+    Tensor&& to_expand1,
+    const Tensor& to_expand2,
+    const Tensor& to_expand3) = delete;
+inline std::tuple<
+    c10::MaybeOwned<Tensor>,
+    c10::MaybeOwned<Tensor>,
+    c10::MaybeOwned<Tensor>>
+expand_outplace(
+    const Tensor& to_expand1,
+    Tensor&& to_expand2,
+    const Tensor& to_expand3) = delete;
+inline std::tuple<
+    c10::MaybeOwned<Tensor>,
+    c10::MaybeOwned<Tensor>,
+    c10::MaybeOwned<Tensor>>
+expand_outplace(
+    Tensor&& to_expand1,
+    Tensor&& to_expand2,
+    const Tensor& to_expand3) = delete;
+inline std::tuple<
+    c10::MaybeOwned<Tensor>,
+    c10::MaybeOwned<Tensor>,
+    c10::MaybeOwned<Tensor>>
+expand_outplace(
+    const Tensor& to_expand1,
+    const Tensor& to_expand2,
+    Tensor&& to_expand3) = delete;
+inline std::tuple<
+    c10::MaybeOwned<Tensor>,
+    c10::MaybeOwned<Tensor>,
+    c10::MaybeOwned<Tensor>>
+expand_outplace(
+    Tensor&& to_expand1,
+    const Tensor& to_expand2,
+    Tensor&& to_expand3) = delete;
+inline std::tuple<
+    c10::MaybeOwned<Tensor>,
+    c10::MaybeOwned<Tensor>,
+    c10::MaybeOwned<Tensor>>
+expand_outplace(
+    const Tensor& to_expand1,
+    Tensor&& to_expand2,
+    Tensor&& to_expand3) = delete;
+inline std::tuple<
+    c10::MaybeOwned<Tensor>,
+    c10::MaybeOwned<Tensor>,
+    c10::MaybeOwned<Tensor>>
+expand_outplace(Tensor&& to_expand1, Tensor&& to_expand2, Tensor&& to_expand3) =
+    delete;
+
+inline std::tuple<
+    c10::MaybeOwned<Tensor>,
+    c10::MaybeOwned<Tensor>,
+    c10::MaybeOwned<Tensor>>
+expand_outplace(
+    const Tensor& to_expand1,
+    const Tensor& to_expand2,
+    const Tensor& to_expand3,
+    const char* api_name) {
+  check_defined({to_expand1, to_expand2, to_expand3}, api_name);
+  return expand_outplace(to_expand1, to_expand2, to_expand3);
+}
+
+inline std::tuple<
+    c10::MaybeOwned<Tensor>,
+    c10::MaybeOwned<Tensor>,
+    c10::MaybeOwned<Tensor>>
+expand_outplace(
+    Tensor&& to_expand1,
+    const Tensor& to_expand2,
+    const Tensor& to_expand3,
+    const char* api_name) = delete;
+inline std::tuple<
+    c10::MaybeOwned<Tensor>,
+    c10::MaybeOwned<Tensor>,
+    c10::MaybeOwned<Tensor>>
+expand_outplace(
+    const Tensor& to_expand1,
+    Tensor&& to_expand2,
+    const Tensor& to_expand3,
+    const char* api_name) = delete;
+inline std::tuple<
+    c10::MaybeOwned<Tensor>,
+    c10::MaybeOwned<Tensor>,
+    c10::MaybeOwned<Tensor>>
+expand_outplace(
+    Tensor&& to_expand1,
+    Tensor&& to_expand2,
+    const Tensor& to_expand3,
+    const char* api_name) = delete;
+inline std::tuple<
+    c10::MaybeOwned<Tensor>,
+    c10::MaybeOwned<Tensor>,
+    c10::MaybeOwned<Tensor>>
+expand_outplace(
+    const Tensor& to_expand1,
+    const Tensor& to_expand2,
+    Tensor&& to_expand3,
+    const char* api_name) = delete;
+inline std::tuple<
+    c10::MaybeOwned<Tensor>,
+    c10::MaybeOwned<Tensor>,
+    c10::MaybeOwned<Tensor>>
+expand_outplace(
+    Tensor&& to_expand1,
+    const Tensor& to_expand2,
+    Tensor&& to_expand3,
+    const char* api_name) = delete;
+inline std::tuple<
+    c10::MaybeOwned<Tensor>,
+    c10::MaybeOwned<Tensor>,
+    c10::MaybeOwned<Tensor>>
+expand_outplace(
+    const Tensor& to_expand1,
+    Tensor&& to_expand2,
+    Tensor&& to_expand3,
+    const char* api_name) = delete;
+inline std::tuple<
+    c10::MaybeOwned<Tensor>,
+    c10::MaybeOwned<Tensor>,
+    c10::MaybeOwned<Tensor>>
+expand_outplace(
+    Tensor&& to_expand1,
+    Tensor&& to_expand2,
+    Tensor&& to_expand3,
+    const char* api_name) = delete;
+
+inline c10::MaybeOwned<Tensor> expand_size(
+    const Tensor& to_expand,
+    IntArrayRef sizes) {
+  if (to_expand.sizes().equals(sizes)) {
+    return c10::MaybeOwned<Tensor>::borrowed(to_expand);
+  }
+
+  return c10::MaybeOwned<Tensor>::owned(to_expand.expand(sizes));
+}
+
+inline c10::MaybeOwned<Tensor> expand_size(
+    Tensor&& to_expand,
+    IntArrayRef sizes) = delete;
+
+inline c10::MaybeOwned<Tensor> expand_size(
+    const Tensor& to_expand,
+    IntArrayRef sizes,
+    const char* api_name) {
+  check_defined({to_expand}, api_name);
+  return expand_size(to_expand, sizes);
+}
+
+inline c10::MaybeOwned<Tensor> expand_size(
+    Tensor&& to_expand,
+    IntArrayRef sizes,
+    const char* api_name) = delete;
+
+inline std::vector<Tensor> expand_outplace(TensorList to_expand) {
+  // expands a list of Tensors; ignores undefined (null) tensors
+  bool first = true;
+  SymDimVector sizes;
+  for (const auto i : c10::irange(to_expand.size())) {
+    if (!to_expand[i].defined()) {
+      continue;
+    } else if (first) {
+      sizes = to_expand[i].sym_sizes();
+      first = false;
+    } else {
+      sizes = infer_size_symdimvector(sizes, to_expand[i].sym_sizes());
+    }
+  }
+
+  std::vector<Tensor> result(to_expand.size());
+  for (const auto i : c10::irange(to_expand.size())) {
+    if (!to_expand[i].defined()) {
+      continue;
+    } else if (to_expand[i].sym_sizes().equals(sizes)) {
+      result[i] = to_expand[i];
+    } else {
+      result[i] = to_expand[i].expand_symint(sizes);
+    }
+  }
+  return result;
+}
+
+template <typename T>
+inline Tensor _sum_to(
+    Tensor tensor,
+    const c10::ArrayRef<T> shape,
+    bool always_return_non_view = false) {
+  if (shape.size() == 0) {
+    return tensor.sum();
+  }
+
+  auto sizes = at::symint::sizes<T>(tensor);
+  c10::SmallVector<int64_t, 8> reduce_dims;
+  const int64_t leading_dims = sizes.size() - shape.size();
+  for (const auto i : c10::irange(leading_dims)) {
+    reduce_dims.push_back(i);
+  }
+  for (int64_t i = leading_dims; i < static_cast<int64_t>(sizes.size()); ++i) {
+    if (TORCH_GUARD_SIZE_OBLIVIOUS(sym_eq(shape[i - leading_dims], 1)) &&
+        TORCH_GUARD_SIZE_OBLIVIOUS(sym_ne(sizes[i], 1))) {
+      reduce_dims.push_back(i);
+    }
+  }
+
+  if (!reduce_dims.empty()) {
+    tensor = tensor.sum(reduce_dims, /*keepdim=*/true);
+  }
+
+  if (always_return_non_view) {
+    // This is only actually used by the functionalization pass.
+    // We want to be able to guarantee that this function doesn't return a view
+    // of the input.
+    return leading_dims > 0 ? at::symint::view_copy<T>(tensor, shape)
+                            : tensor.clone();
+  } else {
+    return leading_dims > 0 ? at::symint::view<T>(tensor, shape) : tensor;
+  }
+}
+
+inline Tensor sum_to(
+    Tensor tensor,
+    const c10::SymIntArrayRef shape,
+    bool always_return_non_view = false) {
+  return _sum_to(std::move(tensor), shape, always_return_non_view);
+}
+
+// Sums `tensor` repeatedly to produce a tensor of shape `shape`.
+// Precondition: is_expandable_to(shape, tensor.sizes()) must be true
+inline Tensor sum_to(
+    Tensor tensor,
+    const IntArrayRef shape,
+    bool always_return_non_view = false) {
+  return _sum_to(std::move(tensor), shape, always_return_non_view);
+}
+
+inline bool is_expandable_to(
+    SymIntArrayRef shape,
+    c10::SymIntArrayRef desired) {
+  size_t ndim = shape.size();
+  size_t target_dim = desired.size();
+  if (ndim > target_dim) {
+    return false;
+  }
+  for (const auto i : c10::irange(ndim)) {
+    const auto& size = shape[ndim - i - 1];
+    const auto& target = desired[target_dim - i - 1];
+    if (size != target && size != 1) {
+      return false;
+    }
+  }
+  return true;
+}
+
+inline bool is_expandable_to(IntArrayRef shape, IntArrayRef desired) {
+  auto sym_shape = c10::SymIntArrayRef(
+      reinterpret_cast<const c10::SymInt*>(shape.data()), shape.size());
+  auto sym_desired = c10::SymIntArrayRef(
+      reinterpret_cast<const c10::SymInt*>(desired.data()), desired.size());
+  return is_expandable_to(sym_shape, sym_desired);
+}
+
+} // namespace at

lib/python3.12/site-packages/torch/include/ATen/Formatting.h

@@ -0,0 +1 @@
+#include <ATen/core/Formatting.h>

lib/python3.12/site-packages/torch/include/ATen/FuncTorchTLS.h

@@ -0,0 +1,46 @@
+#pragma once
+
+#include <c10/macros/Macros.h>
+#include <memory>
+
+namespace at::functorch {
+
+// NOTE [functorch TLS in pytorch/pytorch]
+//
+// functorch lives out-of-tree. However, it has some TLS that needs to be
+// propagated. The solution for that is we store a pointer to the TLS
+// inside pytorch/pytorch and extend FuncTorchTLSBase inside functorch to
+// include whatever functorch needs.
+//
+// We need to store a pointer due to the indirection:
+// inside functorch, we will create a subclass of FunctorchTLSBase called
+// FuncTorchTLSImpl that actually contains metadata, like the DynamicLayerStack.
+// FuncTorchTLSBase doesn't have any metadata because it hasn't been defined
+// yet.
+//
+// Here in pytorch/pytorch, we will pass around FuncTorchTLSBase*, but inside
+// functorch, we will assign a FuncTorchTLSImpl* to the FunctorchTLSBase*.
+// We can't directly pass around FunctorchTLSBase (without a pointer) because
+// FuncTorchTLSImpl does not fit inside a FuncTorchTLSBase by virtue of having
+// more elements.
+struct TORCH_API FuncTorchTLSBase {
+  virtual ~FuncTorchTLSBase() = default;
+  virtual std::unique_ptr<FuncTorchTLSBase> deepcopy() const = 0;
+
+  virtual int64_t checkSupportsSingleLevelAutogradFunction() const = 0;
+  virtual void checkSupportsCppAutogradFunction() const = 0;
+  virtual void checkSupportsInplaceRequiresGrad() const = 0;
+  virtual void checkSupportsRetainGrad() const = 0;
+};
+
+// returns deepcopy of the functorch tls
+TORCH_API std::unique_ptr<FuncTorchTLSBase> getCopyOfFuncTorchTLS();
+
+// sets the functorch tls. always does a deep copy.
+TORCH_API void setFuncTorchTLS(
+    const std::shared_ptr<const FuncTorchTLSBase>& state);
+
+// get a mutable reference to the functorch tls
+TORCH_API std::unique_ptr<FuncTorchTLSBase>& functorchTLSAccessor();
+
+} // namespace at::functorch

lib/python3.12/site-packages/torch/include/ATen/FunctionalStorageImpl.h

@@ -0,0 +1,208 @@
+#pragma once
+
+#include <ATen/Tensor.h>
+
+#include <utility>
+
+namespace at::functionalization {
+
+// See Note [Functionalization Pass In Core]
+
+// ViewMeta is a class used by the functionalization pass to navigate between
+// a base tensor and a view tensor.
+// For example, if I call `b = a.view1(...)`
+// the functionalization pass will generate and store a ViewMeta on b that looks
+// like:
+//
+// ViewMeta(
+//   [<captures>](const Tensor& base, int64_t mutated_view_idx) {
+//     return base.view1(...);
+//   },
+//   [<captures>](const at::Tensor& base, const at::Tensor& mutated_view,
+//   int64_t mutated_view_idx) -> at::Tensor {
+//     return at::functionalization::impl::view1_inverse(base, mutated_view,
+//     ...);
+//   }
+//
+// The forward_fn lambda describes how to replay view1 on a tensor.
+//
+// The reverse_fn lambda describes how, given a tensor that is already a view,
+// how to get the corresponding base tensor. See Note [Functionalization Pass:
+// View Inverses] for details.
+struct ViewMeta {
+  ViewMeta(
+      std::function<Tensor(const Tensor&, int64_t)> forward,
+      std::function<Tensor(const Tensor&, const Tensor&, int64_t)> reverse,
+      bool has_symbolic_inputs,
+      bool is_multi_output = false,
+      bool is_as_strided = false,
+      int64_t out_idx = 0)
+      : forward_fn(std::move(forward)),
+        reverse_fn(std::move(reverse)),
+        out_index(out_idx),
+        is_multi_output(is_multi_output),
+        is_as_strided(is_as_strided),
+        has_symbolic_inputs(has_symbolic_inputs) {}
+
+  std::function<Tensor(const Tensor&, int64_t)> forward_fn;
+  std::function<Tensor(const Tensor&, const Tensor&, int64_t)> reverse_fn;
+  // See Note [out_idx in ViewMeta]
+  int64_t out_index;
+
+  // Tells us if this is a multi-output view
+  bool is_multi_output;
+
+  bool is_as_strided;
+
+  // Tells us if this view operation has any symbolic inputs
+  bool has_symbolic_inputs;
+
+  // Returns a copy of the current ViewMeta, if out_idx matches the current
+  // out_index. Otherwise, returns a new ViewMeta with the same forward/reverse
+  // functions, but a new out index.
+  ViewMeta to_out_idx(int64_t out_idx);
+};
+
+// FunctionalStorageImpl is a subclass of StorageImpl used by the
+// functionalization pass. It has no underlying data (similar to meta storage).
+// It also knows how to reflect mutations to tensors in the absence of a valid
+// data pointer.
+//
+// A storage represents the state shared by (potentially multiple) views of the
+// same tensor. For example, in the following code:
+//
+// b = a.view1(...)
+// c = b.view2(...)
+// b.add_(1)
+// --> storage.add_update(b, {view1_meta})
+//
+// The call to add_(1) will result in a call to alias.add_update(b,
+// {view1_meta}), queueing up the mutation from b onto the alias. Later, suppose
+// c is used in an expression (e.g. you try to print c, or pass it to an
+// operator). Doing so will involve "syncing" c. First we apply any pending
+// updates to the alias, and then we regenerate c by replaying its views off of
+// the updated alias. E.g:
+//
+// print(str(c))
+// --> c.sync_()
+//     --> alias.apply_updates() // after this, the alias will be updated to
+//     reflect the mutation to b
+struct TORCH_API FunctionalStorageImpl : public c10::StorageImpl {
+ public:
+  struct Update {
+    // NOLINTNEXTLINE(cppcoreguidelines-avoid-const-or-ref-data-members)
+    const at::Tensor new_val;
+    // NOLINTNEXTLINE(cppcoreguidelines-avoid-const-or-ref-data-members)
+    const std::vector<ViewMeta> view_metas;
+  };
+
+  explicit FunctionalStorageImpl(const Tensor& value);
+
+  void add_update(
+      const Tensor& updated_val,
+      const std::vector<ViewMeta>& view_metas);
+  bool apply_updates();
+  const Tensor& base() {
+    return base_;
+  }
+  size_t generation() const {
+    return generation_;
+  }
+  void freeze() {
+    frozen_ = true;
+  }
+
+  c10::SymInt get_storage_size(bool before) {
+    if (before) {
+      return original_storage_size_;
+    } else {
+      return curr_storage_size_;
+    }
+  }
+
+  ~FunctionalStorageImpl() override = default;
+
+  void mark_mutation() {
+    mutation_counter_++;
+  }
+  void mark_mutation_during_no_grad_or_inference_mode() {
+    mutation_counter_during_no_grad_or_inference_mode_++;
+  }
+  void mark_mutation_hidden_from_autograd() {
+    mutation_counter_hidden_from_autograd_++;
+  }
+
+  bool are_all_mutations_under_no_grad_or_inference_mode() const {
+    auto non_autograd_mutations =
+        mutation_counter_during_no_grad_or_inference_mode_ +
+        mutation_counter_hidden_from_autograd_;
+    // The <= is because both counters will technically be incremented, if we
+    // perform e.g. a triton kernel mutation under no_grad
+    return mutation_counter_ <= non_autograd_mutations;
+  }
+
+  bool are_all_mutations_hidden_from_autograd() const {
+    // mutations under no_grad / inference_mode are technically not hidden from
+    // autograd - they change the version counter
+    return mutation_counter_ <= mutation_counter_hidden_from_autograd_;
+  }
+
+  void mark_inductor_storage_resize(c10::SymInt new_size) {
+    inductor_storage_resized_ = true;
+    curr_storage_size_ = std::move(new_size);
+  }
+
+  bool was_inductor_storage_resized() {
+    return inductor_storage_resized_;
+  }
+
+ private:
+  // NB: base_ should always point to a tensor BELOW the current
+  // functionalization layer. This is mainly to avoid reference cycles. e.g.
+  // given `b = a.view(...)` Both a.storage_ and b.storage_ are a
+  // FunctionStorageImpl containing an Walualias, with contains a Tensor
+  // `base_`. In this case (where a and b are FunctionalTensorWrapper's), base_
+  // should point not to a, but to a's unwrapped value, a.value_` See Note
+  // [Functionalization: Walualias Removal] for a diagram that shows this
+  // visually.
+  at::Tensor base_;
+  std::vector<Update> updates_;
+  // generation_ gets incremented every time a mutation is queued onto the
+  // alias. It is used to determine if a given tensor is "up to date", or if it
+  // needs to be regenerated from the alias.
+  size_t generation_ = 0;
+  // If frozen, no more mutations are allowed on this storage.  Once frozen, a
+  // storage cannot be unfrozen.
+  bool frozen_ = false;
+
+  // These mutation counters are bumped on the storage
+  // whenever a FunctionalTensorWrapper experiences a mutation.
+  // When the mutation is under no_grad, or comes from a triton kernel, we also
+  // bump the corresponding during_no_grad or hidden_from_autograd counters. Why
+  // do we need to detect these two situations separately from "normal" input
+  // mutations? (1) "normal" input mutations can mutate autograd metadata like
+  // .grad_fn,
+  //     in which case they need to be replayed outside of the compiled graph
+  // (2) "no_grad" input mutations are generally safe to keep in the graph (and
+  // compile),
+  //     but they bump the tensor's VC, so we need to mark_dirty() on the inputs
+  //     in torch.compile
+  // (3) mutations that are fully hidden from autograd (e.g. from a triton
+  // kernel)
+  //     do not mutate any autograd state, and be fully kept in the graph
+  // When we detect that an input was mutated, we need to be able to tell if:
+  // (1) all of the mutations were from triton kernels
+  // (2) all of the mutations were under no_grad
+  uint64_t mutation_counter_during_no_grad_or_inference_mode_ = 0;
+  uint64_t mutation_counter_ = 0;
+  uint64_t mutation_counter_hidden_from_autograd_ = 0;
+
+  // Used to tell if:
+  // (1) There were any storage resizes on a graph input
+  // (2) The original/curr storage size tell us if these resizes result in a nop
+  bool inductor_storage_resized_ = false;
+  c10::SymInt original_storage_size_;
+  c10::SymInt curr_storage_size_;
+};
+
+} // namespace at::functionalization

lib/python3.12/site-packages/torch/include/ATen/FunctionalTensorWrapper.h

@@ -0,0 +1,454 @@
+
+#pragma once
+
+#include <ATen/ArrayRef.h>
+#include <ATen/FunctionalStorageImpl.h>
+#include <ATen/core/IListRef.h>
+#include <ATen/core/List.h>
+#include <ATen/core/boxing/BoxedKernel.h>
+#include <ATen/core/boxing/impl/boxing.h>
+#include <ATen/core/dispatch/Dispatcher.h>
+
+#include <c10/core/DispatchKey.h>
+
+namespace at {
+
+// Note [Functionalization Pass In Core]
+// The Functionalization pass is used to remove aliasing from a pytorch program.
+//
+// This is useful for backends that don't support aliasing, like XLA and Vulkan.
+// It's also necessary in order to remove mutation from a program, which is
+// needed in Functorch.
+//
+// Consider this program:
+// a = torch.ones(...)
+// b = a.view(...)
+// b.add_(1)
+//
+// In this program, b is meant to alias with a due to the use of view(). At the
+// end of the program, both a and b are full of 2's. However, backends that
+// don't support aliasing aren't able to correctly implement the view()
+// operator. Instead, they can opt into the Functionalization pass, which will
+// sit between the user and the backend, and provide the necessary aliasing
+// logic.
+//
+// The functionalization pass will turn the above program into a slightly
+// different program that has the same semantics, transparently to the user,
+// that backends like XLA/Vulkan are able to implement a = torch.ones(...) b =
+// a.view_copy(...)  # view() replaced with view_copy(). Backends like
+// XLA/Vulkan can implement this! b.add_(1) a.add_(1)  # Our functionalization
+// pass machinery knows that a and b are aliased - it applies b's mutation to a
+// too.
+//
+// So, how does the functionalization pass keep track of which tensors are
+// aliased? The pass works by wrapping EVERY tensor in the program inside of a
+// FunctionalTensorWrapper, which knows about its alias'd tensors.
+//
+// See Note [Functionalization: Alias Removal] for details on the aliasing
+// machinery. See Note [Functionalization: Mutation Removal] for details on
+// mutation removal.
+struct TORCH_API FunctionalTensorWrapper : public c10::TensorImpl {
+  explicit FunctionalTensorWrapper(const Tensor& value);
+  // Additional constructor to create a FunctionalTensorWrapper directly from an
+  // underlying tensor that was created from a view. For example, the code b =
+  // a.view1() will generate a constructor call to FunctionalTensorWrapper(b, a,
+  // view1_meta)
+  explicit FunctionalTensorWrapper(
+      const Tensor& view_value,
+      const FunctionalTensorWrapper* base,
+      const functionalization::ViewMeta& meta);
+
+  // Get the underlying, actual tensor, that doesn't know anything about
+  // functionalization.
+  const Tensor& value() const {
+    return value_;
+  }
+  // The concept of "level" is only ever important to functorch; it's exposed
+  // here as more of a hook for functorch to use.
+  int64_t level() const {
+    return level_;
+  }
+  void set_level(int64_t level) {
+    level_ = level;
+  }
+  bool has_metadata_mutation() const {
+    return has_metadata_mutation_;
+  }
+
+  void mark_mutation() {
+    functional_storage_impl()->mark_mutation();
+  }
+  // Denotes a mutation that's hidden from autograd,
+  // e.g. for the purposes of passing a tensor to a triton kernel
+  void mark_mutation_hidden_from_autograd() {
+    functional_storage_impl()->mark_mutation_hidden_from_autograd();
+  }
+  void mark_mutation_during_no_grad_or_inference_mode() {
+    functional_storage_impl()->mark_mutation_during_no_grad_or_inference_mode();
+  }
+  // Are all the mutations happening to the tensor hidden from autograd
+  bool are_all_mutations_hidden_from_autograd() const {
+    return functional_storage_impl()->are_all_mutations_hidden_from_autograd();
+  }
+  // Did all mutations happen under no_grad or inference_mode
+  // (We also need to ignore mutations fully hidden from autograd here)
+  bool are_all_mutations_under_no_grad_or_inference_mode() const {
+    return functional_storage_impl()
+        ->are_all_mutations_under_no_grad_or_inference_mode();
+  }
+
+  void maybe_mark_symbolic(const functionalization::ViewMeta& meta) {
+    is_symbolic_ = is_symbolic_ | meta.has_symbolic_inputs;
+  }
+
+  bool is_symbolic() const {
+    return is_symbolic_;
+  }
+
+  // Runs the forward_fn of every ViewMeta collected in the current instance
+  // to some other base.
+  Tensor apply_view_metas(const Tensor& base);
+
+  // Sync's the underlying tensor with its alias, if it's out of date. This
+  // involves two steps: 1) Apply any pending updates/mutations to the alias 2)
+  // Replay the views (if any) to regenerate the current tensor off of the
+  // updated alias.
+  void sync_();
+  // Performs step (1) of the sync. This is its own public API because it's
+  // needed by view_inplace ops like transpose_. See Note [Functionalization
+  // Pass - Inplace View Ops]
+  void regenerate_from_base();
+  // Performs step (2) of the sync. This is its own public API because it's
+  // needed by functorch. functorch wants to make sure that all input tensors to
+  // a functionalized program have been properly synced so it can properly
+  // propagate mutations to inputs. It can't just call sync_(), because the
+  // FunctionalTensorWrapper will look like it has no aliases and sync_ will be
+  // a noop. We use the reference count on storage_ to determine if the wrapper
+  // is aliased, and by the time functorch is ready to propagate updates to
+  // inputs, any intermediate views of the input created by the program will
+  // have been deallocated. This function also returns whether or not the base
+  // actually had any updates to apply.
+  bool apply_updates();
+  // Takes the current state of value_ and snapshots it, sending it as a pending
+  // update to the alias.
+  void commit_update();
+  // When any tensor is mutated, the tensor increments its alias's "generation".
+  // Separately, each tensor maintains its own "generation" counter, which is
+  // used to determine if it's up-to-date with its alias. The act of syncing a
+  // tensor will set a tensor's generation equal to its alias's generation.
+  bool is_up_to_date() const;
+  // Freezes the storage of this tensor, preventing subsequent mutations
+  void freeze_storage() const;
+  // Every FunctionalTensorWrapper contains a vector<ViewMeta> objects
+  // describing the series of view ops that ran to generate the current tensor
+  // from the base tensor. This method is used by inplace-view ops like
+  // transpose_. It appends a ViewMeta to the existing stack, and refreshes the
+  // tensor by replaying the views off of the alias.
+  void mutate_view_meta(const at::functionalization::ViewMeta& meta);
+
+  // Custom implementation of self.set_(src)
+  void set__impl(const FunctionalTensorWrapper* other);
+
+  // Custom implementation of resize_storage_bytes_(self, new_size)
+  void storage_resize_(const c10::SymInt& new_size);
+
+  // Returns whether the current tensor's data was ever mutated
+  bool has_data_mutation();
+  //
+  // Returns whether the current FunctionalTensorWrapper
+  // experienced a set_() call.
+  bool was_storage_changed() {
+    return was_storage_changed_;
+  }
+
+  void set_storage_changed() {
+    was_storage_changed_ = true;
+  }
+
+  // A FunctionalTensor is considered a base if its not a view of another
+  // tensor.
+  bool isBaseTensor() const {
+    return view_metas_.empty();
+  }
+
+  c10::SymInt get_storage_size(bool before) {
+    return functional_storage_impl()->get_storage_size(before);
+  }
+
+  // Returns whether the FunctionalTensor experienced an
+  // untyped_storage().resize_() call
+  bool was_inductor_storage_resized() {
+    return functional_storage_impl()->was_inductor_storage_resized();
+  }
+
+  // The functionalization pass can be used to remove mutations.
+  // It does so by replacing any mutation op with it's corresponding
+  // out-of-place op, followed by a call to replace_(). e.g:
+  //
+  // a.add_(1)
+  //
+  // will turn into:
+  //
+  // tmp = a.add(1)
+  // a.replace_(tmp)
+  //
+  // replace_() swaps out the wrapped tensor, value_, with tmp.
+  void replace_(const Tensor& other, bool from_lazy_regenerate = false);
+
+  bool is_multi_output_view() {
+    return is_multi_output_view_;
+  }
+
+  // See Note[resize_() in functionalization pass]
+  void maybe_replace_storage(const Tensor& other);
+
+  // Replaces the storage with a new functional storage,
+  // and clears the view_metas_ stack.
+  // WARNING: Calling this function will sever the aliasing relationship between
+  // the current FunctionalTensorWrapper and any of its outstanding aliases.
+  // Please only call if you know what you're doing.
+  void _unsafe_reset_storage();
+
+  c10::intrusive_ptr<TensorImpl> shallow_copy_and_detach(
+      const c10::VariableVersion& version_counter,
+      bool allow_tensor_metadata_change) const override;
+
+  c10::intrusive_ptr<TensorImpl> shallow_copy_and_detach(
+      c10::VariableVersion&& version_counter,
+      bool allow_tensor_metadata_change) const override;
+
+  ~FunctionalTensorWrapper() override = default;
+
+  // FunctionalTensorWrapper overrides all custom size/stride function,
+  // so that if the inner tensor has a custom implementation
+  // we make sure to call that implementation.
+  at::IntArrayRef sizes_custom() const override;
+  at::IntArrayRef strides_custom() const override;
+  int64_t dim_custom() const override;
+  int64_t numel_custom() const override;
+  bool is_contiguous_custom(at::MemoryFormat memory_format) const override;
+  c10::SymIntArrayRef sym_sizes_custom() const override;
+  c10::SymInt sym_size_custom(int64_t d) const override;
+  c10::SymIntArrayRef sym_strides_custom() const override;
+  c10::SymInt sym_storage_offset_custom() const override;
+  c10::Device device_custom() const override;
+  c10::Layout layout_impl() const override;
+
+ private:
+  const char* tensorimpl_type_name() const override;
+  void set_constructor_metadata();
+  functionalization::FunctionalStorageImpl* functional_storage_impl() const;
+
+  // This is used to re-implement shallow_copy_and_detach for
+  // FunctionalTensorWrapper. The implementation is identical, but we just need
+  // to return a subclass instead of a plain TensorImpl.
+  // TODO: maybe it's possible to arrange for that to happen automatically
+  // without an override here?
+  template <typename VariableVersion>
+  c10::intrusive_ptr<TensorImpl> shallow_copy_and_detach_core(
+      VariableVersion&& version_counter,
+      bool allow_tensor_metadata_change) const;
+
+  void shallow_copy_from(const c10::intrusive_ptr<TensorImpl>& impl) override;
+  void copy_tensor_metadata_and_refresh(
+      const FunctionalTensorWrapper* src_impl,
+      FunctionalTensorWrapper* dest_impl,
+      const c10::VariableVersion& version_counter,
+      bool allow_tensor_metadata_change) const;
+
+  // Note that value is not taken by reference: internally, the wrapper will
+  // change the value tensor that it points to over time.
+  Tensor value_;
+  int64_t level_{};
+  // These two counters are used for identifying
+  // whether all the mutations on a given tensor are hidden from autograd or
+  // not. If we have an input mutation that is hidden from autograd, then once
+  // we convert the input mutation to a copy_() we know it will be safe to hide
+  // the copy_() from autograd as well.
+  bool has_metadata_mutation_ = false;
+  bool is_multi_output_view_ = false;
+  // Did the tensor experience a set_() call.
+  bool was_storage_changed_ = false;
+  // Did the tensor experience any view operation with symbolic int.
+  bool is_symbolic_ = false;
+
+  size_t generation_ = 0;
+  std::vector<at::functionalization::ViewMeta> view_metas_;
+
+ protected:
+  static void copy_tensor_metadata(
+      const FunctionalTensorWrapper* src_impl,
+      FunctionalTensorWrapper* dest_impl,
+      const c10::VariableVersion& version_counter,
+      bool allow_tensor_metadata_change);
+};
+
+// Utility functions for the functionalization pass.
+
+namespace functionalization {
+namespace impl {
+
+TORCH_API inline FunctionalTensorWrapper* unsafeGetFunctionalWrapper(
+    const Tensor& tensor) {
+  auto functional_impl =
+      static_cast<FunctionalTensorWrapper*>(tensor.unsafeGetTensorImpl());
+  TORCH_INTERNAL_ASSERT_DEBUG_ONLY(functional_impl != nullptr);
+  return functional_impl;
+}
+
+TORCH_API bool isBaseTensor(const at::Tensor& tensor);
+
+TORCH_API bool isFunctionalTensor(const at::Tensor& tensor);
+TORCH_API bool isFunctionalTensor(const std::optional<Tensor>& t);
+TORCH_API bool isFunctionalTensor(
+    const c10::List<std::optional<Tensor>>& t_list);
+TORCH_API bool isFunctionalTensor(ITensorListRef list);
+
+TORCH_API Tensor to_functional_tensor(const Tensor& tensor);
+TORCH_API std::optional<Tensor> to_functional_tensor(
+    const std::optional<Tensor>& tensor);
+TORCH_API c10::List<std::optional<Tensor>> to_functional_tensor(
+    const c10::List<std::optional<Tensor>>& t_list);
+TORCH_API std::vector<Tensor> to_functional_tensor(ITensorListRef t_list);
+
+TORCH_API void freeze_functional_tensor(const Tensor& tensor);
+
+TORCH_API Tensor
+from_functional_tensor(const Tensor& tensor, bool assert_functional = true);
+TORCH_API std::optional<Tensor> from_functional_tensor(
+    const std::optional<Tensor>& t,
+    bool assert_functional = true);
+TORCH_API c10::List<std::optional<Tensor>> from_functional_tensor(
+    const c10::List<std::optional<Tensor>>& t_list);
+TORCH_API std::vector<Tensor> from_functional_tensor(ITensorListRef t_list);
+
+TORCH_API void sync(const at::Tensor& t);
+TORCH_API void sync(const std::optional<Tensor>& t);
+TORCH_API void sync(const c10::List<std::optional<Tensor>>& t_list);
+TORCH_API void sync(ITensorListRef t_list);
+
+TORCH_API void replace_(const Tensor& functional_tensor, const Tensor& other);
+TORCH_API void replace_(
+    const ITensorListRef functional_tensor,
+    ITensorListRef other);
+
+TORCH_API void commit_update(const Tensor& functional_tensor);
+TORCH_API void commit_update(ITensorListRef functional_tensor);
+
+TORCH_API void unsafe_reset_storage(const Tensor& functional_tensor);
+
+TORCH_API void mark_mutation_hidden_from_autograd(
+    const Tensor& functional_tensor);
+
+TORCH_API bool are_all_mutations_hidden_from_autograd(
+    const Tensor& functional_tensor);
+
+TORCH_API bool are_all_mutations_under_no_grad_or_inference_mode(
+    const Tensor& functional_tensor);
+
+// These two methods are XLA-specific logic and are no-ops
+// for the normal functionalization flow.
+TORCH_API void propagate_xla_data(
+    const Tensor& functional_tensor,
+    const Tensor& other);
+TORCH_API void propagate_xla_data(
+    const ITensorListRef functional_tensor,
+    ITensorListRef other);
+
+TORCH_API void propagate_xla_data_direct(
+    const Tensor& tensor,
+    const Tensor& other);
+TORCH_API void propagate_xla_data_direct(
+    const ITensorListRef tensor,
+    ITensorListRef other);
+
+Tensor create_functional_tensor_with_view_meta(
+    const Tensor& view_to_wrap,
+    const Tensor& base,
+    functionalization::ViewMeta meta,
+    int64_t out_idx = 0);
+std::vector<Tensor> create_functional_tensor_with_view_meta(
+    ITensorListRef view_to_wrap,
+    const Tensor& base,
+    const functionalization::ViewMeta& meta);
+
+void mutate_view_meta(
+    const Tensor& self,
+    const functionalization::ViewMeta& meta);
+
+void set_sizes_strides_offset(const Tensor& out, const Tensor& meta_out);
+void set_sizes_strides_offset(
+    const std::vector<Tensor>& outs,
+    const std::vector<Tensor>& meta_outs);
+
+//  ~~~~~ TLS used in functionalization ~~~~~
+
+TORCH_API bool getFunctionalizationReapplyViewsTLS();
+TORCH_API void setFunctionalizationReapplyViewsTLS(bool reapply_views);
+
+class TORCH_API FunctionalizationReapplyViewsGuard {
+ public:
+  FunctionalizationReapplyViewsGuard(bool reapply_views)
+      : prev_(getFunctionalizationReapplyViewsTLS()) {
+    setFunctionalizationReapplyViewsTLS(reapply_views);
+  }
+
+  ~FunctionalizationReapplyViewsGuard() {
+    setFunctionalizationReapplyViewsTLS(prev_);
+  }
+
+  FunctionalizationReapplyViewsGuard(
+      const FunctionalizationReapplyViewsGuard&) = delete;
+  FunctionalizationReapplyViewsGuard operator=(
+      const FunctionalizationReapplyViewsGuard&) = delete;
+  FunctionalizationReapplyViewsGuard(FunctionalizationReapplyViewsGuard&&) =
+      delete;
+  FunctionalizationReapplyViewsGuard operator=(
+      FunctionalizationReapplyViewsGuard&&) = delete;
+
+ private:
+  bool prev_;
+};
+
+} // namespace impl
+
+// Helper function to call an out-of-place composite aten kernel that may use
+// mutations / views internally, and functionalize them.
+TORCH_API void functionalize_op_helper(
+    const c10::OperatorHandle& op,
+    torch::jit::Stack* stack);
+
+template <class Op, bool symint, class ReturnType, class... ParameterTypes>
+struct _functionalize_aten_op final {};
+
+template <class Op, bool symint, class ReturnType, class... ParameterTypes>
+struct _functionalize_aten_op<Op, symint, ReturnType(ParameterTypes...)> final {
+  static ReturnType call(
+      typename c10::maybe_keep_symint<symint, ParameterTypes>::type... args) {
+    using FuncType = ReturnType(
+        typename c10::maybe_keep_symint<symint, ParameterTypes>::type...);
+    auto op = c10::Dispatcher::singleton()
+                  .findSchemaOrThrow(
+                      (const char*)Op::name, (const char*)Op::overload_name)
+                  .typed<FuncType>();
+
+    return c10::impl::BoxedKernelWrapper<FuncType>::call(
+        c10::BoxedKernel::makeFromFunction<functionalize_op_helper>(),
+        op,
+        // BoxedKernelWrapper knows to ignore this keyset argument,
+        // because functionalize_op_helper doesn't take in a DispatchKeySet
+        c10::DispatchKeySet(),
+        args...);
+  }
+};
+
+template <class Op>
+using functionalize_aten_op =
+    _functionalize_aten_op<Op, false, typename Op::schema>;
+
+template <class Op>
+using functionalize_aten_op_symint =
+    _functionalize_aten_op<Op, true, typename Op::schema>;
+
+} // namespace functionalization
+} // namespace at

lib/python3.12/site-packages/torch/include/ATen/Functions.h

@@ -0,0 +1,1458 @@
+#pragma once
+
+// @generated by torchgen/gen.py from Functions.h
+
+#ifdef TORCH_ASSERT_NO_OPERATORS
+#error This change adds a dependency on native_functions.yaml,            \
+  meaning the file will need to be re-compiled every time an operator     \
+  is changed or added. Consider if your change would be better placed in  \
+  another file, or if a more specific header might achieve the same goal. \
+  See NOTE: [Tensor vs. TensorBase]
+#endif
+
+#if defined(AT_PER_OPERATOR_HEADERS) && defined(TORCH_ASSERT_ONLY_METHOD_OPERATORS)
+#error This change adds a dependency on all pytorch operators, meaning the     \
+  file will need to be re-compiled every time an operator is changed or added. \
+  Consider including a specific operator from <ATen/ops/{my_operator}.h> and   \
+  see NOTE [TORCH_ASSERT_ONLY_METHOD_OPERATORS].
+#endif
+
+// NOTE: [TORCH_ASSERT_ONLY_METHOD_OPERATORS]
+//
+// In ATen, certain generated headers files include the definitions of
+// every single operator in PyTorch. Unfortunately this means every
+// time an operator signature is updated or changed in
+// native_functions.yaml, you (and every other PyTorch developer) need
+// to recompile every source file that includes any of these headers.
+//
+// To break up these header dependencies, and improve incremental
+// build times for all PyTorch developers. These headers are split
+// into per-operator headers in the `ATen/ops` folder. This limits
+// incremental builds to only changes to methods of `Tensor`, or files
+// that use the specific operator being changed. With `at::sum` as an
+// example, you should include
+//
+//   <ATen/ops/sum.h>               // instead of ATen/Functions.h
+//   <ATen/ops/sum_native.h>        // instead of ATen/NativeFunctions.h
+//   <ATen/ops/sum_ops.h>           // instead of ATen/Operators.h
+//   <ATen/ops/sum_cpu_dispatch.h>  // instead of ATen/CPUFunctions.h
+//
+// However, even if you're careful to use this in your own code.
+// `Functions.h` might be included indirectly through another header
+// without you realising. To avoid this, you can add
+//
+//   #define TORCH_ASSERT_ONLY_METHOD_OPERATORS
+//
+// to the top of your source file. This way any time the non-specific
+// headers are included, the compiler will error out.
+//
+// Also, be aware that `ops` are not available in all build
+// configurations (namely fb-internal) so you must guard these
+// includes with `#ifdef AT_PER_OPERATOR_HEADERS`. e.g.
+//
+//   #ifndef AT_PER_OPERATOR_HEADERS
+//   #include <ATen/Functions.h>
+//   #else
+//   #include <ATen/ops/sum.h>
+//   #endif
+
+#include <ATen/Context.h>
+#include <ATen/DeviceGuard.h>
+#include <ATen/TensorUtils.h>
+#include <ATen/TracerMode.h>
+#include <ATen/core/Generator.h>
+#include <ATen/core/Reduction.h>
+#include <c10/core/SymInt.h>
+#include <ATen/core/Tensor.h>
+#include <c10/core/Scalar.h>
+#include <c10/core/Storage.h>
+#include <c10/core/TensorOptions.h>
+#include <c10/util/Deprecated.h>
+#include <optional>
+#include <c10/util/OptionalArrayRef.h>
+
+#include <ATen/ops/from_blob.h>
+#include <ATen/ops/tensor.h>
+
+#include <ATen/ops/_adaptive_avg_pool2d.h>
+#include <ATen/ops/_adaptive_avg_pool2d_backward.h>
+#include <ATen/ops/_adaptive_avg_pool3d.h>
+#include <ATen/ops/_adaptive_avg_pool3d_backward.h>
+#include <ATen/ops/_add_batch_dim.h>
+#include <ATen/ops/_add_relu.h>
+#include <ATen/ops/_addmm_activation.h>
+#include <ATen/ops/_aminmax.h>
+#include <ATen/ops/_amp_foreach_non_finite_check_and_unscale.h>
+#include <ATen/ops/_amp_update_scale.h>
+#include <ATen/ops/_assert_async.h>
+#include <ATen/ops/_assert_scalar.h>
+#include <ATen/ops/_assert_tensor_metadata.h>
+#include <ATen/ops/_autocast_to_full_precision.h>
+#include <ATen/ops/_autocast_to_reduced_precision.h>
+#include <ATen/ops/_backward.h>
+#include <ATen/ops/_batch_norm_impl_index.h>
+#include <ATen/ops/_batch_norm_impl_index_backward.h>
+#include <ATen/ops/_batch_norm_no_update.h>
+#include <ATen/ops/_batch_norm_with_update.h>
+#include <ATen/ops/_cast_Byte.h>
+#include <ATen/ops/_cast_Char.h>
+#include <ATen/ops/_cast_Double.h>
+#include <ATen/ops/_cast_Float.h>
+#include <ATen/ops/_cast_Half.h>
+#include <ATen/ops/_cast_Int.h>
+#include <ATen/ops/_cast_Long.h>
+#include <ATen/ops/_cast_Short.h>
+#include <ATen/ops/_cdist_backward.h>
+#include <ATen/ops/_cdist_forward.h>
+#include <ATen/ops/_cholesky_solve_helper.h>
+#include <ATen/ops/_choose_qparams_per_tensor.h>
+#include <ATen/ops/_chunk_cat.h>
+#include <ATen/ops/_coalesce.h>
+#include <ATen/ops/_coalesced.h>
+#include <ATen/ops/_compute_linear_combination.h>
+#include <ATen/ops/_conj.h>
+#include <ATen/ops/_conj_copy.h>
+#include <ATen/ops/_conj_physical.h>
+#include <ATen/ops/_conv_depthwise2d.h>
+#include <ATen/ops/_convert_indices_from_coo_to_csr.h>
+#include <ATen/ops/_convert_indices_from_csr_to_coo.h>
+#include <ATen/ops/_convert_weight_to_int4pack.h>
+#include <ATen/ops/_convert_weight_to_int4pack_for_cpu.h>
+#include <ATen/ops/_convolution.h>
+#include <ATen/ops/_convolution_double_backward.h>
+#include <ATen/ops/_convolution_mode.h>
+#include <ATen/ops/_copy_from.h>
+#include <ATen/ops/_copy_from_and_resize.h>
+#include <ATen/ops/_cslt_compress.h>
+#include <ATen/ops/_cslt_sparse_mm.h>
+#include <ATen/ops/_cslt_sparse_mm_search.h>
+#include <ATen/ops/_ctc_loss.h>
+#include <ATen/ops/_ctc_loss_backward.h>
+#include <ATen/ops/_cudnn_ctc_loss.h>
+#include <ATen/ops/_cudnn_init_dropout_state.h>
+#include <ATen/ops/_cudnn_rnn.h>
+#include <ATen/ops/_cudnn_rnn_backward.h>
+#include <ATen/ops/_cudnn_rnn_flatten_weight.h>
+#include <ATen/ops/_cufft_clear_plan_cache.h>
+#include <ATen/ops/_cufft_get_plan_cache_max_size.h>
+#include <ATen/ops/_cufft_get_plan_cache_size.h>
+#include <ATen/ops/_cufft_set_plan_cache_max_size.h>
+#include <ATen/ops/_cummax_helper.h>
+#include <ATen/ops/_cummin_helper.h>
+#include <ATen/ops/_debug_has_internal_overlap.h>
+#include <ATen/ops/_dimI.h>
+#include <ATen/ops/_dimV.h>
+#include <ATen/ops/_dim_arange.h>
+#include <ATen/ops/_dirichlet_grad.h>
+#include <ATen/ops/_efficient_attention_backward.h>
+#include <ATen/ops/_efficient_attention_forward.h>
+#include <ATen/ops/_efficientzerotensor.h>
+#include <ATen/ops/_embedding_bag.h>
+#include <ATen/ops/_embedding_bag_backward.h>
+#include <ATen/ops/_embedding_bag_dense_backward.h>
+#include <ATen/ops/_embedding_bag_forward_only.h>
+#include <ATen/ops/_embedding_bag_per_sample_weights_backward.h>
+#include <ATen/ops/_embedding_bag_sparse_backward.h>
+#include <ATen/ops/_empty_affine_quantized.h>
+#include <ATen/ops/_empty_per_channel_affine_quantized.h>
+#include <ATen/ops/_euclidean_dist.h>
+#include <ATen/ops/_fake_quantize_learnable_per_channel_affine.h>
+#include <ATen/ops/_fake_quantize_learnable_per_channel_affine_backward.h>
+#include <ATen/ops/_fake_quantize_learnable_per_tensor_affine.h>
+#include <ATen/ops/_fake_quantize_learnable_per_tensor_affine_backward.h>
+#include <ATen/ops/_fake_quantize_per_tensor_affine_cachemask_tensor_qparams.h>
+#include <ATen/ops/_fft_c2c.h>
+#include <ATen/ops/_fft_c2r.h>
+#include <ATen/ops/_fft_r2c.h>
+#include <ATen/ops/_fill_mem_eff_dropout_mask.h>
+#include <ATen/ops/_flash_attention_backward.h>
+#include <ATen/ops/_flash_attention_forward.h>
+#include <ATen/ops/_foobar.h>
+#include <ATen/ops/_foreach_abs.h>
+#include <ATen/ops/_foreach_acos.h>
+#include <ATen/ops/_foreach_add.h>
+#include <ATen/ops/_foreach_addcdiv.h>
+#include <ATen/ops/_foreach_addcmul.h>
+#include <ATen/ops/_foreach_asin.h>
+#include <ATen/ops/_foreach_atan.h>
+#include <ATen/ops/_foreach_ceil.h>
+#include <ATen/ops/_foreach_clamp_max.h>
+#include <ATen/ops/_foreach_clamp_min.h>
+#include <ATen/ops/_foreach_copy.h>
+#include <ATen/ops/_foreach_cos.h>
+#include <ATen/ops/_foreach_cosh.h>
+#include <ATen/ops/_foreach_div.h>
+#include <ATen/ops/_foreach_erf.h>
+#include <ATen/ops/_foreach_erfc.h>
+#include <ATen/ops/_foreach_exp.h>
+#include <ATen/ops/_foreach_expm1.h>
+#include <ATen/ops/_foreach_floor.h>
+#include <ATen/ops/_foreach_frac.h>
+#include <ATen/ops/_foreach_lerp.h>
+#include <ATen/ops/_foreach_lgamma.h>
+#include <ATen/ops/_foreach_log.h>
+#include <ATen/ops/_foreach_log10.h>
+#include <ATen/ops/_foreach_log1p.h>
+#include <ATen/ops/_foreach_log2.h>
+#include <ATen/ops/_foreach_max.h>
+#include <ATen/ops/_foreach_maximum.h>
+#include <ATen/ops/_foreach_minimum.h>
+#include <ATen/ops/_foreach_mul.h>
+#include <ATen/ops/_foreach_neg.h>
+#include <ATen/ops/_foreach_norm.h>
+#include <ATen/ops/_foreach_pow.h>
+#include <ATen/ops/_foreach_reciprocal.h>
+#include <ATen/ops/_foreach_round.h>
+#include <ATen/ops/_foreach_rsqrt.h>
+#include <ATen/ops/_foreach_sigmoid.h>
+#include <ATen/ops/_foreach_sign.h>
+#include <ATen/ops/_foreach_sin.h>
+#include <ATen/ops/_foreach_sinh.h>
+#include <ATen/ops/_foreach_sqrt.h>
+#include <ATen/ops/_foreach_sub.h>
+#include <ATen/ops/_foreach_tan.h>
+#include <ATen/ops/_foreach_tanh.h>
+#include <ATen/ops/_foreach_trunc.h>
+#include <ATen/ops/_foreach_zero.h>
+#include <ATen/ops/_functional_assert_async.h>
+#include <ATen/ops/_functional_assert_scalar.h>
+#include <ATen/ops/_functional_sym_constrain_range.h>
+#include <ATen/ops/_functional_sym_constrain_range_for_size.h>
+#include <ATen/ops/_fused_adagrad.h>
+#include <ATen/ops/_fused_adam.h>
+#include <ATen/ops/_fused_adamw.h>
+#include <ATen/ops/_fused_dropout.h>
+#include <ATen/ops/_fused_moving_avg_obs_fq_helper.h>
+#include <ATen/ops/_fused_sdp_choice.h>
+#include <ATen/ops/_fused_sgd.h>
+#include <ATen/ops/_fw_primal.h>
+#include <ATen/ops/_fw_primal_copy.h>
+#include <ATen/ops/_gather_sparse_backward.h>
+#include <ATen/ops/_grid_sampler_2d_cpu_fallback.h>
+#include <ATen/ops/_grid_sampler_2d_cpu_fallback_backward.h>
+#include <ATen/ops/_has_compatible_shallow_copy_type.h>
+#include <ATen/ops/_has_same_storage_numel.h>
+#include <ATen/ops/_histogramdd_bin_edges.h>
+#include <ATen/ops/_histogramdd_from_bin_cts.h>
+#include <ATen/ops/_histogramdd_from_bin_tensors.h>
+#include <ATen/ops/_index_put_impl.h>
+#include <ATen/ops/_indices.h>
+#include <ATen/ops/_indices_copy.h>
+#include <ATen/ops/_int_mm.h>
+#include <ATen/ops/_is_all_true.h>
+#include <ATen/ops/_is_any_true.h>
+#include <ATen/ops/_is_zerotensor.h>
+#include <ATen/ops/_jagged_to_padded_dense_forward.h>
+#include <ATen/ops/_lazy_clone.h>
+#include <ATen/ops/_linalg_check_errors.h>
+#include <ATen/ops/_linalg_det.h>
+#include <ATen/ops/_linalg_eigh.h>
+#include <ATen/ops/_linalg_eigvals.h>
+#include <ATen/ops/_linalg_slogdet.h>
+#include <ATen/ops/_linalg_solve_ex.h>
+#include <ATen/ops/_linalg_svd.h>
+#include <ATen/ops/_local_scalar_dense.h>
+#include <ATen/ops/_log_softmax.h>
+#include <ATen/ops/_log_softmax_backward_data.h>
+#include <ATen/ops/_logcumsumexp.h>
+#include <ATen/ops/_lstm_mps.h>
+#include <ATen/ops/_lu_with_info.h>
+#include <ATen/ops/_make_dep_token.h>
+#include <ATen/ops/_make_dual.h>
+#include <ATen/ops/_make_dual_copy.h>
+#include <ATen/ops/_make_per_channel_quantized_tensor.h>
+#include <ATen/ops/_make_per_tensor_quantized_tensor.h>
+#include <ATen/ops/_masked_scale.h>
+#include <ATen/ops/_masked_softmax.h>
+#include <ATen/ops/_masked_softmax_backward.h>
+#include <ATen/ops/_mixed_dtypes_linear.h>
+#include <ATen/ops/_mkldnn_reshape.h>
+#include <ATen/ops/_mkldnn_transpose.h>
+#include <ATen/ops/_mps_convolution.h>
+#include <ATen/ops/_mps_convolution_transpose.h>
+#include <ATen/ops/_native_batch_norm_legit.h>
+#include <ATen/ops/_native_batch_norm_legit_no_training.h>
+#include <ATen/ops/_native_multi_head_attention.h>
+#include <ATen/ops/_neg_view.h>
+#include <ATen/ops/_neg_view_copy.h>
+#include <ATen/ops/_nested_compute_contiguous_strides_offsets.h>
+#include <ATen/ops/_nested_from_padded.h>
+#include <ATen/ops/_nested_from_padded_and_nested_example.h>
+#include <ATen/ops/_nested_from_padded_tensor.h>
+#include <ATen/ops/_nested_get_jagged_dummy.h>
+#include <ATen/ops/_nested_get_lengths.h>
+#include <ATen/ops/_nested_get_max_seqlen.h>
+#include <ATen/ops/_nested_get_min_seqlen.h>
+#include <ATen/ops/_nested_get_offsets.h>
+#include <ATen/ops/_nested_get_ragged_idx.h>
+#include <ATen/ops/_nested_get_values.h>
+#include <ATen/ops/_nested_get_values_copy.h>
+#include <ATen/ops/_nested_select_backward.h>
+#include <ATen/ops/_nested_sum_backward.h>
+#include <ATen/ops/_nested_tensor_from_mask.h>
+#include <ATen/ops/_nested_tensor_from_mask_left_aligned.h>
+#include <ATen/ops/_nested_tensor_from_tensor_list.h>
+#include <ATen/ops/_nested_tensor_size.h>
+#include <ATen/ops/_nested_tensor_softmax_with_shape.h>
+#include <ATen/ops/_nested_tensor_storage_offsets.h>
+#include <ATen/ops/_nested_tensor_strides.h>
+#include <ATen/ops/_nested_view_from_buffer.h>
+#include <ATen/ops/_nested_view_from_buffer_copy.h>
+#include <ATen/ops/_nested_view_from_jagged.h>
+#include <ATen/ops/_nested_view_from_jagged_copy.h>
+#include <ATen/ops/_new_zeros_with_same_feature_meta.h>
+#include <ATen/ops/_nnpack_available.h>
+#include <ATen/ops/_nnpack_spatial_convolution.h>
+#include <ATen/ops/_nnz.h>
+#include <ATen/ops/_pack_padded_sequence.h>
+#include <ATen/ops/_pack_padded_sequence_backward.h>
+#include <ATen/ops/_pad_circular.h>
+#include <ATen/ops/_pad_enum.h>
+#include <ATen/ops/_pad_packed_sequence.h>
+#include <ATen/ops/_padded_dense_to_jagged_forward.h>
+#include <ATen/ops/_pdist_backward.h>
+#include <ATen/ops/_pdist_forward.h>
+#include <ATen/ops/_pin_memory.h>
+#include <ATen/ops/_prelu_kernel.h>
+#include <ATen/ops/_prelu_kernel_backward.h>
+#include <ATen/ops/_print.h>
+#include <ATen/ops/_propagate_xla_data.h>
+#include <ATen/ops/_remove_batch_dim.h>
+#include <ATen/ops/_reshape_alias.h>
+#include <ATen/ops/_reshape_alias_copy.h>
+#include <ATen/ops/_reshape_copy.h>
+#include <ATen/ops/_reshape_from_tensor.h>
+#include <ATen/ops/_resize_output.h>
+#include <ATen/ops/_rowwise_prune.h>
+#include <ATen/ops/_safe_softmax.h>
+#include <ATen/ops/_sample_dirichlet.h>
+#include <ATen/ops/_saturate_weight_to_fp16.h>
+#include <ATen/ops/_scaled_dot_product_attention_math.h>
+#include <ATen/ops/_scaled_dot_product_attention_math_for_mps.h>
+#include <ATen/ops/_scaled_dot_product_cudnn_attention.h>
+#include <ATen/ops/_scaled_dot_product_cudnn_attention_backward.h>
+#include <ATen/ops/_scaled_dot_product_efficient_attention.h>
+#include <ATen/ops/_scaled_dot_product_efficient_attention_backward.h>
+#include <ATen/ops/_scaled_dot_product_flash_attention.h>
+#include <ATen/ops/_scaled_dot_product_flash_attention_backward.h>
+#include <ATen/ops/_scaled_dot_product_flash_attention_for_cpu.h>
+#include <ATen/ops/_scaled_dot_product_flash_attention_for_cpu_backward.h>
+#include <ATen/ops/_scaled_dot_product_fused_attention_overrideable.h>
+#include <ATen/ops/_scaled_dot_product_fused_attention_overrideable_backward.h>
+#include <ATen/ops/_scaled_mm.h>
+#include <ATen/ops/_segment_reduce_backward.h>
+#include <ATen/ops/_shape_as_tensor.h>
+#include <ATen/ops/_slow_conv2d_backward.h>
+#include <ATen/ops/_slow_conv2d_forward.h>
+#include <ATen/ops/_sobol_engine_draw.h>
+#include <ATen/ops/_sobol_engine_ff.h>
+#include <ATen/ops/_sobol_engine_initialize_state.h>
+#include <ATen/ops/_sobol_engine_scramble.h>
+#include <ATen/ops/_softmax.h>
+#include <ATen/ops/_softmax_backward_data.h>
+#include <ATen/ops/_sparse_addmm.h>
+#include <ATen/ops/_sparse_broadcast_to.h>
+#include <ATen/ops/_sparse_broadcast_to_copy.h>
+#include <ATen/ops/_sparse_bsc_tensor_unsafe.h>
+#include <ATen/ops/_sparse_bsr_tensor_unsafe.h>
+#include <ATen/ops/_sparse_compressed_tensor_unsafe.h>
+#include <ATen/ops/_sparse_compressed_tensor_with_dims.h>
+#include <ATen/ops/_sparse_coo_tensor_unsafe.h>
+#include <ATen/ops/_sparse_coo_tensor_with_dims.h>
+#include <ATen/ops/_sparse_coo_tensor_with_dims_and_tensors.h>
+#include <ATen/ops/_sparse_csc_tensor_unsafe.h>
+#include <ATen/ops/_sparse_csr_prod.h>
+#include <ATen/ops/_sparse_csr_sum.h>
+#include <ATen/ops/_sparse_csr_tensor_unsafe.h>
+#include <ATen/ops/_sparse_log_softmax.h>
+#include <ATen/ops/_sparse_log_softmax_backward_data.h>
+#include <ATen/ops/_sparse_mask_projection.h>
+#include <ATen/ops/_sparse_mm.h>
+#include <ATen/ops/_sparse_mm_reduce_impl.h>
+#include <ATen/ops/_sparse_mm_reduce_impl_backward.h>
+#include <ATen/ops/_sparse_semi_structured_addmm.h>
+#include <ATen/ops/_sparse_semi_structured_apply.h>
+#include <ATen/ops/_sparse_semi_structured_apply_dense.h>
+#include <ATen/ops/_sparse_semi_structured_linear.h>
+#include <ATen/ops/_sparse_semi_structured_mm.h>
+#include <ATen/ops/_sparse_semi_structured_tile.h>
+#include <ATen/ops/_sparse_softmax.h>
+#include <ATen/ops/_sparse_softmax_backward_data.h>
+#include <ATen/ops/_sparse_sparse_matmul.h>
+#include <ATen/ops/_sparse_sum.h>
+#include <ATen/ops/_sparse_sum_backward.h>
+#include <ATen/ops/_spdiags.h>
+#include <ATen/ops/_spsolve.h>
+#include <ATen/ops/_stack.h>
+#include <ATen/ops/_standard_gamma.h>
+#include <ATen/ops/_standard_gamma_grad.h>
+#include <ATen/ops/_test_ambiguous_defaults.h>
+#include <ATen/ops/_test_autograd_multiple_dispatch.h>
+#include <ATen/ops/_test_autograd_multiple_dispatch_view.h>
+#include <ATen/ops/_test_autograd_multiple_dispatch_view_copy.h>
+#include <ATen/ops/_test_check_tensor.h>
+#include <ATen/ops/_test_functorch_fallback.h>
+#include <ATen/ops/_test_optional_filled_intlist.h>
+#include <ATen/ops/_test_optional_floatlist.h>
+#include <ATen/ops/_test_optional_intlist.h>
+#include <ATen/ops/_test_parallel_materialize.h>
+#include <ATen/ops/_test_serialization_subcmul.h>
+#include <ATen/ops/_test_string_default.h>
+#include <ATen/ops/_test_warn_in_autograd.h>
+#include <ATen/ops/_thnn_differentiable_gru_cell_backward.h>
+#include <ATen/ops/_thnn_differentiable_lstm_cell_backward.h>
+#include <ATen/ops/_thnn_fused_gru_cell.h>
+#include <ATen/ops/_thnn_fused_gru_cell_backward.h>
+#include <ATen/ops/_thnn_fused_lstm_cell.h>
+#include <ATen/ops/_thnn_fused_lstm_cell_backward.h>
+#include <ATen/ops/_thnn_fused_lstm_cell_backward_impl.h>
+#include <ATen/ops/_to_copy.h>
+#include <ATen/ops/_to_cpu.h>
+#include <ATen/ops/_to_dense.h>
+#include <ATen/ops/_to_sparse.h>
+#include <ATen/ops/_to_sparse_bsc.h>
+#include <ATen/ops/_to_sparse_bsr.h>
+#include <ATen/ops/_to_sparse_csc.h>
+#include <ATen/ops/_to_sparse_csr.h>
+#include <ATen/ops/_to_sparse_semi_structured.h>
+#include <ATen/ops/_transform_bias_rescale_qkv.h>
+#include <ATen/ops/_transformer_encoder_layer_fwd.h>
+#include <ATen/ops/_trilinear.h>
+#include <ATen/ops/_triton_multi_head_attention.h>
+#include <ATen/ops/_triton_scaled_dot_attention.h>
+#include <ATen/ops/_unique.h>
+#include <ATen/ops/_unique2.h>
+#include <ATen/ops/_unpack_dual.h>
+#include <ATen/ops/_unsafe_index.h>
+#include <ATen/ops/_unsafe_index_put.h>
+#include <ATen/ops/_unsafe_masked_index.h>
+#include <ATen/ops/_unsafe_masked_index_put_accumulate.h>
+#include <ATen/ops/_unsafe_view.h>
+#include <ATen/ops/_upsample_bicubic2d_aa.h>
+#include <ATen/ops/_upsample_bicubic2d_aa_backward.h>
+#include <ATen/ops/_upsample_bilinear2d_aa.h>
+#include <ATen/ops/_upsample_bilinear2d_aa_backward.h>
+#include <ATen/ops/_upsample_nearest_exact1d.h>
+#include <ATen/ops/_upsample_nearest_exact1d_backward.h>
+#include <ATen/ops/_upsample_nearest_exact2d.h>
+#include <ATen/ops/_upsample_nearest_exact2d_backward.h>
+#include <ATen/ops/_upsample_nearest_exact3d.h>
+#include <ATen/ops/_upsample_nearest_exact3d_backward.h>
+#include <ATen/ops/_use_cudnn_ctc_loss.h>
+#include <ATen/ops/_use_cudnn_rnn_flatten_weight.h>
+#include <ATen/ops/_validate_compressed_sparse_indices.h>
+#include <ATen/ops/_validate_sparse_bsc_tensor_args.h>
+#include <ATen/ops/_validate_sparse_bsr_tensor_args.h>
+#include <ATen/ops/_validate_sparse_compressed_tensor_args.h>
+#include <ATen/ops/_validate_sparse_coo_tensor_args.h>
+#include <ATen/ops/_validate_sparse_csc_tensor_args.h>
+#include <ATen/ops/_validate_sparse_csr_tensor_args.h>
+#include <ATen/ops/_values.h>
+#include <ATen/ops/_values_copy.h>
+#include <ATen/ops/_version.h>
+#include <ATen/ops/_weight_int4pack_mm.h>
+#include <ATen/ops/_weight_int4pack_mm_for_cpu.h>
+#include <ATen/ops/_weight_int8pack_mm.h>
+#include <ATen/ops/_weight_norm.h>
+#include <ATen/ops/_weight_norm_differentiable_backward.h>
+#include <ATen/ops/_weight_norm_interface.h>
+#include <ATen/ops/_weight_norm_interface_backward.h>
+#include <ATen/ops/_wrapped_linear_prepack.h>
+#include <ATen/ops/_wrapped_quantized_linear_prepacked.h>
+#include <ATen/ops/abs.h>
+#include <ATen/ops/absolute.h>
+#include <ATen/ops/acos.h>
+#include <ATen/ops/acosh.h>
+#include <ATen/ops/adaptive_avg_pool1d.h>
+#include <ATen/ops/adaptive_avg_pool2d.h>
+#include <ATen/ops/adaptive_avg_pool3d.h>
+#include <ATen/ops/adaptive_avg_pool3d_backward.h>
+#include <ATen/ops/adaptive_max_pool1d.h>
+#include <ATen/ops/adaptive_max_pool2d.h>
+#include <ATen/ops/adaptive_max_pool2d_backward.h>
+#include <ATen/ops/adaptive_max_pool3d.h>
+#include <ATen/ops/adaptive_max_pool3d_backward.h>
+#include <ATen/ops/add.h>
+#include <ATen/ops/addbmm.h>
+#include <ATen/ops/addcdiv.h>
+#include <ATen/ops/addcmul.h>
+#include <ATen/ops/addmm.h>
+#include <ATen/ops/addmv.h>
+#include <ATen/ops/addr.h>
+#include <ATen/ops/adjoint.h>
+#include <ATen/ops/affine_grid_generator.h>
+#include <ATen/ops/affine_grid_generator_backward.h>
+#include <ATen/ops/alias.h>
+#include <ATen/ops/alias_copy.h>
+#include <ATen/ops/align_as.h>
+#include <ATen/ops/align_tensors.h>
+#include <ATen/ops/align_to.h>
+#include <ATen/ops/all.h>
+#include <ATen/ops/allclose.h>
+#include <ATen/ops/alpha_dropout.h>
+#include <ATen/ops/amax.h>
+#include <ATen/ops/amin.h>
+#include <ATen/ops/aminmax.h>
+#include <ATen/ops/and.h>
+#include <ATen/ops/angle.h>
+#include <ATen/ops/any.h>
+#include <ATen/ops/arange.h>
+#include <ATen/ops/arccos.h>
+#include <ATen/ops/arccosh.h>
+#include <ATen/ops/arcsin.h>
+#include <ATen/ops/arcsinh.h>
+#include <ATen/ops/arctan.h>
+#include <ATen/ops/arctan2.h>
+#include <ATen/ops/arctanh.h>
+#include <ATen/ops/argmax.h>
+#include <ATen/ops/argmin.h>
+#include <ATen/ops/argsort.h>
+#include <ATen/ops/argwhere.h>
+#include <ATen/ops/as_strided.h>
+#include <ATen/ops/as_strided_copy.h>
+#include <ATen/ops/as_strided_scatter.h>
+#include <ATen/ops/asin.h>
+#include <ATen/ops/asinh.h>
+#include <ATen/ops/atan.h>
+#include <ATen/ops/atan2.h>
+#include <ATen/ops/atanh.h>
+#include <ATen/ops/atleast_1d.h>
+#include <ATen/ops/atleast_2d.h>
+#include <ATen/ops/atleast_3d.h>
+#include <ATen/ops/avg_pool1d.h>
+#include <ATen/ops/avg_pool2d.h>
+#include <ATen/ops/avg_pool2d_backward.h>
+#include <ATen/ops/avg_pool3d.h>
+#include <ATen/ops/avg_pool3d_backward.h>
+#include <ATen/ops/baddbmm.h>
+#include <ATen/ops/bartlett_window.h>
+#include <ATen/ops/batch_norm.h>
+#include <ATen/ops/batch_norm_backward.h>
+#include <ATen/ops/batch_norm_backward_elemt.h>
+#include <ATen/ops/batch_norm_backward_reduce.h>
+#include <ATen/ops/batch_norm_elemt.h>
+#include <ATen/ops/batch_norm_gather_stats.h>
+#include <ATen/ops/batch_norm_gather_stats_with_counts.h>
+#include <ATen/ops/batch_norm_stats.h>
+#include <ATen/ops/batch_norm_update_stats.h>
+#include <ATen/ops/bernoulli.h>
+#include <ATen/ops/bilinear.h>
+#include <ATen/ops/binary_cross_entropy.h>
+#include <ATen/ops/binary_cross_entropy_backward.h>
+#include <ATen/ops/binary_cross_entropy_with_logits.h>
+#include <ATen/ops/bincount.h>
+#include <ATen/ops/binomial.h>
+#include <ATen/ops/bitwise_and.h>
+#include <ATen/ops/bitwise_left_shift.h>
+#include <ATen/ops/bitwise_not.h>
+#include <ATen/ops/bitwise_or.h>
+#include <ATen/ops/bitwise_right_shift.h>
+#include <ATen/ops/bitwise_xor.h>
+#include <ATen/ops/blackman_window.h>
+#include <ATen/ops/block_diag.h>
+#include <ATen/ops/bmm.h>
+#include <ATen/ops/broadcast_tensors.h>
+#include <ATen/ops/broadcast_to.h>
+#include <ATen/ops/bucketize.h>
+#include <ATen/ops/can_cast.h>
+#include <ATen/ops/cartesian_prod.h>
+#include <ATen/ops/cat.h>
+#include <ATen/ops/cauchy.h>
+#include <ATen/ops/ccol_indices.h>
+#include <ATen/ops/ccol_indices_copy.h>
+#include <ATen/ops/cdist.h>
+#include <ATen/ops/ceil.h>
+#include <ATen/ops/celu.h>
+#include <ATen/ops/chain_matmul.h>
+#include <ATen/ops/chalf.h>
+#include <ATen/ops/channel_shuffle.h>
+#include <ATen/ops/cholesky.h>
+#include <ATen/ops/cholesky_inverse.h>
+#include <ATen/ops/cholesky_solve.h>
+#include <ATen/ops/choose_qparams_optimized.h>
+#include <ATen/ops/chunk.h>
+#include <ATen/ops/clamp.h>
+#include <ATen/ops/clamp_max.h>
+#include <ATen/ops/clamp_min.h>
+#include <ATen/ops/clip.h>
+#include <ATen/ops/clone.h>
+#include <ATen/ops/coalesce.h>
+#include <ATen/ops/col2im.h>
+#include <ATen/ops/col_indices.h>
+#include <ATen/ops/col_indices_copy.h>
+#include <ATen/ops/column_stack.h>
+#include <ATen/ops/combinations.h>
+#include <ATen/ops/complex.h>
+#include <ATen/ops/concat.h>
+#include <ATen/ops/concatenate.h>
+#include <ATen/ops/conj.h>
+#include <ATen/ops/conj_physical.h>
+#include <ATen/ops/constant_pad_nd.h>
+#include <ATen/ops/contiguous.h>
+#include <ATen/ops/conv1d.h>
+#include <ATen/ops/conv2d.h>
+#include <ATen/ops/conv3d.h>
+#include <ATen/ops/conv_depthwise3d.h>
+#include <ATen/ops/conv_tbc.h>
+#include <ATen/ops/conv_tbc_backward.h>
+#include <ATen/ops/conv_transpose1d.h>
+#include <ATen/ops/conv_transpose2d.h>
+#include <ATen/ops/conv_transpose3d.h>
+#include <ATen/ops/convolution.h>
+#include <ATen/ops/convolution_backward.h>
+#include <ATen/ops/convolution_backward_overrideable.h>
+#include <ATen/ops/convolution_overrideable.h>
+#include <ATen/ops/copy.h>
+#include <ATen/ops/copy_sparse_to_sparse.h>
+#include <ATen/ops/copysign.h>
+#include <ATen/ops/corrcoef.h>
+#include <ATen/ops/cos.h>
+#include <ATen/ops/cosh.h>
+#include <ATen/ops/cosine_embedding_loss.h>
+#include <ATen/ops/cosine_similarity.h>
+#include <ATen/ops/count_nonzero.h>
+#include <ATen/ops/cov.h>
+#include <ATen/ops/cross.h>
+#include <ATen/ops/cross_entropy_loss.h>
+#include <ATen/ops/crow_indices.h>
+#include <ATen/ops/crow_indices_copy.h>
+#include <ATen/ops/ctc_loss.h>
+#include <ATen/ops/cudnn_affine_grid_generator.h>
+#include <ATen/ops/cudnn_affine_grid_generator_backward.h>
+#include <ATen/ops/cudnn_batch_norm.h>
+#include <ATen/ops/cudnn_batch_norm_backward.h>
+#include <ATen/ops/cudnn_convolution.h>
+#include <ATen/ops/cudnn_convolution_add_relu.h>
+#include <ATen/ops/cudnn_convolution_relu.h>
+#include <ATen/ops/cudnn_convolution_transpose.h>
+#include <ATen/ops/cudnn_grid_sampler.h>
+#include <ATen/ops/cudnn_grid_sampler_backward.h>
+#include <ATen/ops/cudnn_is_acceptable.h>
+#include <ATen/ops/cummax.h>
+#include <ATen/ops/cummaxmin_backward.h>
+#include <ATen/ops/cummin.h>
+#include <ATen/ops/cumprod.h>
+#include <ATen/ops/cumprod_backward.h>
+#include <ATen/ops/cumsum.h>
+#include <ATen/ops/cumulative_trapezoid.h>
+#include <ATen/ops/data.h>
+#include <ATen/ops/deg2rad.h>
+#include <ATen/ops/dense_dim.h>
+#include <ATen/ops/dequantize.h>
+#include <ATen/ops/det.h>
+#include <ATen/ops/detach.h>
+#include <ATen/ops/detach_copy.h>
+#include <ATen/ops/diag.h>
+#include <ATen/ops/diag_embed.h>
+#include <ATen/ops/diagflat.h>
+#include <ATen/ops/diagonal.h>
+#include <ATen/ops/diagonal_backward.h>
+#include <ATen/ops/diagonal_copy.h>
+#include <ATen/ops/diagonal_scatter.h>
+#include <ATen/ops/diff.h>
+#include <ATen/ops/digamma.h>
+#include <ATen/ops/dist.h>
+#include <ATen/ops/div.h>
+#include <ATen/ops/divide.h>
+#include <ATen/ops/dot.h>
+#include <ATen/ops/dropout.h>
+#include <ATen/ops/dsplit.h>
+#include <ATen/ops/dstack.h>
+#include <ATen/ops/einsum.h>
+#include <ATen/ops/elu.h>
+#include <ATen/ops/elu_backward.h>
+#include <ATen/ops/embedding.h>
+#include <ATen/ops/embedding_backward.h>
+#include <ATen/ops/embedding_bag.h>
+#include <ATen/ops/embedding_dense_backward.h>
+#include <ATen/ops/embedding_renorm.h>
+#include <ATen/ops/embedding_sparse_backward.h>
+#include <ATen/ops/empty.h>
+#include <ATen/ops/empty_like.h>
+#include <ATen/ops/empty_permuted.h>
+#include <ATen/ops/empty_quantized.h>
+#include <ATen/ops/empty_strided.h>
+#include <ATen/ops/eq.h>
+#include <ATen/ops/equal.h>
+#include <ATen/ops/erf.h>
+#include <ATen/ops/erfc.h>
+#include <ATen/ops/erfinv.h>
+#include <ATen/ops/exp.h>
+#include <ATen/ops/exp2.h>
+#include <ATen/ops/expand.h>
+#include <ATen/ops/expand_as.h>
+#include <ATen/ops/expand_copy.h>
+#include <ATen/ops/expm1.h>
+#include <ATen/ops/exponential.h>
+#include <ATen/ops/eye.h>
+#include <ATen/ops/fake_quantize_per_channel_affine.h>
+#include <ATen/ops/fake_quantize_per_channel_affine_cachemask.h>
+#include <ATen/ops/fake_quantize_per_channel_affine_cachemask_backward.h>
+#include <ATen/ops/fake_quantize_per_tensor_affine.h>
+#include <ATen/ops/fake_quantize_per_tensor_affine_cachemask.h>
+#include <ATen/ops/fake_quantize_per_tensor_affine_cachemask_backward.h>
+#include <ATen/ops/fbgemm_linear_fp16_weight.h>
+#include <ATen/ops/fbgemm_linear_fp16_weight_fp32_activation.h>
+#include <ATen/ops/fbgemm_linear_int8_weight.h>
+#include <ATen/ops/fbgemm_linear_int8_weight_fp32_activation.h>
+#include <ATen/ops/fbgemm_linear_quantize_weight.h>
+#include <ATen/ops/fbgemm_pack_gemm_matrix_fp16.h>
+#include <ATen/ops/fbgemm_pack_quantized_matrix.h>
+#include <ATen/ops/feature_alpha_dropout.h>
+#include <ATen/ops/feature_dropout.h>
+#include <ATen/ops/fft_fft.h>
+#include <ATen/ops/fft_fft2.h>
+#include <ATen/ops/fft_fftfreq.h>
+#include <ATen/ops/fft_fftn.h>
+#include <ATen/ops/fft_fftshift.h>
+#include <ATen/ops/fft_hfft.h>
+#include <ATen/ops/fft_hfft2.h>
+#include <ATen/ops/fft_hfftn.h>
+#include <ATen/ops/fft_ifft.h>
+#include <ATen/ops/fft_ifft2.h>
+#include <ATen/ops/fft_ifftn.h>
+#include <ATen/ops/fft_ifftshift.h>
+#include <ATen/ops/fft_ihfft.h>
+#include <ATen/ops/fft_ihfft2.h>
+#include <ATen/ops/fft_ihfftn.h>
+#include <ATen/ops/fft_irfft.h>
+#include <ATen/ops/fft_irfft2.h>
+#include <ATen/ops/fft_irfftn.h>
+#include <ATen/ops/fft_rfft.h>
+#include <ATen/ops/fft_rfft2.h>
+#include <ATen/ops/fft_rfftfreq.h>
+#include <ATen/ops/fft_rfftn.h>
+#include <ATen/ops/fill.h>
+#include <ATen/ops/fill_diagonal.h>
+#include <ATen/ops/fix.h>
+#include <ATen/ops/flatten.h>
+#include <ATen/ops/flatten_dense_tensors.h>
+#include <ATen/ops/flip.h>
+#include <ATen/ops/fliplr.h>
+#include <ATen/ops/flipud.h>
+#include <ATen/ops/float_power.h>
+#include <ATen/ops/floor.h>
+#include <ATen/ops/floor_divide.h>
+#include <ATen/ops/fmax.h>
+#include <ATen/ops/fmin.h>
+#include <ATen/ops/fmod.h>
+#include <ATen/ops/frac.h>
+#include <ATen/ops/fractional_max_pool2d.h>
+#include <ATen/ops/fractional_max_pool2d_backward.h>
+#include <ATen/ops/fractional_max_pool3d.h>
+#include <ATen/ops/fractional_max_pool3d_backward.h>
+#include <ATen/ops/frexp.h>
+#include <ATen/ops/frobenius_norm.h>
+#include <ATen/ops/from_file.h>
+#include <ATen/ops/full.h>
+#include <ATen/ops/full_like.h>
+#include <ATen/ops/fused_moving_avg_obs_fake_quant.h>
+#include <ATen/ops/gather.h>
+#include <ATen/ops/gather_backward.h>
+#include <ATen/ops/gcd.h>
+#include <ATen/ops/ge.h>
+#include <ATen/ops/gelu.h>
+#include <ATen/ops/gelu_backward.h>
+#include <ATen/ops/geometric.h>
+#include <ATen/ops/geqrf.h>
+#include <ATen/ops/ger.h>
+#include <ATen/ops/glu.h>
+#include <ATen/ops/glu_backward.h>
+#include <ATen/ops/glu_backward_jvp.h>
+#include <ATen/ops/glu_jvp.h>
+#include <ATen/ops/gradient.h>
+#include <ATen/ops/greater.h>
+#include <ATen/ops/greater_equal.h>
+#include <ATen/ops/grid_sampler.h>
+#include <ATen/ops/grid_sampler_2d.h>
+#include <ATen/ops/grid_sampler_2d_backward.h>
+#include <ATen/ops/grid_sampler_3d.h>
+#include <ATen/ops/grid_sampler_3d_backward.h>
+#include <ATen/ops/group_norm.h>
+#include <ATen/ops/gru.h>
+#include <ATen/ops/gru_cell.h>
+#include <ATen/ops/gt.h>
+#include <ATen/ops/hamming_window.h>
+#include <ATen/ops/hann_window.h>
+#include <ATen/ops/hardshrink.h>
+#include <ATen/ops/hardshrink_backward.h>
+#include <ATen/ops/hardsigmoid.h>
+#include <ATen/ops/hardsigmoid_backward.h>
+#include <ATen/ops/hardswish.h>
+#include <ATen/ops/hardswish_backward.h>
+#include <ATen/ops/hardtanh.h>
+#include <ATen/ops/hardtanh_backward.h>
+#include <ATen/ops/heaviside.h>
+#include <ATen/ops/hinge_embedding_loss.h>
+#include <ATen/ops/histc.h>
+#include <ATen/ops/histogram.h>
+#include <ATen/ops/histogramdd.h>
+#include <ATen/ops/hsplit.h>
+#include <ATen/ops/hspmm.h>
+#include <ATen/ops/hstack.h>
+#include <ATen/ops/huber_loss.h>
+#include <ATen/ops/huber_loss_backward.h>
+#include <ATen/ops/hypot.h>
+#include <ATen/ops/i0.h>
+#include <ATen/ops/igamma.h>
+#include <ATen/ops/igammac.h>
+#include <ATen/ops/im2col.h>
+#include <ATen/ops/imag.h>
+#include <ATen/ops/index.h>
+#include <ATen/ops/index_add.h>
+#include <ATen/ops/index_copy.h>
+#include <ATen/ops/index_fill.h>
+#include <ATen/ops/index_put.h>
+#include <ATen/ops/index_reduce.h>
+#include <ATen/ops/index_select.h>
+#include <ATen/ops/index_select_backward.h>
+#include <ATen/ops/indices.h>
+#include <ATen/ops/indices_copy.h>
+#include <ATen/ops/infinitely_differentiable_gelu_backward.h>
+#include <ATen/ops/inner.h>
+#include <ATen/ops/instance_norm.h>
+#include <ATen/ops/int_repr.h>
+#include <ATen/ops/inverse.h>
+#include <ATen/ops/is_coalesced.h>
+#include <ATen/ops/is_complex.h>
+#include <ATen/ops/is_conj.h>
+#include <ATen/ops/is_distributed.h>
+#include <ATen/ops/is_floating_point.h>
+#include <ATen/ops/is_inference.h>
+#include <ATen/ops/is_leaf.h>
+#include <ATen/ops/is_neg.h>
+#include <ATen/ops/is_nonzero.h>
+#include <ATen/ops/is_pinned.h>
+#include <ATen/ops/is_same_size.h>
+#include <ATen/ops/is_set_to.h>
+#include <ATen/ops/is_signed.h>
+#include <ATen/ops/is_vulkan_available.h>
+#include <ATen/ops/isclose.h>
+#include <ATen/ops/isfinite.h>
+#include <ATen/ops/isin.h>
+#include <ATen/ops/isinf.h>
+#include <ATen/ops/isnan.h>
+#include <ATen/ops/isneginf.h>
+#include <ATen/ops/isposinf.h>
+#include <ATen/ops/isreal.h>
+#include <ATen/ops/istft.h>
+#include <ATen/ops/item.h>
+#include <ATen/ops/kaiser_window.h>
+#include <ATen/ops/kl_div.h>
+#include <ATen/ops/kron.h>
+#include <ATen/ops/kthvalue.h>
+#include <ATen/ops/l1_loss.h>
+#include <ATen/ops/layer_norm.h>
+#include <ATen/ops/lcm.h>
+#include <ATen/ops/ldexp.h>
+#include <ATen/ops/le.h>
+#include <ATen/ops/leaky_relu.h>
+#include <ATen/ops/leaky_relu_backward.h>
+#include <ATen/ops/lerp.h>
+#include <ATen/ops/less.h>
+#include <ATen/ops/less_equal.h>
+#include <ATen/ops/lgamma.h>
+#include <ATen/ops/lift.h>
+#include <ATen/ops/lift_fresh.h>
+#include <ATen/ops/lift_fresh_copy.h>
+#include <ATen/ops/linalg_cholesky.h>
+#include <ATen/ops/linalg_cholesky_ex.h>
+#include <ATen/ops/linalg_cond.h>
+#include <ATen/ops/linalg_cross.h>
+#include <ATen/ops/linalg_det.h>
+#include <ATen/ops/linalg_diagonal.h>
+#include <ATen/ops/linalg_eig.h>
+#include <ATen/ops/linalg_eigh.h>
+#include <ATen/ops/linalg_eigvals.h>
+#include <ATen/ops/linalg_eigvalsh.h>
+#include <ATen/ops/linalg_householder_product.h>
+#include <ATen/ops/linalg_inv.h>
+#include <ATen/ops/linalg_inv_ex.h>
+#include <ATen/ops/linalg_ldl_factor.h>
+#include <ATen/ops/linalg_ldl_factor_ex.h>
+#include <ATen/ops/linalg_ldl_solve.h>
+#include <ATen/ops/linalg_lstsq.h>
+#include <ATen/ops/linalg_lu.h>
+#include <ATen/ops/linalg_lu_factor.h>
+#include <ATen/ops/linalg_lu_factor_ex.h>
+#include <ATen/ops/linalg_lu_solve.h>
+#include <ATen/ops/linalg_matmul.h>
+#include <ATen/ops/linalg_matrix_exp.h>
+#include <ATen/ops/linalg_matrix_norm.h>
+#include <ATen/ops/linalg_matrix_power.h>
+#include <ATen/ops/linalg_matrix_rank.h>
+#include <ATen/ops/linalg_multi_dot.h>
+#include <ATen/ops/linalg_norm.h>
+#include <ATen/ops/linalg_pinv.h>
+#include <ATen/ops/linalg_qr.h>
+#include <ATen/ops/linalg_slogdet.h>
+#include <ATen/ops/linalg_solve.h>
+#include <ATen/ops/linalg_solve_ex.h>
+#include <ATen/ops/linalg_solve_triangular.h>
+#include <ATen/ops/linalg_svd.h>
+#include <ATen/ops/linalg_svdvals.h>
+#include <ATen/ops/linalg_tensorinv.h>
+#include <ATen/ops/linalg_tensorsolve.h>
+#include <ATen/ops/linalg_vander.h>
+#include <ATen/ops/linalg_vecdot.h>
+#include <ATen/ops/linalg_vector_norm.h>
+#include <ATen/ops/linear.h>
+#include <ATen/ops/linear_backward.h>
+#include <ATen/ops/linspace.h>
+#include <ATen/ops/log.h>
+#include <ATen/ops/log10.h>
+#include <ATen/ops/log1p.h>
+#include <ATen/ops/log2.h>
+#include <ATen/ops/log_normal.h>
+#include <ATen/ops/log_sigmoid.h>
+#include <ATen/ops/log_sigmoid_backward.h>
+#include <ATen/ops/log_sigmoid_forward.h>
+#include <ATen/ops/log_softmax.h>
+#include <ATen/ops/logaddexp.h>
+#include <ATen/ops/logaddexp2.h>
+#include <ATen/ops/logcumsumexp.h>
+#include <ATen/ops/logdet.h>
+#include <ATen/ops/logical_and.h>
+#include <ATen/ops/logical_not.h>
+#include <ATen/ops/logical_or.h>
+#include <ATen/ops/logical_xor.h>
+#include <ATen/ops/logit.h>
+#include <ATen/ops/logit_backward.h>
+#include <ATen/ops/logspace.h>
+#include <ATen/ops/logsumexp.h>
+#include <ATen/ops/lshift.h>
+#include <ATen/ops/lstm.h>
+#include <ATen/ops/lstm_cell.h>
+#include <ATen/ops/lstm_mps_backward.h>
+#include <ATen/ops/lt.h>
+#include <ATen/ops/lu_solve.h>
+#include <ATen/ops/lu_unpack.h>
+#include <ATen/ops/mH.h>
+#include <ATen/ops/mT.h>
+#include <ATen/ops/margin_ranking_loss.h>
+#include <ATen/ops/masked_fill.h>
+#include <ATen/ops/masked_scatter.h>
+#include <ATen/ops/masked_scatter_backward.h>
+#include <ATen/ops/masked_select.h>
+#include <ATen/ops/masked_select_backward.h>
+#include <ATen/ops/matmul.h>
+#include <ATen/ops/matmul_backward.h>
+#include <ATen/ops/matrix_H.h>
+#include <ATen/ops/matrix_exp.h>
+#include <ATen/ops/matrix_exp_backward.h>
+#include <ATen/ops/matrix_power.h>
+#include <ATen/ops/max.h>
+#include <ATen/ops/max_pool1d.h>
+#include <ATen/ops/max_pool1d_with_indices.h>
+#include <ATen/ops/max_pool2d.h>
+#include <ATen/ops/max_pool2d_backward.h>
+#include <ATen/ops/max_pool2d_with_indices.h>
+#include <ATen/ops/max_pool2d_with_indices_backward.h>
+#include <ATen/ops/max_pool3d.h>
+#include <ATen/ops/max_pool3d_with_indices.h>
+#include <ATen/ops/max_pool3d_with_indices_backward.h>
+#include <ATen/ops/max_unpool2d.h>
+#include <ATen/ops/max_unpool3d.h>
+#include <ATen/ops/maximum.h>
+#include <ATen/ops/mean.h>
+#include <ATen/ops/median.h>
+#include <ATen/ops/meshgrid.h>
+#include <ATen/ops/min.h>
+#include <ATen/ops/minimum.h>
+#include <ATen/ops/miopen_batch_norm.h>
+#include <ATen/ops/miopen_batch_norm_backward.h>
+#include <ATen/ops/miopen_convolution.h>
+#include <ATen/ops/miopen_convolution_add_relu.h>
+#include <ATen/ops/miopen_convolution_relu.h>
+#include <ATen/ops/miopen_convolution_transpose.h>
+#include <ATen/ops/miopen_depthwise_convolution.h>
+#include <ATen/ops/miopen_rnn.h>
+#include <ATen/ops/miopen_rnn_backward.h>
+#include <ATen/ops/mish.h>
+#include <ATen/ops/mish_backward.h>
+#include <ATen/ops/mkldnn_adaptive_avg_pool2d.h>
+#include <ATen/ops/mkldnn_adaptive_avg_pool2d_backward.h>
+#include <ATen/ops/mkldnn_convolution.h>
+#include <ATen/ops/mkldnn_linear.h>
+#include <ATen/ops/mkldnn_linear_backward.h>
+#include <ATen/ops/mkldnn_linear_backward_input.h>
+#include <ATen/ops/mkldnn_linear_backward_weights.h>
+#include <ATen/ops/mkldnn_max_pool2d.h>
+#include <ATen/ops/mkldnn_max_pool2d_backward.h>
+#include <ATen/ops/mkldnn_max_pool3d.h>
+#include <ATen/ops/mkldnn_max_pool3d_backward.h>
+#include <ATen/ops/mkldnn_reorder_conv2d_weight.h>
+#include <ATen/ops/mkldnn_reorder_conv3d_weight.h>
+#include <ATen/ops/mkldnn_rnn_layer.h>
+#include <ATen/ops/mkldnn_rnn_layer_backward.h>
+#include <ATen/ops/mm.h>
+#include <ATen/ops/mode.h>
+#include <ATen/ops/moveaxis.h>
+#include <ATen/ops/movedim.h>
+#include <ATen/ops/mps_convolution_backward.h>
+#include <ATen/ops/mps_convolution_transpose_backward.h>
+#include <ATen/ops/mse_loss.h>
+#include <ATen/ops/mse_loss_backward.h>
+#include <ATen/ops/msort.h>
+#include <ATen/ops/mul.h>
+#include <ATen/ops/multi_margin_loss.h>
+#include <ATen/ops/multi_margin_loss_backward.h>
+#include <ATen/ops/multilabel_margin_loss.h>
+#include <ATen/ops/multilabel_margin_loss_backward.h>
+#include <ATen/ops/multilabel_margin_loss_forward.h>
+#include <ATen/ops/multinomial.h>
+#include <ATen/ops/multiply.h>
+#include <ATen/ops/mv.h>
+#include <ATen/ops/mvlgamma.h>
+#include <ATen/ops/nan_to_num.h>
+#include <ATen/ops/nanmean.h>
+#include <ATen/ops/nanmedian.h>
+#include <ATen/ops/nanquantile.h>
+#include <ATen/ops/nansum.h>
+#include <ATen/ops/narrow.h>
+#include <ATen/ops/narrow_copy.h>
+#include <ATen/ops/native_batch_norm.h>
+#include <ATen/ops/native_batch_norm_backward.h>
+#include <ATen/ops/native_channel_shuffle.h>
+#include <ATen/ops/native_dropout.h>
+#include <ATen/ops/native_dropout_backward.h>
+#include <ATen/ops/native_group_norm.h>
+#include <ATen/ops/native_group_norm_backward.h>
+#include <ATen/ops/native_layer_norm.h>
+#include <ATen/ops/native_layer_norm_backward.h>
+#include <ATen/ops/native_norm.h>
+#include <ATen/ops/ne.h>
+#include <ATen/ops/neg.h>
+#include <ATen/ops/negative.h>
+#include <ATen/ops/nested_to_padded_tensor.h>
+#include <ATen/ops/new_empty.h>
+#include <ATen/ops/new_empty_strided.h>
+#include <ATen/ops/new_full.h>
+#include <ATen/ops/new_ones.h>
+#include <ATen/ops/new_zeros.h>
+#include <ATen/ops/nextafter.h>
+#include <ATen/ops/nll_loss.h>
+#include <ATen/ops/nll_loss2d.h>
+#include <ATen/ops/nll_loss2d_backward.h>
+#include <ATen/ops/nll_loss2d_forward.h>
+#include <ATen/ops/nll_loss_backward.h>
+#include <ATen/ops/nll_loss_forward.h>
+#include <ATen/ops/nll_loss_nd.h>
+#include <ATen/ops/nonzero.h>
+#include <ATen/ops/nonzero_numpy.h>
+#include <ATen/ops/nonzero_static.h>
+#include <ATen/ops/norm.h>
+#include <ATen/ops/norm_except_dim.h>
+#include <ATen/ops/normal.h>
+#include <ATen/ops/not_equal.h>
+#include <ATen/ops/nuclear_norm.h>
+#include <ATen/ops/numpy_T.h>
+#include <ATen/ops/one_hot.h>
+#include <ATen/ops/ones.h>
+#include <ATen/ops/ones_like.h>
+#include <ATen/ops/or.h>
+#include <ATen/ops/orgqr.h>
+#include <ATen/ops/ormqr.h>
+#include <ATen/ops/outer.h>
+#include <ATen/ops/output_nr.h>
+#include <ATen/ops/pad.h>
+#include <ATen/ops/pad_sequence.h>
+#include <ATen/ops/pairwise_distance.h>
+#include <ATen/ops/pdist.h>
+#include <ATen/ops/permute.h>
+#include <ATen/ops/permute_copy.h>
+#include <ATen/ops/pin_memory.h>
+#include <ATen/ops/pinverse.h>
+#include <ATen/ops/pixel_shuffle.h>
+#include <ATen/ops/pixel_unshuffle.h>
+#include <ATen/ops/poisson.h>
+#include <ATen/ops/poisson_nll_loss.h>
+#include <ATen/ops/polar.h>
+#include <ATen/ops/polygamma.h>
+#include <ATen/ops/positive.h>
+#include <ATen/ops/pow.h>
+#include <ATen/ops/prelu.h>
+#include <ATen/ops/prod.h>
+#include <ATen/ops/promote_types.h>
+#include <ATen/ops/put.h>
+#include <ATen/ops/q_per_channel_axis.h>
+#include <ATen/ops/q_per_channel_scales.h>
+#include <ATen/ops/q_per_channel_zero_points.h>
+#include <ATen/ops/q_scale.h>
+#include <ATen/ops/q_zero_point.h>
+#include <ATen/ops/qr.h>
+#include <ATen/ops/qscheme.h>
+#include <ATen/ops/quantile.h>
+#include <ATen/ops/quantize_per_channel.h>
+#include <ATen/ops/quantize_per_tensor.h>
+#include <ATen/ops/quantize_per_tensor_dynamic.h>
+#include <ATen/ops/quantized_batch_norm.h>
+#include <ATen/ops/quantized_gru_cell.h>
+#include <ATen/ops/quantized_lstm_cell.h>
+#include <ATen/ops/quantized_max_pool1d.h>
+#include <ATen/ops/quantized_max_pool2d.h>
+#include <ATen/ops/quantized_max_pool3d.h>
+#include <ATen/ops/quantized_rnn_relu_cell.h>
+#include <ATen/ops/quantized_rnn_tanh_cell.h>
+#include <ATen/ops/rad2deg.h>
+#include <ATen/ops/rand.h>
+#include <ATen/ops/rand_like.h>
+#include <ATen/ops/randint.h>
+#include <ATen/ops/randint_like.h>
+#include <ATen/ops/randn.h>
+#include <ATen/ops/randn_like.h>
+#include <ATen/ops/random.h>
+#include <ATen/ops/randperm.h>
+#include <ATen/ops/range.h>
+#include <ATen/ops/ravel.h>
+#include <ATen/ops/real.h>
+#include <ATen/ops/reciprocal.h>
+#include <ATen/ops/record_stream.h>
+#include <ATen/ops/refine_names.h>
+#include <ATen/ops/reflection_pad1d.h>
+#include <ATen/ops/reflection_pad1d_backward.h>
+#include <ATen/ops/reflection_pad2d.h>
+#include <ATen/ops/reflection_pad2d_backward.h>
+#include <ATen/ops/reflection_pad3d.h>
+#include <ATen/ops/reflection_pad3d_backward.h>
+#include <ATen/ops/relu.h>
+#include <ATen/ops/relu6.h>
+#include <ATen/ops/remainder.h>
+#include <ATen/ops/rename.h>
+#include <ATen/ops/renorm.h>
+#include <ATen/ops/repeat.h>
+#include <ATen/ops/repeat_interleave.h>
+#include <ATen/ops/replication_pad1d.h>
+#include <ATen/ops/replication_pad1d_backward.h>
+#include <ATen/ops/replication_pad2d.h>
+#include <ATen/ops/replication_pad2d_backward.h>
+#include <ATen/ops/replication_pad3d.h>
+#include <ATen/ops/replication_pad3d_backward.h>
+#include <ATen/ops/requires_grad.h>
+#include <ATen/ops/reshape.h>
+#include <ATen/ops/reshape_as.h>
+#include <ATen/ops/resize.h>
+#include <ATen/ops/resize_as.h>
+#include <ATen/ops/resize_as_sparse.h>
+#include <ATen/ops/resolve_conj.h>
+#include <ATen/ops/resolve_neg.h>
+#include <ATen/ops/result_type.h>
+#include <ATen/ops/retain_grad.h>
+#include <ATen/ops/retains_grad.h>
+#include <ATen/ops/rms_norm.h>
+#include <ATen/ops/rnn_relu.h>
+#include <ATen/ops/rnn_relu_cell.h>
+#include <ATen/ops/rnn_tanh.h>
+#include <ATen/ops/rnn_tanh_cell.h>
+#include <ATen/ops/roll.h>
+#include <ATen/ops/rot90.h>
+#include <ATen/ops/round.h>
+#include <ATen/ops/row_indices.h>
+#include <ATen/ops/row_indices_copy.h>
+#include <ATen/ops/row_stack.h>
+#include <ATen/ops/rrelu.h>
+#include <ATen/ops/rrelu_with_noise.h>
+#include <ATen/ops/rrelu_with_noise_backward.h>
+#include <ATen/ops/rshift.h>
+#include <ATen/ops/rsqrt.h>
+#include <ATen/ops/rsub.h>
+#include <ATen/ops/scalar_tensor.h>
+#include <ATen/ops/scaled_dot_product_attention.h>
+#include <ATen/ops/scatter.h>
+#include <ATen/ops/scatter_add.h>
+#include <ATen/ops/scatter_reduce.h>
+#include <ATen/ops/searchsorted.h>
+#include <ATen/ops/segment_reduce.h>
+#include <ATen/ops/select.h>
+#include <ATen/ops/select_backward.h>
+#include <ATen/ops/select_copy.h>
+#include <ATen/ops/select_scatter.h>
+#include <ATen/ops/selu.h>
+#include <ATen/ops/set.h>
+#include <ATen/ops/set_data.h>
+#include <ATen/ops/sgn.h>
+#include <ATen/ops/sigmoid.h>
+#include <ATen/ops/sigmoid_backward.h>
+#include <ATen/ops/sign.h>
+#include <ATen/ops/signbit.h>
+#include <ATen/ops/silu.h>
+#include <ATen/ops/silu_backward.h>
+#include <ATen/ops/sin.h>
+#include <ATen/ops/sinc.h>
+#include <ATen/ops/sinh.h>
+#include <ATen/ops/size.h>
+#include <ATen/ops/slice.h>
+#include <ATen/ops/slice_backward.h>
+#include <ATen/ops/slice_copy.h>
+#include <ATen/ops/slice_inverse.h>
+#include <ATen/ops/slice_scatter.h>
+#include <ATen/ops/slogdet.h>
+#include <ATen/ops/slow_conv3d.h>
+#include <ATen/ops/slow_conv3d_forward.h>
+#include <ATen/ops/slow_conv_dilated2d.h>
+#include <ATen/ops/slow_conv_dilated3d.h>
+#include <ATen/ops/slow_conv_transpose2d.h>
+#include <ATen/ops/slow_conv_transpose3d.h>
+#include <ATen/ops/smm.h>
+#include <ATen/ops/smooth_l1_loss.h>
+#include <ATen/ops/smooth_l1_loss_backward.h>
+#include <ATen/ops/soft_margin_loss.h>
+#include <ATen/ops/soft_margin_loss_backward.h>
+#include <ATen/ops/softmax.h>
+#include <ATen/ops/softplus.h>
+#include <ATen/ops/softplus_backward.h>
+#include <ATen/ops/softshrink.h>
+#include <ATen/ops/softshrink_backward.h>
+#include <ATen/ops/sort.h>
+#include <ATen/ops/sparse_bsc_tensor.h>
+#include <ATen/ops/sparse_bsr_tensor.h>
+#include <ATen/ops/sparse_compressed_tensor.h>
+#include <ATen/ops/sparse_coo_tensor.h>
+#include <ATen/ops/sparse_csc_tensor.h>
+#include <ATen/ops/sparse_csr_tensor.h>
+#include <ATen/ops/sparse_dim.h>
+#include <ATen/ops/sparse_mask.h>
+#include <ATen/ops/sparse_resize.h>
+#include <ATen/ops/sparse_resize_and_clear.h>
+#include <ATen/ops/sparse_sampled_addmm.h>
+#include <ATen/ops/special_airy_ai.h>
+#include <ATen/ops/special_bessel_j0.h>
+#include <ATen/ops/special_bessel_j1.h>
+#include <ATen/ops/special_bessel_y0.h>
+#include <ATen/ops/special_bessel_y1.h>
+#include <ATen/ops/special_chebyshev_polynomial_t.h>
+#include <ATen/ops/special_chebyshev_polynomial_u.h>
+#include <ATen/ops/special_chebyshev_polynomial_v.h>
+#include <ATen/ops/special_chebyshev_polynomial_w.h>
+#include <ATen/ops/special_digamma.h>
+#include <ATen/ops/special_entr.h>
+#include <ATen/ops/special_erf.h>
+#include <ATen/ops/special_erfc.h>
+#include <ATen/ops/special_erfcx.h>
+#include <ATen/ops/special_erfinv.h>
+#include <ATen/ops/special_exp2.h>
+#include <ATen/ops/special_expit.h>
+#include <ATen/ops/special_expm1.h>
+#include <ATen/ops/special_gammainc.h>
+#include <ATen/ops/special_gammaincc.h>
+#include <ATen/ops/special_gammaln.h>
+#include <ATen/ops/special_hermite_polynomial_h.h>
+#include <ATen/ops/special_hermite_polynomial_he.h>
+#include <ATen/ops/special_i0.h>
+#include <ATen/ops/special_i0e.h>
+#include <ATen/ops/special_i1.h>
+#include <ATen/ops/special_i1e.h>
+#include <ATen/ops/special_laguerre_polynomial_l.h>
+#include <ATen/ops/special_legendre_polynomial_p.h>
+#include <ATen/ops/special_log1p.h>
+#include <ATen/ops/special_log_ndtr.h>
+#include <ATen/ops/special_log_softmax.h>
+#include <ATen/ops/special_logit.h>
+#include <ATen/ops/special_logsumexp.h>
+#include <ATen/ops/special_modified_bessel_i0.h>
+#include <ATen/ops/special_modified_bessel_i1.h>
+#include <ATen/ops/special_modified_bessel_k0.h>
+#include <ATen/ops/special_modified_bessel_k1.h>
+#include <ATen/ops/special_multigammaln.h>
+#include <ATen/ops/special_ndtr.h>
+#include <ATen/ops/special_ndtri.h>
+#include <ATen/ops/special_polygamma.h>
+#include <ATen/ops/special_psi.h>
+#include <ATen/ops/special_round.h>
+#include <ATen/ops/special_scaled_modified_bessel_k0.h>
+#include <ATen/ops/special_scaled_modified_bessel_k1.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_t.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_u.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_v.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_w.h>
+#include <ATen/ops/special_sinc.h>
+#include <ATen/ops/special_softmax.h>
+#include <ATen/ops/special_spherical_bessel_j0.h>
+#include <ATen/ops/special_xlog1py.h>
+#include <ATen/ops/special_xlogy.h>
+#include <ATen/ops/special_zeta.h>
+#include <ATen/ops/split.h>
+#include <ATen/ops/split_copy.h>
+#include <ATen/ops/split_with_sizes.h>
+#include <ATen/ops/split_with_sizes_copy.h>
+#include <ATen/ops/sqrt.h>
+#include <ATen/ops/square.h>
+#include <ATen/ops/squeeze.h>
+#include <ATen/ops/squeeze_copy.h>
+#include <ATen/ops/sspaddmm.h>
+#include <ATen/ops/stack.h>
+#include <ATen/ops/std.h>
+#include <ATen/ops/std_mean.h>
+#include <ATen/ops/stft.h>
+#include <ATen/ops/stride.h>
+#include <ATen/ops/sub.h>
+#include <ATen/ops/subtract.h>
+#include <ATen/ops/sum.h>
+#include <ATen/ops/sum_to_size.h>
+#include <ATen/ops/svd.h>
+#include <ATen/ops/swapaxes.h>
+#include <ATen/ops/swapdims.h>
+#include <ATen/ops/sym_constrain_range.h>
+#include <ATen/ops/sym_constrain_range_for_size.h>
+#include <ATen/ops/sym_numel.h>
+#include <ATen/ops/sym_size.h>
+#include <ATen/ops/sym_storage_offset.h>
+#include <ATen/ops/sym_stride.h>
+#include <ATen/ops/t.h>
+#include <ATen/ops/t_copy.h>
+#include <ATen/ops/take.h>
+#include <ATen/ops/take_along_dim.h>
+#include <ATen/ops/tan.h>
+#include <ATen/ops/tanh.h>
+#include <ATen/ops/tanh_backward.h>
+#include <ATen/ops/tensor_split.h>
+#include <ATen/ops/tensordot.h>
+#include <ATen/ops/thnn_conv2d.h>
+#include <ATen/ops/threshold.h>
+#include <ATen/ops/threshold_backward.h>
+#include <ATen/ops/tile.h>
+#include <ATen/ops/to.h>
+#include <ATen/ops/to_dense.h>
+#include <ATen/ops/to_dense_backward.h>
+#include <ATen/ops/to_mkldnn.h>
+#include <ATen/ops/to_mkldnn_backward.h>
+#include <ATen/ops/to_padded_tensor.h>
+#include <ATen/ops/to_sparse.h>
+#include <ATen/ops/to_sparse_bsc.h>
+#include <ATen/ops/to_sparse_bsr.h>
+#include <ATen/ops/to_sparse_csc.h>
+#include <ATen/ops/to_sparse_csr.h>
+#include <ATen/ops/topk.h>
+#include <ATen/ops/trace.h>
+#include <ATen/ops/trace_backward.h>
+#include <ATen/ops/transpose.h>
+#include <ATen/ops/transpose_copy.h>
+#include <ATen/ops/trapezoid.h>
+#include <ATen/ops/trapz.h>
+#include <ATen/ops/triangular_solve.h>
+#include <ATen/ops/tril.h>
+#include <ATen/ops/tril_indices.h>
+#include <ATen/ops/triplet_margin_loss.h>
+#include <ATen/ops/triu.h>
+#include <ATen/ops/triu_indices.h>
+#include <ATen/ops/true_divide.h>
+#include <ATen/ops/trunc.h>
+#include <ATen/ops/type_as.h>
+#include <ATen/ops/unbind.h>
+#include <ATen/ops/unbind_copy.h>
+#include <ATen/ops/unflatten.h>
+#include <ATen/ops/unflatten_dense_tensors.h>
+#include <ATen/ops/unfold.h>
+#include <ATen/ops/unfold_backward.h>
+#include <ATen/ops/unfold_copy.h>
+#include <ATen/ops/uniform.h>
+#include <ATen/ops/unique_consecutive.h>
+#include <ATen/ops/unique_dim.h>
+#include <ATen/ops/unique_dim_consecutive.h>
+#include <ATen/ops/unsafe_chunk.h>
+#include <ATen/ops/unsafe_split.h>
+#include <ATen/ops/unsafe_split_with_sizes.h>
+#include <ATen/ops/unsqueeze.h>
+#include <ATen/ops/unsqueeze_copy.h>
+#include <ATen/ops/upsample_bicubic2d.h>
+#include <ATen/ops/upsample_bicubic2d_backward.h>
+#include <ATen/ops/upsample_bilinear2d.h>
+#include <ATen/ops/upsample_bilinear2d_backward.h>
+#include <ATen/ops/upsample_linear1d.h>
+#include <ATen/ops/upsample_linear1d_backward.h>
+#include <ATen/ops/upsample_nearest1d.h>
+#include <ATen/ops/upsample_nearest1d_backward.h>
+#include <ATen/ops/upsample_nearest2d.h>
+#include <ATen/ops/upsample_nearest2d_backward.h>
+#include <ATen/ops/upsample_nearest3d.h>
+#include <ATen/ops/upsample_nearest3d_backward.h>
+#include <ATen/ops/upsample_trilinear3d.h>
+#include <ATen/ops/upsample_trilinear3d_backward.h>
+#include <ATen/ops/value_selecting_reduction_backward.h>
+#include <ATen/ops/values.h>
+#include <ATen/ops/values_copy.h>
+#include <ATen/ops/vander.h>
+#include <ATen/ops/var.h>
+#include <ATen/ops/var_mean.h>
+#include <ATen/ops/vdot.h>
+#include <ATen/ops/view.h>
+#include <ATen/ops/view_as.h>
+#include <ATen/ops/view_as_complex.h>
+#include <ATen/ops/view_as_complex_copy.h>
+#include <ATen/ops/view_as_real.h>
+#include <ATen/ops/view_as_real_copy.h>
+#include <ATen/ops/view_copy.h>
+#include <ATen/ops/vsplit.h>
+#include <ATen/ops/vstack.h>
+#include <ATen/ops/where.h>
+#include <ATen/ops/xlogy.h>
+#include <ATen/ops/xor.h>
+#include <ATen/ops/zero.h>
+#include <ATen/ops/zeros.h>
+#include <ATen/ops/zeros_like.h>
+
+namespace at {
+
+
+
+// Special C++ only overloads for std()-like functions (See gh-40287)
+// These are needed because int -> bool conversion takes precedence over int -> IntArrayRef
+// So, for example std(0) would select the std(unbiased=False) overload
+TORCH_API inline Tensor var(const Tensor& self, int dim) {
+  return at::var(self, IntArrayRef{dim});
+}
+TORCH_API inline std::tuple<Tensor, Tensor> var_mean(const Tensor& self, int dim) {
+  return at::var_mean(self, IntArrayRef{dim});
+}
+TORCH_API inline Tensor std(const Tensor& self, int dim) {
+  return at::std(self, IntArrayRef{dim});
+}
+TORCH_API inline std::tuple<Tensor, Tensor> std_mean(const Tensor& self, int dim) {
+  return at::std_mean(self, IntArrayRef{dim});
+}
+
+inline int64_t numel(const Tensor& tensor) {
+  return tensor.numel();
+}
+
+inline int64_t size(const Tensor& tensor, int64_t dim) {
+  return tensor.size(dim);
+}
+
+inline int64_t stride(const Tensor& tensor, int64_t dim) {
+  return tensor.stride(dim);
+}
+
+inline bool is_complex(const Tensor& tensor) {
+  return tensor.is_complex();
+}
+
+inline bool is_floating_point(const Tensor& tensor) {
+  return tensor.is_floating_point();
+}
+
+inline bool is_signed(const Tensor& tensor) {
+  return tensor.is_signed();
+}
+
+inline bool is_inference(const Tensor& tensor) {
+  return tensor.is_inference();
+}
+
+inline bool _is_zerotensor(const Tensor& tensor) {
+  return tensor._is_zerotensor();
+}
+
+inline bool is_conj(const Tensor& tensor) {
+  return tensor.is_conj();
+}
+
+inline Tensor conj(const Tensor& tensor) {
+  return tensor.conj();
+}
+
+inline bool is_neg(const Tensor& tensor) {
+  return tensor.is_neg();
+}
+
+}

lib/python3.12/site-packages/torch/include/ATen/InferSize.h

@@ -0,0 +1,88 @@
+#pragma once
+
+#include <ATen/DimVector.h>
+#include <c10/core/ScalarType.h>
+#include <c10/core/SymIntArrayRef.h>
+#include <c10/util/DimVector.h>
+#include <optional>
+#include <sstream>
+#include <vector>
+
+namespace at {
+
+// Infers the size of a dim with size -1, if it exists. Also checks that new
+// shape is compatible with the number of elements.
+//
+// templated to handle std::vector<int64_t> and DimVector use cases, see
+// below
+//
+template <typename InputArrayRef, typename NumelType, typename ResultVec>
+inline void infer_size_impl(
+    InputArrayRef shape,
+    NumelType numel,
+    ResultVec& res) {
+  NumelType newsize = 1;
+  // N.B. this is an index, not a sym dim!
+  std::optional<int64_t> infer_dim;
+  for (int64_t dim = 0, ndim = shape.size(); dim != ndim; dim++) {
+    if (shape[dim] == -1) {
+      if (infer_dim) {
+        throw std::runtime_error("only one dimension can be inferred");
+      }
+      infer_dim = dim;
+    } else if (shape[dim] >= 0) {
+      newsize *= shape[dim];
+    } else {
+      TORCH_CHECK(false, "invalid shape dimension ", shape[dim]);
+    }
+  }
+
+  if (TORCH_GUARD_SIZE_OBLIVIOUS(sym_eq(numel, newsize)) ||
+      (infer_dim && newsize > 0 && numel % newsize == 0)) {
+    if (infer_dim) {
+      // We have a degree of freedom here to select the dimension size; follow
+      // NumPy semantics and just bail.  However, a nice error message is needed
+      // because users often use `view` as a way to flatten & unflatten
+      // dimensions and will otherwise be confused why
+      //   empty_tensor.view( 0, 0)
+      // works yet
+      //   empty_tensor.view(-1, 0)
+      // doesn't.
+      TORCH_CHECK(
+          newsize != 0,
+          "cannot reshape tensor of 0 elements into shape ",
+          shape,
+          " because the unspecified dimension size -1 can be any "
+          "value and is ambiguous");
+      res[*infer_dim] = numel / newsize;
+    }
+    return;
+  }
+
+  std::ostringstream ss;
+  ss << "shape '" << shape << "' is invalid for input of size " << numel;
+  throw std::runtime_error(ss.str());
+}
+
+inline std::vector<int64_t> infer_size(IntArrayRef shape, int64_t numel) {
+  auto res = shape.vec();
+  infer_size_impl(shape, numel, res);
+  return res;
+}
+
+inline at::DimVector infer_size_dv(IntArrayRef shape, int64_t numel) {
+  auto res = at::DimVector(shape);
+  infer_size_impl(shape, numel, res);
+  return res;
+}
+
+inline at::SymDimVector infer_size_dv(
+    c10::SymIntArrayRef shape,
+    c10::SymInt numel) {
+  auto res = at::SymDimVector(shape);
+  infer_size_impl<c10::SymIntArrayRef, c10::SymInt, at::SymDimVector>(
+      shape, std::move(numel), res);
+  return res;
+}
+
+} // namespace at

lib/python3.12/site-packages/torch/include/ATen/InitialTensorOptions.h

@@ -0,0 +1,15 @@
+#pragma once
+
+#include <c10/core/TensorOptions.h>
+
+namespace at {
+
+// Represents the initial TensorOptions, before the "defaults" are ever changed.
+// This is designed to be used in library code, where the explicit devices,
+// dtypes, etc. are known. NOTE: this is not a stable API.
+inline TensorOptions initialTensorOptions() {
+  return TensorOptions(kCPU).dtype(kFloat).layout(kStrided).requires_grad(
+      false);
+}
+
+} // namespace at

lib/python3.12/site-packages/torch/include/ATen/Layout.h

@@ -0,0 +1,2 @@
+#pragma once
+#include <c10/core/Layout.h>

lib/python3.12/site-packages/torch/include/ATen/LegacyBatchedTensorImpl.h

@@ -0,0 +1,160 @@
+#pragma once
+
+#include <bitset>
+
+#include <ATen/ArrayRef.h>
+#include <ATen/SmallVector.h>
+#include <ATen/Tensor.h>
+
+namespace at {
+
+// We assume this in a few other places in the codebase,
+// but there isn't a centralized definition.
+constexpr int64_t kVmapMaxTensorDims = 64;
+
+// The valid vmap levels range from [0, 64). This effectively means that we
+// support a maximum of 64 nested vmaps.
+constexpr int64_t kVmapNumLevels = 64;
+
+// Store this number of elements of BatchDims on the stack. Most people will
+// probably use <= 5 nested vmaps, but adjust this number as necessary.
+constexpr int64_t kBatchDimsStackSize = 5;
+
+// a BatchDim represents a "private" dimension on a Tensor created inside of
+// vmap. It is a (level, dim) tuple, with the `dim` indicating which dimension
+// is being vmap'ed over and the `level` being an identifier for which vmap
+// said dimension was created inside. The `dim` corresponds to a "physical
+// dim" - it is a dimension index on the underlying physical tensor that is
+// being vmapped over.
+struct BatchDim {
+  BatchDim(int64_t level, int64_t dim) : dim_(dim), level_(level) {}
+  int64_t dim() const {
+    return dim_;
+  }
+  int64_t level() const {
+    return level_;
+  }
+
+ private:
+  int64_t dim_;
+  int64_t level_;
+};
+
+using BatchDims = SmallVector<BatchDim, kBatchDimsStackSize>;
+using BatchDimsRef = ArrayRef<BatchDim>;
+
+// A BatchedTensorImpl holds an underlying Tensor and a list of BatchDim
+// NB: We use the term "BatchedTensor" to mean a Tensor that is backed with a
+// BatchedTensorImpl.
+//
+// The batch dimensions are treated as being "private"; they are not
+// user-visible. For example, in the following Tensor,
+//    bt = BatchedTensorImpl(ones(2, 3, 5, 7), [(lvl=1, dim=0), (lvl=2, dim=1)])
+// dimensions 0 and 1 are batch dimensions.
+//
+// bt.sizes() returns (5, 7); bt.sum(0) performs a reduction over the (public)
+// dim 0, which is equivalent to dim 3 in the underlying ones(2, 3, 5, 7)
+// tensor.
+struct TORCH_API BatchedTensorImpl : public c10::TensorImpl {
+  explicit BatchedTensorImpl(Tensor value, BatchDims bdims);
+
+  // Returns a reference to BatchDims that represent which dimensions of this
+  // tensor are private.
+  BatchDimsRef bdims() const {
+    return bdims_;
+  }
+
+  // BatchedTensorImpl wraps a Tensor
+  const Tensor& value() const {
+    return value_;
+  }
+
+  // Given a public dimension index, return the dimension index in the
+  // underlying value() tensor. For example, if we have
+  //    bt = BatchedTensorImpl(ones(2, 3, 5, 7), [(lvl=1, dim=0), (lvl=2,
+  //    dim=2)])
+  // bt.actualDim(0) -> 1
+  // bt.actualDim(1) -> 3
+  // bt.actualDim(2) -> Error
+  int64_t actualDim(int64_t dim, bool wrap_dim = true) const;
+
+  // We have to override this because we opted into CustomStrides
+  IntArrayRef strides_custom() const override;
+  // Override a bunch of methods inherited from TensorImpl to return error
+  // messages.
+  bool is_contiguous_custom(at::MemoryFormat memory_format) const override;
+  void set_size(int64_t dim, int64_t new_size) override;
+  void set_stride(int64_t dim, int64_t new_stride) override;
+  void set_storage_offset(int64_t storage_offset) override;
+#ifdef DEBUG
+  bool has_storage() const override;
+#endif
+
+ private:
+  // see NOTE: [BatchedTensorImpl levels invariant]
+  void checkInvariants() const;
+  const char* tensorimpl_type_name() const override;
+
+  Tensor value_;
+
+  // Note: [BatchedTensorImpl levels invariant]
+  // There is an invariant that the BatchDims must be stored in increasing
+  // `level` order. That is, for i < j, bdims_[i].level must be less than
+  // bdims_[j].level.
+  BatchDims bdims_;
+};
+
+// NB: We use the term "BatchedTensor" to mean a Tensor that is backed with a
+// BatchedTensorImpl.
+inline bool isBatchedTensor(const Tensor& tensor) {
+  return tensor.unsafeGetTensorImpl()->key_set().has(DispatchKey::Batched);
+}
+
+// It is unsafe to call this on a Tensor that is not backed by a
+// BatchedTensorImpl. Please use `maybeGetBatchedImpl` whenever possible.
+inline BatchedTensorImpl* unsafeGetBatchedImpl(const Tensor& tensor) {
+  return static_cast<BatchedTensorImpl*>(tensor.unsafeGetTensorImpl());
+}
+
+inline BatchedTensorImpl* maybeGetBatchedImpl(const Tensor& tensor) {
+  if (!isBatchedTensor(tensor)) {
+    return nullptr;
+  }
+  return unsafeGetBatchedImpl(tensor);
+}
+
+// Returns a bitset. If bit i is set, then that means dim i is a batchdim.
+inline std::bitset<kVmapMaxTensorDims> createBatchDimBitset(
+    BatchDimsRef bdims) {
+  std::bitset<kVmapMaxTensorDims> is_bdim;
+  for (const auto& bdim : bdims) {
+    is_bdim.set(bdim.dim());
+  }
+  return is_bdim;
+}
+
+// Creates a bitset for all of the levels present in `bdims`
+inline std::bitset<kVmapNumLevels> createVmapLevelsBitset(BatchDimsRef bdims) {
+  std::bitset<kVmapNumLevels> result;
+  for (const auto& bdim : bdims) {
+    result.set(bdim.level());
+  }
+  return result;
+}
+
+inline std::ostream& operator<<(std::ostream& out, const BatchDim& bdim) {
+  out << "(lvl=" << bdim.level() << ", dim=" << bdim.dim() << ")";
+  return out;
+}
+
+// Use this to construct a BatchedTensor from a regular Tensor
+TORCH_API Tensor makeBatched(Tensor tensor, BatchDims bdims);
+
+// Adds a batch dim to `tensor`, returning a BatchedTensor
+TORCH_API Tensor addBatchDim(Tensor tensor, int64_t level, int64_t dim);
+
+// Checks if an inplace operation on self and other is "vmap compatible".
+// See NOTE: [vmap-incompatible in-place operations] for the definition of this.
+TORCH_API bool inplaceIsVmapCompatible(const Tensor& self, const Tensor& other);
+
+} // namespace at

lib/python3.12/site-packages/torch/include/ATen/LegacyVmapMode.h

@@ -0,0 +1,26 @@
+#pragma once
+
+#include <c10/core/impl/LocalDispatchKeySet.h>
+
+namespace at::impl {
+
+// VmapMode contains a thread local count of how many nested vmaps
+// we are currently inside. That number is known as the `vmap level`.
+// VmapMode is used in the implementation of the Python `torch.vmap` API.
+//
+// NOTE: this is NOT the c++ api for torch.vmap. That doesn't exist yet.
+
+struct TORCH_API VmapMode {
+  // Returns the vmap level, aka the count of how many nested vmaps we're in.
+  static int64_t current_vmap_level();
+
+  // Increment the count of nested vmaps. If this causes the vmap level to be
+  // greater than 0, then it enables DispatchKey::VmapMode on all tensors.
+  static int64_t increment_nesting();
+
+  // Decrements the count of nested vmaps. If this causes the vmap level to be
+  // equal to 0, then it disables DispatchKey::VmapMode on all tensors.
+  static int64_t decrement_nesting();
+};
+
+} // namespace at::impl

lib/python3.12/site-packages/torch/include/ATen/LegacyVmapTransforms.h

@@ -0,0 +1,183 @@
+#pragma once
+
+#include <ATen/LegacyBatchedTensorImpl.h>
+#include <ATen/core/IListRef.h>
+
+namespace at {
+
+// This file contains abstractions used for transforming *logical* vmap
+// arguments into *physical* arguments. (Keep reading for definitions of these
+// terms).
+
+// NOTE: [Logical vs physical args]
+// Consider the following vmap.
+//   vmap(vmap(func, in_dims=(2,)), in_dims=(0,))(torch.ones(2, 3, 4))
+// This would produce a BatchedTensor wrapping a Tensor of size [2, 3, 4],
+// with batch dims 0 and 2:
+//   BatchedTensor(ones(2, 3, 4), bdims=[(lvl=1,dim=0),(lvl=2,dim=2)])
+//
+// We say the *logical* view of the tensor has size [3] -- tensors inside
+// `func` appear to have size [3].
+// However, the *physical* underlying tensor (the one passed to vmap) has size
+// [2, 3, 4].
+//
+// This notion of logical vs physical also extends to non-tensor arguments.
+// Consider the previous tensor; let's assume the user called
+// `torch.sum(tensor, dim=0)` inside of `func`. Then the logical
+// dimension they are reducing over is dim 0 but the physical dim is dim 1
+// (the first non-batch dimension)
+
+// Forward declared; see NOTE: [What is a VmapPhysicalView?]
+struct VmapPhysicalView;
+
+// Most PyTorch operators take 4 or fewer inputs.
+constexpr int64_t kVmapTransformStaticInputSize = 4;
+using VmapPhysicalViewVec =
+    SmallVector<VmapPhysicalView, kVmapTransformStaticInputSize>;
+
+// Pytorch generally advertises good performance for <= 5 dims.
+// (see ATen/core/DimVector.h). We add a few extra dims (~3) for vmap
+// dimensions to get 8. Adjust this number as necessary
+constexpr int64_t kVmapStaticDimVecSize = 8;
+using VmapDimVector = SmallVector<int64_t, kVmapStaticDimVecSize>;
+using VmapSymDimVector = SmallVector<c10::SymInt, kVmapStaticDimVecSize>;
+
+// NOTE: [What is an VmapTransform?]
+// An *VmapTransform* converts logical views of tensors to physical views.
+//
+// Batching rules use VmapTransforms to convert logical arguments to
+// physical arguments, then call one or more at:: operator that handles the
+// physical arguments, and then converts the physical result back to a logical
+// argument.
+
+// VmapTransform for operators that take tensors with multiple batch dims.
+// Given one or more logical views on Tensors, `logicalToPhysical`
+// permutes all of the batch dims to the front of the tensor, aligns
+// and expands the batch dims to match each other (according to their `level`),
+// and returns a VmapPhysicalView on the tensor(s).
+struct TORCH_API MultiBatchVmapTransform {
+  static VmapPhysicalView logicalToPhysical(const Tensor& logical_tensor);
+  static VmapPhysicalViewVec logicalToPhysical(ITensorListRef logical_tensors);
+};
+
+// VmapTransform for operators that broadcast all inputs.
+// Given some logical views on Tensors, `logicalToPhysical`:
+// - permutes all of the batch dims to the front of the tensors
+// - aligns all the batch dims to the collective levels of all of the tensors.
+//   If a tensor does not have a batch dim for a vmap level, then it receives
+//   a size-one dimension for said level.
+// - aligns the non-batch dims to have the same dimensionality, adding extra
+//   size-1 dimensions in between the batch dimensions and the non-batch
+//   dimensions so that the batch dimensions are lined up from the right.
+//
+// For example: given inputs of size (B, 2) and (B, 3, 2) where B is the batch
+// dimension, BroadcastingVmapTransform returns VmapPhysicalViews that wrap
+// tensors of size (B, 1, 2) and (B, 3, 2).
+//
+// Given inputs of size (B, 2) and (2,), BroadcastingVmapTransform returns
+// VmapPhysicalViews wrapping tensors of size (B, 2) and (1, 2). We don't
+// actually *need* to return a tensor of size (1, 2) for the second tensor
+// because the broadcasting operation takes care of that for us, but we do
+// it anyways to keep things simple.
+struct TORCH_API BroadcastingVmapTransform {
+  static VmapPhysicalViewVec logicalToPhysical(TensorList logical_tensors);
+};
+
+// Forward declared, if you're reading this file head to toe, don't worry about
+// it yet.
+struct VmapPhysicalToLogicalMap;
+
+// NOTE: [What is a VmapPhysicalView?]
+// VmapPhysicalView represents a physical view on a Tensor.
+//
+// One can use it to further convert logical dimension indices, logical shapes,
+// and more to their physical variants, or convert a new (physical) tensor into
+// a logical BatchedTensor. (TODO(rzou): some of these are not yet implemented).
+//
+// VmapPhysicalView stores a physical tensor with all of its batch dimensions at
+// the front and some levels that correspond to said batch dimensions.
+//
+// The levels bitset specifies which vmap levels correspond to the batch
+// dimensions at the front of the tensor. In particular, the number of set bits
+// corresponds to the number of batch dimensions on `tensor` and the rightmost
+// bit of `levels` specifies the maximum number of nested vmaps we are in at
+// this point in time.
+// For example, given:
+//   physical_view = VmapPhysicalView(tensor=ones(2, 3, 4, 5, 6), levels={1, 3})
+//
+// Rightmost bit of `levels` is 3 indicating the number of nested vmaps less
+// than or equal to 3.
+//   bitset: 010100
+//              ^
+//              |
+//   levels: 012345
+struct TORCH_API VmapPhysicalView {
+  VmapPhysicalView(Tensor&& tensor, std::bitset<kVmapNumLevels> levels)
+      : levels_(levels), tensor_(std::move(tensor)) {
+    TORCH_INTERNAL_ASSERT(!isBatchedTensor(tensor_));
+  }
+
+  Tensor& tensor() {
+    return tensor_;
+  }
+  const Tensor& tensor() const {
+    return tensor_;
+  }
+
+  // Maps logical dim indices to physical dim indices. Also does dim wrapping.
+  //
+  // For example, given:
+  //   physical_view = VmapPhysicalView(tensor=ones(2, 3, 4, 5), levels={1, 3})
+  //
+  // Then physical_view.getPhysicalDims({0, 1}) returns {2, 3}.
+  // This is because the size of levels tell us that the first two dimensions
+  // of `tensor_` are batch dimensions, so a logical dim of `n` is actually
+  // a physical dim of `n + 2`.
+  VmapDimVector getPhysicalDims(OptionalIntArrayRef logical_dims) const;
+  int64_t getPhysicalDim(int64_t logical_dim) const;
+
+  // Returns a VmapPhysicalToLogicalMap object. This can be used for
+  // mapping a physical tensor to a new logical tensor (BatchedTensor)
+  VmapPhysicalToLogicalMap getPhysicalToLogicalMap() const;
+
+  // Maps a logical shape to a physical shape by pre-pending the batch
+  // sizes to the logical shape.
+  VmapDimVector getPhysicalShape(IntArrayRef logical_shape) const;
+
+  int64_t numBatchDims() const;
+
+ private:
+  int64_t numLogicalDims() const;
+
+  std::bitset<kVmapNumLevels> levels_;
+  Tensor tensor_;
+};
+
+// Convenience struct used for mapping a physical tensor (a non-BatchedTensor)
+// to a logical one (BatchedTensor). It holds some levels that are used to do
+// the mapping and assumes that the batch dimensions in the physical tensor all
+// occur at the front of the tensor.
+struct TORCH_API VmapPhysicalToLogicalMap {
+  VmapPhysicalToLogicalMap(std::bitset<kVmapNumLevels> levels)
+      : levels_(levels) {}
+
+  // Maps a physical tensor to a new logical tensor (BatchedTensor).
+  // Assumes that all of the "batch dimensions" are at the front
+  // of the physical tensor. For example, given:
+  // - x = rank-4 Tensor with size 2, 3, 5, 7
+  // - levels = (2, 4)
+  // Returns:
+  // - BatchedTensor(x, bdims=[(dim=0,lvl=2), (dim=1, lvl=4)])
+  Tensor apply(const Tensor& physical_tensor) const;
+
+  // Given a vector of physical tensors,
+  // 1. maps each tensor to a new logical tensor. Assumes that all of the
+  //    "batch dimensions" are at the front of the physical tensors.
+  // 2. stores the new logical tensors back into the passed-in vector. This is
+  //    to avoid additional dynamic allocations.
+  void applyInplace(std::vector<Tensor>& physical_tensors) const;
+
+  std::bitset<kVmapNumLevels> levels_;
+};
+
+} // namespace at

lib/python3.12/site-packages/torch/include/ATen/LinalgBackend.h

@@ -0,0 +1,31 @@
+#pragma once
+
+#include <c10/util/Exception.h>
+
+#include <ostream>
+#include <string>
+
+namespace at {
+
+enum class LinalgBackend : int8_t { Default, Cusolver, Magma };
+
+inline std::string LinalgBackendToString(at::LinalgBackend backend) {
+  switch (backend) {
+    case LinalgBackend::Default:
+      return "at::LinalgBackend::Default";
+    case LinalgBackend::Cusolver:
+      return "at::LinalgBackend::Cusolver";
+    case LinalgBackend::Magma:
+      return "at::LinalgBackend::Magma";
+    default:
+      TORCH_CHECK(false, "Unknown linalg backend");
+  }
+}
+
+inline std::ostream& operator<<(
+    std::ostream& stream,
+    at::LinalgBackend backend) {
+  return stream << LinalgBackendToString(backend);
+}
+
+} // namespace at

lib/python3.12/site-packages/torch/include/ATen/MapAllocator.h

@@ -0,0 +1,147 @@
+#pragma once
+
+#include <c10/core/Allocator.h>
+#include <c10/util/string_view.h>
+
+namespace at {
+
+enum MappedAllocatorModes {
+  ALLOCATOR_MAPPED_SHARED = 1,
+  ALLOCATOR_MAPPED_SHAREDMEM = 2,
+  ALLOCATOR_MAPPED_EXCLUSIVE = 4,
+  ALLOCATOR_MAPPED_NOCREATE = 8,
+  ALLOCATOR_MAPPED_KEEPFD = 16,
+  ALLOCATOR_MAPPED_FROMFD = 32,
+  ALLOCATOR_MAPPED_UNLINK = 64
+};
+
+// Sentinel value/type to help distinguish the file descriptor constructor from
+// the non-file descriptor constructor
+enum WithFd { WITH_FD };
+
+TORCH_API std::string NewProcessWideShmHandle();
+
+class TORCH_API MapAllocator {
+ public:
+  MapAllocator(std::string_view filename, int flags, size_t size);
+  MapAllocator(
+      WithFd,
+      std::string_view filename,
+      int fd,
+      int flags,
+      size_t size);
+  MapAllocator(const MapAllocator&) = delete;
+  MapAllocator& operator=(const MapAllocator&) = delete;
+  MapAllocator(MapAllocator&&) = delete;
+  MapAllocator& operator=(MapAllocator&&) = delete;
+
+  const char* filename() const {
+    return filename_.c_str();
+  }
+  int fd() const {
+#ifdef _WIN32
+    TORCH_CHECK(false, "MapAllocator::fd() is unsupported on Windows");
+#else
+    return fd_;
+#endif
+  }
+  ptrdiff_t size() const {
+    return size_;
+  }
+  // Return a pointer to the actual data for this allocator
+  // (in the case of the refcounted allocator, this is offset
+  // from the base pointer.)
+  virtual void* data() const {
+    return base_ptr_;
+  }
+
+  int flags() const {
+    return flags_;
+  }
+
+  static MapAllocator* fromDataPtr(const at::DataPtr&);
+  static at::DataPtr makeDataPtr(
+      std::string_view filename,
+      int flags,
+      size_t size,
+      size_t* actual_size_out);
+  static at::DataPtr makeDataPtr(
+      WithFd,
+      const char* filename,
+      int fd,
+      int flags,
+      size_t size,
+      size_t* actual_size_out);
+
+  // Closes the data.  Helps us avoid destructor shenanigans
+  virtual void close();
+
+  // This is very dangerous.  You have to redefine this destructor for each
+  // subclass
+  virtual ~MapAllocator();
+
+ protected:
+  bool closed_ = false;
+  std::string filename_;
+  int flags_ = 0;
+  ptrdiff_t size_; /* mapped size */
+#ifdef _WIN32
+  void* handle_;
+  void* event_;
+  std::string eventname_;
+#else
+  int fd_ = -1;
+#endif
+  void* base_ptr_ = nullptr;
+};
+
+// Base-from-member idiom
+struct TORCH_API RefcountedMapAllocatorArgCheck {
+  RefcountedMapAllocatorArgCheck(int flags);
+};
+
+class TORCH_API RefcountedMapAllocator : private RefcountedMapAllocatorArgCheck,
+                                         public MapAllocator {
+ public:
+  RefcountedMapAllocator(const char* filename, int flags, size_t size);
+  RefcountedMapAllocator(
+      WithFd,
+      const char* filename,
+      int fd,
+      int flags,
+      size_t size);
+
+  static RefcountedMapAllocator* fromDataPtr(const at::DataPtr&);
+  RefcountedMapAllocator(const RefcountedMapAllocator&) = delete;
+  RefcountedMapAllocator(RefcountedMapAllocator&&) = delete;
+  RefcountedMapAllocator& operator=(const RefcountedMapAllocator&) = delete;
+  RefcountedMapAllocator& operator=(RefcountedMapAllocator&&) = delete;
+  static at::DataPtr makeDataPtr(
+      const char* filename,
+      int flags,
+      size_t size,
+      size_t* actual_size_out);
+  static at::DataPtr makeDataPtr(
+      WithFd,
+      const char* filename,
+      int fd,
+      int flags,
+      size_t size,
+      size_t* actual_size_out);
+
+  void* data() const override;
+
+  void incref();
+  int decref();
+  void close() override;
+
+  ~RefcountedMapAllocator() override {
+    RefcountedMapAllocator::close();
+  }
+
+ protected:
+  void checkFlags();
+  void initializeAlloc();
+};
+
+} // namespace at

lib/python3.12/site-packages/torch/include/ATen/MetaFunctions_inl.h

@@ -0,0 +1,326 @@
+#pragma once
+// @generated by torchgen/gen.py from DispatchKeyFunctions_inl.h
+
+// NB: The implementing C++ file is RegisterDispatchKey.cpp
+
+// The only #includes we need are for custom classes that have defaults in the C++ API
+#include <c10/core/MemoryFormat.h>
+#include <c10/core/Scalar.h>
+#include <ATen/core/Reduction.h>
+
+#if defined(AT_PER_OPERATOR_HEADERS) && defined(TORCH_ASSERT_ONLY_METHOD_OPERATORS)
+#error This change adds a dependency on all pytorch operators, meaning the     \
+  file will need to be re-compiled every time an operator is changed or added. \
+  Consider including a specific operator from                                  \
+  <ATen/ops/{my_operator}_meta_dispatch.h>.                   \
+  See NOTE [TORCH_ASSERT_ONLY_METHOD_OPERATORS].
+#endif
+
+#include <ATen/ops/_add_relu_meta_dispatch.h>
+#include <ATen/ops/_addmm_activation_meta_dispatch.h>
+#include <ATen/ops/_amp_update_scale_meta_dispatch.h>
+#include <ATen/ops/_assert_tensor_metadata_meta_dispatch.h>
+#include <ATen/ops/_coalesced_meta_dispatch.h>
+#include <ATen/ops/_convert_indices_from_coo_to_csr_meta_dispatch.h>
+#include <ATen/ops/_convert_indices_from_csr_to_coo_meta_dispatch.h>
+#include <ATen/ops/_ctc_loss_meta_dispatch.h>
+#include <ATen/ops/_efficientzerotensor_meta_dispatch.h>
+#include <ATen/ops/_fill_mem_eff_dropout_mask_meta_dispatch.h>
+#include <ATen/ops/_fused_sdp_choice_meta_dispatch.h>
+#include <ATen/ops/_index_put_impl_meta_dispatch.h>
+#include <ATen/ops/_linalg_det_meta_dispatch.h>
+#include <ATen/ops/_linalg_eigh_meta_dispatch.h>
+#include <ATen/ops/_linalg_slogdet_meta_dispatch.h>
+#include <ATen/ops/_linalg_solve_ex_meta_dispatch.h>
+#include <ATen/ops/_linalg_svd_meta_dispatch.h>
+#include <ATen/ops/_log_softmax_meta_dispatch.h>
+#include <ATen/ops/_log_softmax_backward_data_meta_dispatch.h>
+#include <ATen/ops/_mkldnn_transpose_meta_dispatch.h>
+#include <ATen/ops/_reshape_alias_meta_dispatch.h>
+#include <ATen/ops/_resize_output_meta_dispatch.h>
+#include <ATen/ops/_softmax_meta_dispatch.h>
+#include <ATen/ops/_softmax_backward_data_meta_dispatch.h>
+#include <ATen/ops/_sparse_coo_tensor_with_dims_meta_dispatch.h>
+#include <ATen/ops/_sparse_coo_tensor_with_dims_and_tensors_meta_dispatch.h>
+#include <ATen/ops/_upsample_bicubic2d_aa_meta_dispatch.h>
+#include <ATen/ops/_upsample_bicubic2d_aa_backward_meta_dispatch.h>
+#include <ATen/ops/_upsample_bilinear2d_aa_meta_dispatch.h>
+#include <ATen/ops/_upsample_bilinear2d_aa_backward_meta_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact1d_meta_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact1d_backward_meta_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact2d_meta_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact2d_backward_meta_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact3d_meta_dispatch.h>
+#include <ATen/ops/_upsample_nearest_exact3d_backward_meta_dispatch.h>
+#include <ATen/ops/acos_meta_dispatch.h>
+#include <ATen/ops/acosh_meta_dispatch.h>
+#include <ATen/ops/adaptive_max_pool2d_meta_dispatch.h>
+#include <ATen/ops/adaptive_max_pool2d_backward_meta_dispatch.h>
+#include <ATen/ops/adaptive_max_pool3d_meta_dispatch.h>
+#include <ATen/ops/adaptive_max_pool3d_backward_meta_dispatch.h>
+#include <ATen/ops/add_meta_dispatch.h>
+#include <ATen/ops/addbmm_meta_dispatch.h>
+#include <ATen/ops/addcdiv_meta_dispatch.h>
+#include <ATen/ops/addcmul_meta_dispatch.h>
+#include <ATen/ops/addmm_meta_dispatch.h>
+#include <ATen/ops/addmv_meta_dispatch.h>
+#include <ATen/ops/all_meta_dispatch.h>
+#include <ATen/ops/amax_meta_dispatch.h>
+#include <ATen/ops/amin_meta_dispatch.h>
+#include <ATen/ops/aminmax_meta_dispatch.h>
+#include <ATen/ops/any_meta_dispatch.h>
+#include <ATen/ops/arange_meta_dispatch.h>
+#include <ATen/ops/argmax_meta_dispatch.h>
+#include <ATen/ops/argmin_meta_dispatch.h>
+#include <ATen/ops/as_strided_meta_dispatch.h>
+#include <ATen/ops/asin_meta_dispatch.h>
+#include <ATen/ops/asinh_meta_dispatch.h>
+#include <ATen/ops/atan_meta_dispatch.h>
+#include <ATen/ops/atan2_meta_dispatch.h>
+#include <ATen/ops/atanh_meta_dispatch.h>
+#include <ATen/ops/avg_pool2d_meta_dispatch.h>
+#include <ATen/ops/avg_pool2d_backward_meta_dispatch.h>
+#include <ATen/ops/avg_pool3d_meta_dispatch.h>
+#include <ATen/ops/avg_pool3d_backward_meta_dispatch.h>
+#include <ATen/ops/baddbmm_meta_dispatch.h>
+#include <ATen/ops/bernoulli_meta_dispatch.h>
+#include <ATen/ops/bitwise_and_meta_dispatch.h>
+#include <ATen/ops/bitwise_left_shift_meta_dispatch.h>
+#include <ATen/ops/bitwise_not_meta_dispatch.h>
+#include <ATen/ops/bitwise_or_meta_dispatch.h>
+#include <ATen/ops/bitwise_right_shift_meta_dispatch.h>
+#include <ATen/ops/bitwise_xor_meta_dispatch.h>
+#include <ATen/ops/bmm_meta_dispatch.h>
+#include <ATen/ops/cat_meta_dispatch.h>
+#include <ATen/ops/cauchy_meta_dispatch.h>
+#include <ATen/ops/ceil_meta_dispatch.h>
+#include <ATen/ops/clamp_meta_dispatch.h>
+#include <ATen/ops/clamp_max_meta_dispatch.h>
+#include <ATen/ops/clamp_min_meta_dispatch.h>
+#include <ATen/ops/copy_meta_dispatch.h>
+#include <ATen/ops/copy_sparse_to_sparse_meta_dispatch.h>
+#include <ATen/ops/copysign_meta_dispatch.h>
+#include <ATen/ops/cos_meta_dispatch.h>
+#include <ATen/ops/cosh_meta_dispatch.h>
+#include <ATen/ops/cumprod_meta_dispatch.h>
+#include <ATen/ops/cumsum_meta_dispatch.h>
+#include <ATen/ops/digamma_meta_dispatch.h>
+#include <ATen/ops/div_meta_dispatch.h>
+#include <ATen/ops/elu_meta_dispatch.h>
+#include <ATen/ops/elu_backward_meta_dispatch.h>
+#include <ATen/ops/embedding_renorm_meta_dispatch.h>
+#include <ATen/ops/empty_meta_dispatch.h>
+#include <ATen/ops/empty_strided_meta_dispatch.h>
+#include <ATen/ops/eq_meta_dispatch.h>
+#include <ATen/ops/erf_meta_dispatch.h>
+#include <ATen/ops/erfc_meta_dispatch.h>
+#include <ATen/ops/erfinv_meta_dispatch.h>
+#include <ATen/ops/exp_meta_dispatch.h>
+#include <ATen/ops/exp2_meta_dispatch.h>
+#include <ATen/ops/expm1_meta_dispatch.h>
+#include <ATen/ops/exponential_meta_dispatch.h>
+#include <ATen/ops/eye_meta_dispatch.h>
+#include <ATen/ops/fill_meta_dispatch.h>
+#include <ATen/ops/floor_meta_dispatch.h>
+#include <ATen/ops/floor_divide_meta_dispatch.h>
+#include <ATen/ops/fmax_meta_dispatch.h>
+#include <ATen/ops/fmin_meta_dispatch.h>
+#include <ATen/ops/fmod_meta_dispatch.h>
+#include <ATen/ops/frac_meta_dispatch.h>
+#include <ATen/ops/fractional_max_pool2d_meta_dispatch.h>
+#include <ATen/ops/fractional_max_pool2d_backward_meta_dispatch.h>
+#include <ATen/ops/fractional_max_pool3d_meta_dispatch.h>
+#include <ATen/ops/gather_meta_dispatch.h>
+#include <ATen/ops/gcd_meta_dispatch.h>
+#include <ATen/ops/ge_meta_dispatch.h>
+#include <ATen/ops/gelu_meta_dispatch.h>
+#include <ATen/ops/gelu_backward_meta_dispatch.h>
+#include <ATen/ops/geometric_meta_dispatch.h>
+#include <ATen/ops/glu_meta_dispatch.h>
+#include <ATen/ops/gt_meta_dispatch.h>
+#include <ATen/ops/hardshrink_meta_dispatch.h>
+#include <ATen/ops/hardshrink_backward_meta_dispatch.h>
+#include <ATen/ops/hardsigmoid_meta_dispatch.h>
+#include <ATen/ops/hardsigmoid_backward_meta_dispatch.h>
+#include <ATen/ops/hardswish_meta_dispatch.h>
+#include <ATen/ops/hardtanh_meta_dispatch.h>
+#include <ATen/ops/heaviside_meta_dispatch.h>
+#include <ATen/ops/hypot_meta_dispatch.h>
+#include <ATen/ops/i0_meta_dispatch.h>
+#include <ATen/ops/igamma_meta_dispatch.h>
+#include <ATen/ops/igammac_meta_dispatch.h>
+#include <ATen/ops/index_meta_dispatch.h>
+#include <ATen/ops/index_add_meta_dispatch.h>
+#include <ATen/ops/index_copy_meta_dispatch.h>
+#include <ATen/ops/index_fill_meta_dispatch.h>
+#include <ATen/ops/index_reduce_meta_dispatch.h>
+#include <ATen/ops/isin_meta_dispatch.h>
+#include <ATen/ops/isneginf_meta_dispatch.h>
+#include <ATen/ops/isposinf_meta_dispatch.h>
+#include <ATen/ops/lcm_meta_dispatch.h>
+#include <ATen/ops/le_meta_dispatch.h>
+#include <ATen/ops/leaky_relu_meta_dispatch.h>
+#include <ATen/ops/leaky_relu_backward_meta_dispatch.h>
+#include <ATen/ops/lerp_meta_dispatch.h>
+#include <ATen/ops/lgamma_meta_dispatch.h>
+#include <ATen/ops/linalg_cholesky_ex_meta_dispatch.h>
+#include <ATen/ops/linalg_cross_meta_dispatch.h>
+#include <ATen/ops/linalg_inv_ex_meta_dispatch.h>
+#include <ATen/ops/linalg_ldl_factor_ex_meta_dispatch.h>
+#include <ATen/ops/linalg_ldl_solve_meta_dispatch.h>
+#include <ATen/ops/linalg_lu_meta_dispatch.h>
+#include <ATen/ops/linalg_lu_factor_ex_meta_dispatch.h>
+#include <ATen/ops/linalg_lu_solve_meta_dispatch.h>
+#include <ATen/ops/linalg_qr_meta_dispatch.h>
+#include <ATen/ops/linalg_vector_norm_meta_dispatch.h>
+#include <ATen/ops/linspace_meta_dispatch.h>
+#include <ATen/ops/log_meta_dispatch.h>
+#include <ATen/ops/log10_meta_dispatch.h>
+#include <ATen/ops/log1p_meta_dispatch.h>
+#include <ATen/ops/log2_meta_dispatch.h>
+#include <ATen/ops/log_normal_meta_dispatch.h>
+#include <ATen/ops/logaddexp_meta_dispatch.h>
+#include <ATen/ops/logaddexp2_meta_dispatch.h>
+#include <ATen/ops/logit_meta_dispatch.h>
+#include <ATen/ops/logit_backward_meta_dispatch.h>
+#include <ATen/ops/logspace_meta_dispatch.h>
+#include <ATen/ops/lshift_meta_dispatch.h>
+#include <ATen/ops/lt_meta_dispatch.h>
+#include <ATen/ops/lu_unpack_meta_dispatch.h>
+#include <ATen/ops/masked_fill_meta_dispatch.h>
+#include <ATen/ops/masked_scatter_meta_dispatch.h>
+#include <ATen/ops/max_meta_dispatch.h>
+#include <ATen/ops/max_pool2d_with_indices_meta_dispatch.h>
+#include <ATen/ops/max_pool2d_with_indices_backward_meta_dispatch.h>
+#include <ATen/ops/maximum_meta_dispatch.h>
+#include <ATen/ops/mean_meta_dispatch.h>
+#include <ATen/ops/min_meta_dispatch.h>
+#include <ATen/ops/minimum_meta_dispatch.h>
+#include <ATen/ops/mish_meta_dispatch.h>
+#include <ATen/ops/mm_meta_dispatch.h>
+#include <ATen/ops/mse_loss_meta_dispatch.h>
+#include <ATen/ops/mul_meta_dispatch.h>
+#include <ATen/ops/ne_meta_dispatch.h>
+#include <ATen/ops/neg_meta_dispatch.h>
+#include <ATen/ops/nextafter_meta_dispatch.h>
+#include <ATen/ops/nll_loss_backward_meta_dispatch.h>
+#include <ATen/ops/nll_loss_forward_meta_dispatch.h>
+#include <ATen/ops/norm_meta_dispatch.h>
+#include <ATen/ops/normal_meta_dispatch.h>
+#include <ATen/ops/polygamma_meta_dispatch.h>
+#include <ATen/ops/pow_meta_dispatch.h>
+#include <ATen/ops/prod_meta_dispatch.h>
+#include <ATen/ops/put_meta_dispatch.h>
+#include <ATen/ops/random_meta_dispatch.h>
+#include <ATen/ops/range_meta_dispatch.h>
+#include <ATen/ops/reciprocal_meta_dispatch.h>
+#include <ATen/ops/reflection_pad1d_meta_dispatch.h>
+#include <ATen/ops/reflection_pad1d_backward_meta_dispatch.h>
+#include <ATen/ops/reflection_pad3d_meta_dispatch.h>
+#include <ATen/ops/reflection_pad3d_backward_meta_dispatch.h>
+#include <ATen/ops/relu_meta_dispatch.h>
+#include <ATen/ops/remainder_meta_dispatch.h>
+#include <ATen/ops/renorm_meta_dispatch.h>
+#include <ATen/ops/replication_pad1d_meta_dispatch.h>
+#include <ATen/ops/replication_pad1d_backward_meta_dispatch.h>
+#include <ATen/ops/replication_pad2d_meta_dispatch.h>
+#include <ATen/ops/replication_pad3d_meta_dispatch.h>
+#include <ATen/ops/resize_meta_dispatch.h>
+#include <ATen/ops/resize_as_sparse_meta_dispatch.h>
+#include <ATen/ops/round_meta_dispatch.h>
+#include <ATen/ops/rrelu_with_noise_meta_dispatch.h>
+#include <ATen/ops/rshift_meta_dispatch.h>
+#include <ATen/ops/rsqrt_meta_dispatch.h>
+#include <ATen/ops/scatter_meta_dispatch.h>
+#include <ATen/ops/scatter_add_meta_dispatch.h>
+#include <ATen/ops/scatter_reduce_meta_dispatch.h>
+#include <ATen/ops/set_meta_dispatch.h>
+#include <ATen/ops/sgn_meta_dispatch.h>
+#include <ATen/ops/sigmoid_meta_dispatch.h>
+#include <ATen/ops/sigmoid_backward_meta_dispatch.h>
+#include <ATen/ops/sign_meta_dispatch.h>
+#include <ATen/ops/signbit_meta_dispatch.h>
+#include <ATen/ops/silu_meta_dispatch.h>
+#include <ATen/ops/silu_backward_meta_dispatch.h>
+#include <ATen/ops/sin_meta_dispatch.h>
+#include <ATen/ops/sinc_meta_dispatch.h>
+#include <ATen/ops/sinh_meta_dispatch.h>
+#include <ATen/ops/slow_conv_transpose2d_meta_dispatch.h>
+#include <ATen/ops/smooth_l1_loss_meta_dispatch.h>
+#include <ATen/ops/softplus_meta_dispatch.h>
+#include <ATen/ops/softplus_backward_meta_dispatch.h>
+#include <ATen/ops/softshrink_meta_dispatch.h>
+#include <ATen/ops/softshrink_backward_meta_dispatch.h>
+#include <ATen/ops/sort_meta_dispatch.h>
+#include <ATen/ops/sparse_resize_meta_dispatch.h>
+#include <ATen/ops/sparse_resize_and_clear_meta_dispatch.h>
+#include <ATen/ops/special_airy_ai_meta_dispatch.h>
+#include <ATen/ops/special_bessel_j0_meta_dispatch.h>
+#include <ATen/ops/special_bessel_j1_meta_dispatch.h>
+#include <ATen/ops/special_bessel_y0_meta_dispatch.h>
+#include <ATen/ops/special_bessel_y1_meta_dispatch.h>
+#include <ATen/ops/special_chebyshev_polynomial_t_meta_dispatch.h>
+#include <ATen/ops/special_chebyshev_polynomial_u_meta_dispatch.h>
+#include <ATen/ops/special_chebyshev_polynomial_v_meta_dispatch.h>
+#include <ATen/ops/special_chebyshev_polynomial_w_meta_dispatch.h>
+#include <ATen/ops/special_entr_meta_dispatch.h>
+#include <ATen/ops/special_erfcx_meta_dispatch.h>
+#include <ATen/ops/special_hermite_polynomial_h_meta_dispatch.h>
+#include <ATen/ops/special_hermite_polynomial_he_meta_dispatch.h>
+#include <ATen/ops/special_i0e_meta_dispatch.h>
+#include <ATen/ops/special_i1_meta_dispatch.h>
+#include <ATen/ops/special_i1e_meta_dispatch.h>
+#include <ATen/ops/special_laguerre_polynomial_l_meta_dispatch.h>
+#include <ATen/ops/special_legendre_polynomial_p_meta_dispatch.h>
+#include <ATen/ops/special_log_ndtr_meta_dispatch.h>
+#include <ATen/ops/special_modified_bessel_i0_meta_dispatch.h>
+#include <ATen/ops/special_modified_bessel_i1_meta_dispatch.h>
+#include <ATen/ops/special_modified_bessel_k0_meta_dispatch.h>
+#include <ATen/ops/special_modified_bessel_k1_meta_dispatch.h>
+#include <ATen/ops/special_ndtri_meta_dispatch.h>
+#include <ATen/ops/special_scaled_modified_bessel_k0_meta_dispatch.h>
+#include <ATen/ops/special_scaled_modified_bessel_k1_meta_dispatch.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_t_meta_dispatch.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_u_meta_dispatch.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_v_meta_dispatch.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_w_meta_dispatch.h>
+#include <ATen/ops/special_spherical_bessel_j0_meta_dispatch.h>
+#include <ATen/ops/special_xlog1py_meta_dispatch.h>
+#include <ATen/ops/special_zeta_meta_dispatch.h>
+#include <ATen/ops/sqrt_meta_dispatch.h>
+#include <ATen/ops/sub_meta_dispatch.h>
+#include <ATen/ops/sum_meta_dispatch.h>
+#include <ATen/ops/tan_meta_dispatch.h>
+#include <ATen/ops/tanh_meta_dispatch.h>
+#include <ATen/ops/tanh_backward_meta_dispatch.h>
+#include <ATen/ops/threshold_meta_dispatch.h>
+#include <ATen/ops/threshold_backward_meta_dispatch.h>
+#include <ATen/ops/topk_meta_dispatch.h>
+#include <ATen/ops/triangular_solve_meta_dispatch.h>
+#include <ATen/ops/tril_meta_dispatch.h>
+#include <ATen/ops/triu_meta_dispatch.h>
+#include <ATen/ops/trunc_meta_dispatch.h>
+#include <ATen/ops/unfold_meta_dispatch.h>
+#include <ATen/ops/uniform_meta_dispatch.h>
+#include <ATen/ops/upsample_bicubic2d_meta_dispatch.h>
+#include <ATen/ops/upsample_bicubic2d_backward_meta_dispatch.h>
+#include <ATen/ops/upsample_bilinear2d_meta_dispatch.h>
+#include <ATen/ops/upsample_bilinear2d_backward_meta_dispatch.h>
+#include <ATen/ops/upsample_linear1d_meta_dispatch.h>
+#include <ATen/ops/upsample_linear1d_backward_meta_dispatch.h>
+#include <ATen/ops/upsample_nearest1d_meta_dispatch.h>
+#include <ATen/ops/upsample_nearest1d_backward_meta_dispatch.h>
+#include <ATen/ops/upsample_nearest2d_meta_dispatch.h>
+#include <ATen/ops/upsample_nearest2d_backward_meta_dispatch.h>
+#include <ATen/ops/upsample_nearest3d_meta_dispatch.h>
+#include <ATen/ops/upsample_nearest3d_backward_meta_dispatch.h>
+#include <ATen/ops/upsample_trilinear3d_meta_dispatch.h>
+#include <ATen/ops/upsample_trilinear3d_backward_meta_dispatch.h>
+#include <ATen/ops/view_meta_dispatch.h>
+#include <ATen/ops/view_as_complex_meta_dispatch.h>
+#include <ATen/ops/view_as_real_meta_dispatch.h>
+#include <ATen/ops/xlogy_meta_dispatch.h>
+#include <ATen/ops/zero_meta_dispatch.h>
+
+
+

lib/python3.12/site-packages/torch/include/ATen/MethodOperators.h

@@ -0,0 +1,443 @@
+#pragma once
+
+// @generated by torchgen/gen.py from MethodOperators.h
+
+#ifdef TORCH_ASSERT_NO_OPERATORS
+#error This change adds a dependency on native_functions.yaml,             \
+  meaning the file will need to be re-compiled every time an operator      \
+  is changed or added. Consider if your change would be better placed in   \
+  another file, or if a more specific header might achieve the same goal.  \
+  See NOTE: [Tensor vs. TensorBase]
+#endif
+
+// Forward declarations of any types needed in the operator signatures.
+// We can't directly include these classes because it will cause circular include dependencies.
+// This file is included by TensorBody.h, which defines the Tensor class.
+#include <ATen/core/ATen_fwd.h>
+
+#include <ATen/ops/_addmm_activation_ops.h>
+#include <ATen/ops/_autocast_to_full_precision_ops.h>
+#include <ATen/ops/_autocast_to_reduced_precision_ops.h>
+#include <ATen/ops/_backward_ops.h>
+#include <ATen/ops/_coalesced_ops.h>
+#include <ATen/ops/_conj_ops.h>
+#include <ATen/ops/_conj_physical_ops.h>
+#include <ATen/ops/_dimI_ops.h>
+#include <ATen/ops/_dimV_ops.h>
+#include <ATen/ops/_fw_primal_ops.h>
+#include <ATen/ops/_indices_ops.h>
+#include <ATen/ops/_is_all_true_ops.h>
+#include <ATen/ops/_is_any_true_ops.h>
+#include <ATen/ops/_is_zerotensor_ops.h>
+#include <ATen/ops/_lazy_clone_ops.h>
+#include <ATen/ops/_neg_view_ops.h>
+#include <ATen/ops/_nested_tensor_size_ops.h>
+#include <ATen/ops/_nested_tensor_storage_offsets_ops.h>
+#include <ATen/ops/_nested_tensor_strides_ops.h>
+#include <ATen/ops/_nnz_ops.h>
+#include <ATen/ops/_reshape_alias_ops.h>
+#include <ATen/ops/_sparse_mask_projection_ops.h>
+#include <ATen/ops/_to_dense_ops.h>
+#include <ATen/ops/_to_sparse_bsc_ops.h>
+#include <ATen/ops/_to_sparse_bsr_ops.h>
+#include <ATen/ops/_to_sparse_csc_ops.h>
+#include <ATen/ops/_to_sparse_csr_ops.h>
+#include <ATen/ops/_to_sparse_ops.h>
+#include <ATen/ops/_values_ops.h>
+#include <ATen/ops/_version_ops.h>
+#include <ATen/ops/abs_ops.h>
+#include <ATen/ops/absolute_ops.h>
+#include <ATen/ops/acos_ops.h>
+#include <ATen/ops/acosh_ops.h>
+#include <ATen/ops/add_ops.h>
+#include <ATen/ops/addbmm_ops.h>
+#include <ATen/ops/addcdiv_ops.h>
+#include <ATen/ops/addcmul_ops.h>
+#include <ATen/ops/addmm_ops.h>
+#include <ATen/ops/addmv_ops.h>
+#include <ATen/ops/addr_ops.h>
+#include <ATen/ops/adjoint_ops.h>
+#include <ATen/ops/alias_ops.h>
+#include <ATen/ops/align_as_ops.h>
+#include <ATen/ops/align_to_ops.h>
+#include <ATen/ops/all_ops.h>
+#include <ATen/ops/allclose_ops.h>
+#include <ATen/ops/amax_ops.h>
+#include <ATen/ops/amin_ops.h>
+#include <ATen/ops/aminmax_ops.h>
+#include <ATen/ops/and_ops.h>
+#include <ATen/ops/angle_ops.h>
+#include <ATen/ops/any_ops.h>
+#include <ATen/ops/arccos_ops.h>
+#include <ATen/ops/arccosh_ops.h>
+#include <ATen/ops/arcsin_ops.h>
+#include <ATen/ops/arcsinh_ops.h>
+#include <ATen/ops/arctan2_ops.h>
+#include <ATen/ops/arctan_ops.h>
+#include <ATen/ops/arctanh_ops.h>
+#include <ATen/ops/argmax_ops.h>
+#include <ATen/ops/argmin_ops.h>
+#include <ATen/ops/argsort_ops.h>
+#include <ATen/ops/argwhere_ops.h>
+#include <ATen/ops/as_strided_ops.h>
+#include <ATen/ops/as_strided_scatter_ops.h>
+#include <ATen/ops/asin_ops.h>
+#include <ATen/ops/asinh_ops.h>
+#include <ATen/ops/atan2_ops.h>
+#include <ATen/ops/atan_ops.h>
+#include <ATen/ops/atanh_ops.h>
+#include <ATen/ops/baddbmm_ops.h>
+#include <ATen/ops/bernoulli_ops.h>
+#include <ATen/ops/bincount_ops.h>
+#include <ATen/ops/bitwise_and_ops.h>
+#include <ATen/ops/bitwise_left_shift_ops.h>
+#include <ATen/ops/bitwise_not_ops.h>
+#include <ATen/ops/bitwise_or_ops.h>
+#include <ATen/ops/bitwise_right_shift_ops.h>
+#include <ATen/ops/bitwise_xor_ops.h>
+#include <ATen/ops/bmm_ops.h>
+#include <ATen/ops/broadcast_to_ops.h>
+#include <ATen/ops/cauchy_ops.h>
+#include <ATen/ops/ccol_indices_ops.h>
+#include <ATen/ops/ceil_ops.h>
+#include <ATen/ops/chalf_ops.h>
+#include <ATen/ops/cholesky_inverse_ops.h>
+#include <ATen/ops/cholesky_ops.h>
+#include <ATen/ops/cholesky_solve_ops.h>
+#include <ATen/ops/chunk_ops.h>
+#include <ATen/ops/clamp_max_ops.h>
+#include <ATen/ops/clamp_min_ops.h>
+#include <ATen/ops/clamp_ops.h>
+#include <ATen/ops/clip_ops.h>
+#include <ATen/ops/clone_ops.h>
+#include <ATen/ops/coalesce_ops.h>
+#include <ATen/ops/col_indices_ops.h>
+#include <ATen/ops/conj_ops.h>
+#include <ATen/ops/conj_physical_ops.h>
+#include <ATen/ops/contiguous_ops.h>
+#include <ATen/ops/copy_ops.h>
+#include <ATen/ops/copysign_ops.h>
+#include <ATen/ops/corrcoef_ops.h>
+#include <ATen/ops/cos_ops.h>
+#include <ATen/ops/cosh_ops.h>
+#include <ATen/ops/count_nonzero_ops.h>
+#include <ATen/ops/cov_ops.h>
+#include <ATen/ops/cross_ops.h>
+#include <ATen/ops/crow_indices_ops.h>
+#include <ATen/ops/cummax_ops.h>
+#include <ATen/ops/cummin_ops.h>
+#include <ATen/ops/cumprod_ops.h>
+#include <ATen/ops/cumsum_ops.h>
+#include <ATen/ops/data_ops.h>
+#include <ATen/ops/deg2rad_ops.h>
+#include <ATen/ops/dense_dim_ops.h>
+#include <ATen/ops/dequantize_ops.h>
+#include <ATen/ops/det_ops.h>
+#include <ATen/ops/detach_ops.h>
+#include <ATen/ops/diag_embed_ops.h>
+#include <ATen/ops/diag_ops.h>
+#include <ATen/ops/diagflat_ops.h>
+#include <ATen/ops/diagonal_ops.h>
+#include <ATen/ops/diagonal_scatter_ops.h>
+#include <ATen/ops/diff_ops.h>
+#include <ATen/ops/digamma_ops.h>
+#include <ATen/ops/dist_ops.h>
+#include <ATen/ops/div_ops.h>
+#include <ATen/ops/divide_ops.h>
+#include <ATen/ops/dot_ops.h>
+#include <ATen/ops/dsplit_ops.h>
+#include <ATen/ops/eq_ops.h>
+#include <ATen/ops/equal_ops.h>
+#include <ATen/ops/erf_ops.h>
+#include <ATen/ops/erfc_ops.h>
+#include <ATen/ops/erfinv_ops.h>
+#include <ATen/ops/exp2_ops.h>
+#include <ATen/ops/exp_ops.h>
+#include <ATen/ops/expand_as_ops.h>
+#include <ATen/ops/expand_ops.h>
+#include <ATen/ops/expm1_ops.h>
+#include <ATen/ops/exponential_ops.h>
+#include <ATen/ops/fill_diagonal_ops.h>
+#include <ATen/ops/fill_ops.h>
+#include <ATen/ops/fix_ops.h>
+#include <ATen/ops/flatten_ops.h>
+#include <ATen/ops/flip_ops.h>
+#include <ATen/ops/fliplr_ops.h>
+#include <ATen/ops/flipud_ops.h>
+#include <ATen/ops/float_power_ops.h>
+#include <ATen/ops/floor_divide_ops.h>
+#include <ATen/ops/floor_ops.h>
+#include <ATen/ops/fmax_ops.h>
+#include <ATen/ops/fmin_ops.h>
+#include <ATen/ops/fmod_ops.h>
+#include <ATen/ops/frac_ops.h>
+#include <ATen/ops/frexp_ops.h>
+#include <ATen/ops/gather_ops.h>
+#include <ATen/ops/gcd_ops.h>
+#include <ATen/ops/ge_ops.h>
+#include <ATen/ops/geometric_ops.h>
+#include <ATen/ops/geqrf_ops.h>
+#include <ATen/ops/ger_ops.h>
+#include <ATen/ops/greater_equal_ops.h>
+#include <ATen/ops/greater_ops.h>
+#include <ATen/ops/gt_ops.h>
+#include <ATen/ops/hardshrink_backward_ops.h>
+#include <ATen/ops/hardshrink_ops.h>
+#include <ATen/ops/heaviside_ops.h>
+#include <ATen/ops/histc_ops.h>
+#include <ATen/ops/histogram_ops.h>
+#include <ATen/ops/hsplit_ops.h>
+#include <ATen/ops/hypot_ops.h>
+#include <ATen/ops/i0_ops.h>
+#include <ATen/ops/igamma_ops.h>
+#include <ATen/ops/igammac_ops.h>
+#include <ATen/ops/index_add_ops.h>
+#include <ATen/ops/index_copy_ops.h>
+#include <ATen/ops/index_fill_ops.h>
+#include <ATen/ops/index_ops.h>
+#include <ATen/ops/index_put_ops.h>
+#include <ATen/ops/index_reduce_ops.h>
+#include <ATen/ops/index_select_ops.h>
+#include <ATen/ops/indices_ops.h>
+#include <ATen/ops/inner_ops.h>
+#include <ATen/ops/int_repr_ops.h>
+#include <ATen/ops/inverse_ops.h>
+#include <ATen/ops/is_coalesced_ops.h>
+#include <ATen/ops/is_complex_ops.h>
+#include <ATen/ops/is_conj_ops.h>
+#include <ATen/ops/is_distributed_ops.h>
+#include <ATen/ops/is_floating_point_ops.h>
+#include <ATen/ops/is_inference_ops.h>
+#include <ATen/ops/is_leaf_ops.h>
+#include <ATen/ops/is_neg_ops.h>
+#include <ATen/ops/is_nonzero_ops.h>
+#include <ATen/ops/is_pinned_ops.h>
+#include <ATen/ops/is_same_size_ops.h>
+#include <ATen/ops/is_set_to_ops.h>
+#include <ATen/ops/is_signed_ops.h>
+#include <ATen/ops/isclose_ops.h>
+#include <ATen/ops/isfinite_ops.h>
+#include <ATen/ops/isinf_ops.h>
+#include <ATen/ops/isnan_ops.h>
+#include <ATen/ops/isneginf_ops.h>
+#include <ATen/ops/isposinf_ops.h>
+#include <ATen/ops/isreal_ops.h>
+#include <ATen/ops/istft_ops.h>
+#include <ATen/ops/item_ops.h>
+#include <ATen/ops/kron_ops.h>
+#include <ATen/ops/kthvalue_ops.h>
+#include <ATen/ops/lcm_ops.h>
+#include <ATen/ops/ldexp_ops.h>
+#include <ATen/ops/le_ops.h>
+#include <ATen/ops/lerp_ops.h>
+#include <ATen/ops/less_equal_ops.h>
+#include <ATen/ops/less_ops.h>
+#include <ATen/ops/lgamma_ops.h>
+#include <ATen/ops/log10_ops.h>
+#include <ATen/ops/log1p_ops.h>
+#include <ATen/ops/log2_ops.h>
+#include <ATen/ops/log_normal_ops.h>
+#include <ATen/ops/log_ops.h>
+#include <ATen/ops/log_softmax_ops.h>
+#include <ATen/ops/logaddexp2_ops.h>
+#include <ATen/ops/logaddexp_ops.h>
+#include <ATen/ops/logcumsumexp_ops.h>
+#include <ATen/ops/logdet_ops.h>
+#include <ATen/ops/logical_and_ops.h>
+#include <ATen/ops/logical_not_ops.h>
+#include <ATen/ops/logical_or_ops.h>
+#include <ATen/ops/logical_xor_ops.h>
+#include <ATen/ops/logit_ops.h>
+#include <ATen/ops/logsumexp_ops.h>
+#include <ATen/ops/lshift_ops.h>
+#include <ATen/ops/lt_ops.h>
+#include <ATen/ops/lu_solve_ops.h>
+#include <ATen/ops/mH_ops.h>
+#include <ATen/ops/mT_ops.h>
+#include <ATen/ops/masked_fill_ops.h>
+#include <ATen/ops/masked_scatter_ops.h>
+#include <ATen/ops/masked_select_ops.h>
+#include <ATen/ops/matmul_ops.h>
+#include <ATen/ops/matrix_H_ops.h>
+#include <ATen/ops/matrix_exp_ops.h>
+#include <ATen/ops/matrix_power_ops.h>
+#include <ATen/ops/max_ops.h>
+#include <ATen/ops/maximum_ops.h>
+#include <ATen/ops/mean_ops.h>
+#include <ATen/ops/median_ops.h>
+#include <ATen/ops/min_ops.h>
+#include <ATen/ops/minimum_ops.h>
+#include <ATen/ops/mm_ops.h>
+#include <ATen/ops/mode_ops.h>
+#include <ATen/ops/moveaxis_ops.h>
+#include <ATen/ops/movedim_ops.h>
+#include <ATen/ops/msort_ops.h>
+#include <ATen/ops/mul_ops.h>
+#include <ATen/ops/multinomial_ops.h>
+#include <ATen/ops/multiply_ops.h>
+#include <ATen/ops/mv_ops.h>
+#include <ATen/ops/mvlgamma_ops.h>
+#include <ATen/ops/nan_to_num_ops.h>
+#include <ATen/ops/nanmean_ops.h>
+#include <ATen/ops/nanmedian_ops.h>
+#include <ATen/ops/nanquantile_ops.h>
+#include <ATen/ops/nansum_ops.h>
+#include <ATen/ops/narrow_copy_ops.h>
+#include <ATen/ops/narrow_ops.h>
+#include <ATen/ops/ne_ops.h>
+#include <ATen/ops/neg_ops.h>
+#include <ATen/ops/negative_ops.h>
+#include <ATen/ops/new_empty_ops.h>
+#include <ATen/ops/new_empty_strided_ops.h>
+#include <ATen/ops/new_full_ops.h>
+#include <ATen/ops/new_ones_ops.h>
+#include <ATen/ops/new_zeros_ops.h>
+#include <ATen/ops/nextafter_ops.h>
+#include <ATen/ops/nonzero_numpy_ops.h>
+#include <ATen/ops/nonzero_ops.h>
+#include <ATen/ops/nonzero_static_ops.h>
+#include <ATen/ops/norm_ops.h>
+#include <ATen/ops/normal_ops.h>
+#include <ATen/ops/not_equal_ops.h>
+#include <ATen/ops/numpy_T_ops.h>
+#include <ATen/ops/or_ops.h>
+#include <ATen/ops/orgqr_ops.h>
+#include <ATen/ops/ormqr_ops.h>
+#include <ATen/ops/outer_ops.h>
+#include <ATen/ops/output_nr_ops.h>
+#include <ATen/ops/permute_ops.h>
+#include <ATen/ops/pin_memory_ops.h>
+#include <ATen/ops/pinverse_ops.h>
+#include <ATen/ops/polygamma_ops.h>
+#include <ATen/ops/positive_ops.h>
+#include <ATen/ops/pow_ops.h>
+#include <ATen/ops/prelu_ops.h>
+#include <ATen/ops/prod_ops.h>
+#include <ATen/ops/put_ops.h>
+#include <ATen/ops/q_per_channel_axis_ops.h>
+#include <ATen/ops/q_per_channel_scales_ops.h>
+#include <ATen/ops/q_per_channel_zero_points_ops.h>
+#include <ATen/ops/q_scale_ops.h>
+#include <ATen/ops/q_zero_point_ops.h>
+#include <ATen/ops/qr_ops.h>
+#include <ATen/ops/qscheme_ops.h>
+#include <ATen/ops/quantile_ops.h>
+#include <ATen/ops/rad2deg_ops.h>
+#include <ATen/ops/random_ops.h>
+#include <ATen/ops/ravel_ops.h>
+#include <ATen/ops/reciprocal_ops.h>
+#include <ATen/ops/record_stream_ops.h>
+#include <ATen/ops/refine_names_ops.h>
+#include <ATen/ops/relu_ops.h>
+#include <ATen/ops/remainder_ops.h>
+#include <ATen/ops/rename_ops.h>
+#include <ATen/ops/renorm_ops.h>
+#include <ATen/ops/repeat_interleave_ops.h>
+#include <ATen/ops/repeat_ops.h>
+#include <ATen/ops/requires_grad_ops.h>
+#include <ATen/ops/reshape_as_ops.h>
+#include <ATen/ops/reshape_ops.h>
+#include <ATen/ops/resize_as_ops.h>
+#include <ATen/ops/resize_as_sparse_ops.h>
+#include <ATen/ops/resize_ops.h>
+#include <ATen/ops/resolve_conj_ops.h>
+#include <ATen/ops/resolve_neg_ops.h>
+#include <ATen/ops/retain_grad_ops.h>
+#include <ATen/ops/retains_grad_ops.h>
+#include <ATen/ops/roll_ops.h>
+#include <ATen/ops/rot90_ops.h>
+#include <ATen/ops/round_ops.h>
+#include <ATen/ops/row_indices_ops.h>
+#include <ATen/ops/rshift_ops.h>
+#include <ATen/ops/rsqrt_ops.h>
+#include <ATen/ops/scatter_add_ops.h>
+#include <ATen/ops/scatter_ops.h>
+#include <ATen/ops/scatter_reduce_ops.h>
+#include <ATen/ops/select_ops.h>
+#include <ATen/ops/select_scatter_ops.h>
+#include <ATen/ops/set_data_ops.h>
+#include <ATen/ops/set_ops.h>
+#include <ATen/ops/sgn_ops.h>
+#include <ATen/ops/sigmoid_ops.h>
+#include <ATen/ops/sign_ops.h>
+#include <ATen/ops/signbit_ops.h>
+#include <ATen/ops/sin_ops.h>
+#include <ATen/ops/sinc_ops.h>
+#include <ATen/ops/sinh_ops.h>
+#include <ATen/ops/size_ops.h>
+#include <ATen/ops/slice_inverse_ops.h>
+#include <ATen/ops/slice_ops.h>
+#include <ATen/ops/slice_scatter_ops.h>
+#include <ATen/ops/slogdet_ops.h>
+#include <ATen/ops/smm_ops.h>
+#include <ATen/ops/softmax_ops.h>
+#include <ATen/ops/sort_ops.h>
+#include <ATen/ops/sparse_dim_ops.h>
+#include <ATen/ops/sparse_mask_ops.h>
+#include <ATen/ops/sparse_resize_and_clear_ops.h>
+#include <ATen/ops/sparse_resize_ops.h>
+#include <ATen/ops/split_ops.h>
+#include <ATen/ops/split_with_sizes_ops.h>
+#include <ATen/ops/sqrt_ops.h>
+#include <ATen/ops/square_ops.h>
+#include <ATen/ops/squeeze_ops.h>
+#include <ATen/ops/sspaddmm_ops.h>
+#include <ATen/ops/std_ops.h>
+#include <ATen/ops/stft_ops.h>
+#include <ATen/ops/stride_ops.h>
+#include <ATen/ops/sub_ops.h>
+#include <ATen/ops/subtract_ops.h>
+#include <ATen/ops/sum_ops.h>
+#include <ATen/ops/sum_to_size_ops.h>
+#include <ATen/ops/svd_ops.h>
+#include <ATen/ops/swapaxes_ops.h>
+#include <ATen/ops/swapdims_ops.h>
+#include <ATen/ops/t_ops.h>
+#include <ATen/ops/take_along_dim_ops.h>
+#include <ATen/ops/take_ops.h>
+#include <ATen/ops/tan_ops.h>
+#include <ATen/ops/tanh_ops.h>
+#include <ATen/ops/tensor_split_ops.h>
+#include <ATen/ops/tile_ops.h>
+#include <ATen/ops/to_dense_ops.h>
+#include <ATen/ops/to_mkldnn_ops.h>
+#include <ATen/ops/to_ops.h>
+#include <ATen/ops/to_padded_tensor_ops.h>
+#include <ATen/ops/to_sparse_bsc_ops.h>
+#include <ATen/ops/to_sparse_bsr_ops.h>
+#include <ATen/ops/to_sparse_csc_ops.h>
+#include <ATen/ops/to_sparse_csr_ops.h>
+#include <ATen/ops/to_sparse_ops.h>
+#include <ATen/ops/topk_ops.h>
+#include <ATen/ops/trace_ops.h>
+#include <ATen/ops/transpose_ops.h>
+#include <ATen/ops/triangular_solve_ops.h>
+#include <ATen/ops/tril_ops.h>
+#include <ATen/ops/triu_ops.h>
+#include <ATen/ops/true_divide_ops.h>
+#include <ATen/ops/trunc_ops.h>
+#include <ATen/ops/type_as_ops.h>
+#include <ATen/ops/unbind_ops.h>
+#include <ATen/ops/unflatten_ops.h>
+#include <ATen/ops/unfold_ops.h>
+#include <ATen/ops/uniform_ops.h>
+#include <ATen/ops/unsafe_chunk_ops.h>
+#include <ATen/ops/unsafe_split_ops.h>
+#include <ATen/ops/unsafe_split_with_sizes_ops.h>
+#include <ATen/ops/unsqueeze_ops.h>
+#include <ATen/ops/values_ops.h>
+#include <ATen/ops/var_ops.h>
+#include <ATen/ops/vdot_ops.h>
+#include <ATen/ops/view_as_ops.h>
+#include <ATen/ops/view_ops.h>
+#include <ATen/ops/vsplit_ops.h>
+#include <ATen/ops/where_ops.h>
+#include <ATen/ops/xlogy_ops.h>
+#include <ATen/ops/xor_ops.h>
+#include <ATen/ops/zero_ops.h>
+
+namespace at {
+namespace _ops {
+
+} // namespace _ops
+} // namespace at

lib/python3.12/site-packages/torch/include/ATen/NamedTensor.h

@@ -0,0 +1 @@
+#include <ATen/core/NamedTensor.h>

lib/python3.12/site-packages/torch/include/ATen/NativeFunctions.h

@@ -0,0 +1,1348 @@
+#pragma once
+
+// @generated by torchgen/gen.py from NativeFunctions.h
+
+#ifdef TORCH_ASSERT_NO_OPERATORS
+#error This change adds a dependency on native_functions.yaml,            \
+  meaning the file will need to be re-compiled every time an operator     \
+  is changed or added. Consider if your change would be better placed in  \
+  another file, or if a more specific header might achieve the same goal. \
+  See NOTE: [Tensor vs. TensorBase]
+#endif
+
+#if defined(AT_PER_OPERATOR_HEADERS) && defined(TORCH_ASSERT_ONLY_METHOD_OPERATORS)
+#error This change adds a dependency on all pytorch operators, meaning the      \
+  file will need to be re-compiled every time an operator is changed or added.  \
+  Consider including a specific operator from <ATen/ops/{my_operator}_native.h> \
+  and see NOTE [TORCH_ASSERT_ONLY_METHOD_OPERATORS].
+#endif
+
+#include <c10/core/Scalar.h>
+#include <c10/core/Storage.h>
+#include <c10/core/TensorOptions.h>
+#include <c10/util/Deprecated.h>
+#include <optional>
+#include <c10/core/QScheme.h>
+#include <ATen/core/Reduction.h>
+#include <ATen/core/Tensor.h>
+#include <tuple>
+#include <vector>
+
+#include <ATen/ops/_adaptive_avg_pool2d_native.h>
+#include <ATen/ops/_adaptive_avg_pool2d_backward_native.h>
+#include <ATen/ops/_adaptive_avg_pool3d_native.h>
+#include <ATen/ops/_adaptive_avg_pool3d_backward_native.h>
+#include <ATen/ops/_add_batch_dim_native.h>
+#include <ATen/ops/_add_relu_native.h>
+#include <ATen/ops/_addmm_activation_native.h>
+#include <ATen/ops/_aminmax_native.h>
+#include <ATen/ops/_amp_foreach_non_finite_check_and_unscale_native.h>
+#include <ATen/ops/_amp_update_scale_native.h>
+#include <ATen/ops/_assert_async_native.h>
+#include <ATen/ops/_assert_scalar_native.h>
+#include <ATen/ops/_assert_tensor_metadata_native.h>
+#include <ATen/ops/_autocast_to_full_precision_native.h>
+#include <ATen/ops/_autocast_to_reduced_precision_native.h>
+#include <ATen/ops/_backward_native.h>
+#include <ATen/ops/_batch_norm_impl_index_native.h>
+#include <ATen/ops/_batch_norm_impl_index_backward_native.h>
+#include <ATen/ops/_batch_norm_no_update_native.h>
+#include <ATen/ops/_batch_norm_with_update_native.h>
+#include <ATen/ops/_cast_Byte_native.h>
+#include <ATen/ops/_cast_Char_native.h>
+#include <ATen/ops/_cast_Double_native.h>
+#include <ATen/ops/_cast_Float_native.h>
+#include <ATen/ops/_cast_Half_native.h>
+#include <ATen/ops/_cast_Int_native.h>
+#include <ATen/ops/_cast_Long_native.h>
+#include <ATen/ops/_cast_Short_native.h>
+#include <ATen/ops/_cdist_backward_native.h>
+#include <ATen/ops/_cdist_forward_native.h>
+#include <ATen/ops/_cholesky_solve_helper_native.h>
+#include <ATen/ops/_choose_qparams_per_tensor_native.h>
+#include <ATen/ops/_chunk_cat_native.h>
+#include <ATen/ops/_coalesce_native.h>
+#include <ATen/ops/_coalesced_native.h>
+#include <ATen/ops/_compute_linear_combination_native.h>
+#include <ATen/ops/_conj_native.h>
+#include <ATen/ops/_conj_copy_native.h>
+#include <ATen/ops/_conj_physical_native.h>
+#include <ATen/ops/_conv_depthwise2d_native.h>
+#include <ATen/ops/_convert_indices_from_coo_to_csr_native.h>
+#include <ATen/ops/_convert_indices_from_csr_to_coo_native.h>
+#include <ATen/ops/_convert_weight_to_int4pack_native.h>
+#include <ATen/ops/_convert_weight_to_int4pack_for_cpu_native.h>
+#include <ATen/ops/_convolution_native.h>
+#include <ATen/ops/_convolution_double_backward_native.h>
+#include <ATen/ops/_convolution_mode_native.h>
+#include <ATen/ops/_copy_from_native.h>
+#include <ATen/ops/_copy_from_and_resize_native.h>
+#include <ATen/ops/_cslt_compress_native.h>
+#include <ATen/ops/_cslt_sparse_mm_native.h>
+#include <ATen/ops/_cslt_sparse_mm_search_native.h>
+#include <ATen/ops/_ctc_loss_native.h>
+#include <ATen/ops/_ctc_loss_backward_native.h>
+#include <ATen/ops/_cudnn_ctc_loss_native.h>
+#include <ATen/ops/_cudnn_init_dropout_state_native.h>
+#include <ATen/ops/_cudnn_rnn_native.h>
+#include <ATen/ops/_cudnn_rnn_backward_native.h>
+#include <ATen/ops/_cudnn_rnn_flatten_weight_native.h>
+#include <ATen/ops/_cufft_clear_plan_cache_native.h>
+#include <ATen/ops/_cufft_get_plan_cache_max_size_native.h>
+#include <ATen/ops/_cufft_get_plan_cache_size_native.h>
+#include <ATen/ops/_cufft_set_plan_cache_max_size_native.h>
+#include <ATen/ops/_cummax_helper_native.h>
+#include <ATen/ops/_cummin_helper_native.h>
+#include <ATen/ops/_debug_has_internal_overlap_native.h>
+#include <ATen/ops/_dimI_native.h>
+#include <ATen/ops/_dimV_native.h>
+#include <ATen/ops/_dim_arange_native.h>
+#include <ATen/ops/_dirichlet_grad_native.h>
+#include <ATen/ops/_efficient_attention_backward_native.h>
+#include <ATen/ops/_efficient_attention_forward_native.h>
+#include <ATen/ops/_efficientzerotensor_native.h>
+#include <ATen/ops/_embedding_bag_native.h>
+#include <ATen/ops/_embedding_bag_backward_native.h>
+#include <ATen/ops/_embedding_bag_dense_backward_native.h>
+#include <ATen/ops/_embedding_bag_forward_only_native.h>
+#include <ATen/ops/_embedding_bag_per_sample_weights_backward_native.h>
+#include <ATen/ops/_embedding_bag_sparse_backward_native.h>
+#include <ATen/ops/_empty_affine_quantized_native.h>
+#include <ATen/ops/_empty_per_channel_affine_quantized_native.h>
+#include <ATen/ops/_euclidean_dist_native.h>
+#include <ATen/ops/_fake_quantize_learnable_per_channel_affine_native.h>
+#include <ATen/ops/_fake_quantize_learnable_per_channel_affine_backward_native.h>
+#include <ATen/ops/_fake_quantize_learnable_per_tensor_affine_native.h>
+#include <ATen/ops/_fake_quantize_learnable_per_tensor_affine_backward_native.h>
+#include <ATen/ops/_fake_quantize_per_tensor_affine_cachemask_tensor_qparams_native.h>
+#include <ATen/ops/_fft_c2c_native.h>
+#include <ATen/ops/_fft_c2r_native.h>
+#include <ATen/ops/_fft_r2c_native.h>
+#include <ATen/ops/_fill_mem_eff_dropout_mask_native.h>
+#include <ATen/ops/_flash_attention_backward_native.h>
+#include <ATen/ops/_flash_attention_forward_native.h>
+#include <ATen/ops/_foobar_native.h>
+#include <ATen/ops/_foreach_abs_native.h>
+#include <ATen/ops/_foreach_acos_native.h>
+#include <ATen/ops/_foreach_add_native.h>
+#include <ATen/ops/_foreach_addcdiv_native.h>
+#include <ATen/ops/_foreach_addcmul_native.h>
+#include <ATen/ops/_foreach_asin_native.h>
+#include <ATen/ops/_foreach_atan_native.h>
+#include <ATen/ops/_foreach_ceil_native.h>
+#include <ATen/ops/_foreach_clamp_max_native.h>
+#include <ATen/ops/_foreach_clamp_min_native.h>
+#include <ATen/ops/_foreach_copy_native.h>
+#include <ATen/ops/_foreach_cos_native.h>
+#include <ATen/ops/_foreach_cosh_native.h>
+#include <ATen/ops/_foreach_div_native.h>
+#include <ATen/ops/_foreach_erf_native.h>
+#include <ATen/ops/_foreach_erfc_native.h>
+#include <ATen/ops/_foreach_exp_native.h>
+#include <ATen/ops/_foreach_expm1_native.h>
+#include <ATen/ops/_foreach_floor_native.h>
+#include <ATen/ops/_foreach_frac_native.h>
+#include <ATen/ops/_foreach_lerp_native.h>
+#include <ATen/ops/_foreach_lgamma_native.h>
+#include <ATen/ops/_foreach_log_native.h>
+#include <ATen/ops/_foreach_log10_native.h>
+#include <ATen/ops/_foreach_log1p_native.h>
+#include <ATen/ops/_foreach_log2_native.h>
+#include <ATen/ops/_foreach_max_native.h>
+#include <ATen/ops/_foreach_maximum_native.h>
+#include <ATen/ops/_foreach_minimum_native.h>
+#include <ATen/ops/_foreach_mul_native.h>
+#include <ATen/ops/_foreach_neg_native.h>
+#include <ATen/ops/_foreach_norm_native.h>
+#include <ATen/ops/_foreach_pow_native.h>
+#include <ATen/ops/_foreach_reciprocal_native.h>
+#include <ATen/ops/_foreach_round_native.h>
+#include <ATen/ops/_foreach_rsqrt_native.h>
+#include <ATen/ops/_foreach_sigmoid_native.h>
+#include <ATen/ops/_foreach_sign_native.h>
+#include <ATen/ops/_foreach_sin_native.h>
+#include <ATen/ops/_foreach_sinh_native.h>
+#include <ATen/ops/_foreach_sqrt_native.h>
+#include <ATen/ops/_foreach_sub_native.h>
+#include <ATen/ops/_foreach_tan_native.h>
+#include <ATen/ops/_foreach_tanh_native.h>
+#include <ATen/ops/_foreach_trunc_native.h>
+#include <ATen/ops/_foreach_zero_native.h>
+#include <ATen/ops/_functional_assert_async_native.h>
+#include <ATen/ops/_functional_assert_scalar_native.h>
+#include <ATen/ops/_functional_sym_constrain_range_native.h>
+#include <ATen/ops/_functional_sym_constrain_range_for_size_native.h>
+#include <ATen/ops/_fused_adagrad_native.h>
+#include <ATen/ops/_fused_adam_native.h>
+#include <ATen/ops/_fused_adamw_native.h>
+#include <ATen/ops/_fused_dropout_native.h>
+#include <ATen/ops/_fused_moving_avg_obs_fq_helper_native.h>
+#include <ATen/ops/_fused_sdp_choice_native.h>
+#include <ATen/ops/_fused_sgd_native.h>
+#include <ATen/ops/_fw_primal_native.h>
+#include <ATen/ops/_fw_primal_copy_native.h>
+#include <ATen/ops/_gather_sparse_backward_native.h>
+#include <ATen/ops/_grid_sampler_2d_cpu_fallback_native.h>
+#include <ATen/ops/_grid_sampler_2d_cpu_fallback_backward_native.h>
+#include <ATen/ops/_has_compatible_shallow_copy_type_native.h>
+#include <ATen/ops/_has_same_storage_numel_native.h>
+#include <ATen/ops/_histogramdd_bin_edges_native.h>
+#include <ATen/ops/_histogramdd_from_bin_cts_native.h>
+#include <ATen/ops/_histogramdd_from_bin_tensors_native.h>
+#include <ATen/ops/_index_put_impl_native.h>
+#include <ATen/ops/_indices_native.h>
+#include <ATen/ops/_indices_copy_native.h>
+#include <ATen/ops/_int_mm_native.h>
+#include <ATen/ops/_is_all_true_native.h>
+#include <ATen/ops/_is_any_true_native.h>
+#include <ATen/ops/_is_zerotensor_native.h>
+#include <ATen/ops/_jagged_to_padded_dense_forward_native.h>
+#include <ATen/ops/_lazy_clone_native.h>
+#include <ATen/ops/_linalg_check_errors_native.h>
+#include <ATen/ops/_linalg_det_native.h>
+#include <ATen/ops/_linalg_eigh_native.h>
+#include <ATen/ops/_linalg_eigvals_native.h>
+#include <ATen/ops/_linalg_slogdet_native.h>
+#include <ATen/ops/_linalg_solve_ex_native.h>
+#include <ATen/ops/_linalg_svd_native.h>
+#include <ATen/ops/_local_scalar_dense_native.h>
+#include <ATen/ops/_log_softmax_native.h>
+#include <ATen/ops/_log_softmax_backward_data_native.h>
+#include <ATen/ops/_logcumsumexp_native.h>
+#include <ATen/ops/_lstm_mps_native.h>
+#include <ATen/ops/_lu_with_info_native.h>
+#include <ATen/ops/_make_dep_token_native.h>
+#include <ATen/ops/_make_dual_native.h>
+#include <ATen/ops/_make_dual_copy_native.h>
+#include <ATen/ops/_make_per_channel_quantized_tensor_native.h>
+#include <ATen/ops/_make_per_tensor_quantized_tensor_native.h>
+#include <ATen/ops/_masked_scale_native.h>
+#include <ATen/ops/_masked_softmax_native.h>
+#include <ATen/ops/_masked_softmax_backward_native.h>
+#include <ATen/ops/_mixed_dtypes_linear_native.h>
+#include <ATen/ops/_mkldnn_reshape_native.h>
+#include <ATen/ops/_mkldnn_transpose_native.h>
+#include <ATen/ops/_mps_convolution_native.h>
+#include <ATen/ops/_mps_convolution_transpose_native.h>
+#include <ATen/ops/_native_batch_norm_legit_native.h>
+#include <ATen/ops/_native_batch_norm_legit_no_training_native.h>
+#include <ATen/ops/_native_multi_head_attention_native.h>
+#include <ATen/ops/_neg_view_native.h>
+#include <ATen/ops/_neg_view_copy_native.h>
+#include <ATen/ops/_nested_compute_contiguous_strides_offsets_native.h>
+#include <ATen/ops/_nested_from_padded_native.h>
+#include <ATen/ops/_nested_from_padded_and_nested_example_native.h>
+#include <ATen/ops/_nested_from_padded_tensor_native.h>
+#include <ATen/ops/_nested_get_jagged_dummy_native.h>
+#include <ATen/ops/_nested_get_lengths_native.h>
+#include <ATen/ops/_nested_get_max_seqlen_native.h>
+#include <ATen/ops/_nested_get_min_seqlen_native.h>
+#include <ATen/ops/_nested_get_offsets_native.h>
+#include <ATen/ops/_nested_get_ragged_idx_native.h>
+#include <ATen/ops/_nested_get_values_native.h>
+#include <ATen/ops/_nested_get_values_copy_native.h>
+#include <ATen/ops/_nested_select_backward_native.h>
+#include <ATen/ops/_nested_sum_backward_native.h>
+#include <ATen/ops/_nested_tensor_from_mask_native.h>
+#include <ATen/ops/_nested_tensor_from_mask_left_aligned_native.h>
+#include <ATen/ops/_nested_tensor_from_tensor_list_native.h>
+#include <ATen/ops/_nested_tensor_size_native.h>
+#include <ATen/ops/_nested_tensor_softmax_with_shape_native.h>
+#include <ATen/ops/_nested_tensor_storage_offsets_native.h>
+#include <ATen/ops/_nested_tensor_strides_native.h>
+#include <ATen/ops/_nested_view_from_buffer_native.h>
+#include <ATen/ops/_nested_view_from_buffer_copy_native.h>
+#include <ATen/ops/_nested_view_from_jagged_native.h>
+#include <ATen/ops/_nested_view_from_jagged_copy_native.h>
+#include <ATen/ops/_new_zeros_with_same_feature_meta_native.h>
+#include <ATen/ops/_nnpack_available_native.h>
+#include <ATen/ops/_nnpack_spatial_convolution_native.h>
+#include <ATen/ops/_nnz_native.h>
+#include <ATen/ops/_pack_padded_sequence_native.h>
+#include <ATen/ops/_pack_padded_sequence_backward_native.h>
+#include <ATen/ops/_pad_circular_native.h>
+#include <ATen/ops/_pad_enum_native.h>
+#include <ATen/ops/_pad_packed_sequence_native.h>
+#include <ATen/ops/_padded_dense_to_jagged_forward_native.h>
+#include <ATen/ops/_pdist_backward_native.h>
+#include <ATen/ops/_pdist_forward_native.h>
+#include <ATen/ops/_pin_memory_native.h>
+#include <ATen/ops/_prelu_kernel_native.h>
+#include <ATen/ops/_prelu_kernel_backward_native.h>
+#include <ATen/ops/_print_native.h>
+#include <ATen/ops/_propagate_xla_data_native.h>
+#include <ATen/ops/_remove_batch_dim_native.h>
+#include <ATen/ops/_reshape_alias_native.h>
+#include <ATen/ops/_reshape_alias_copy_native.h>
+#include <ATen/ops/_reshape_copy_native.h>
+#include <ATen/ops/_reshape_from_tensor_native.h>
+#include <ATen/ops/_resize_output_native.h>
+#include <ATen/ops/_rowwise_prune_native.h>
+#include <ATen/ops/_safe_softmax_native.h>
+#include <ATen/ops/_sample_dirichlet_native.h>
+#include <ATen/ops/_saturate_weight_to_fp16_native.h>
+#include <ATen/ops/_scaled_dot_product_attention_math_native.h>
+#include <ATen/ops/_scaled_dot_product_attention_math_for_mps_native.h>
+#include <ATen/ops/_scaled_dot_product_cudnn_attention_native.h>
+#include <ATen/ops/_scaled_dot_product_cudnn_attention_backward_native.h>
+#include <ATen/ops/_scaled_dot_product_efficient_attention_native.h>
+#include <ATen/ops/_scaled_dot_product_efficient_attention_backward_native.h>
+#include <ATen/ops/_scaled_dot_product_flash_attention_native.h>
+#include <ATen/ops/_scaled_dot_product_flash_attention_backward_native.h>
+#include <ATen/ops/_scaled_dot_product_flash_attention_for_cpu_native.h>
+#include <ATen/ops/_scaled_dot_product_flash_attention_for_cpu_backward_native.h>
+#include <ATen/ops/_scaled_dot_product_fused_attention_overrideable_native.h>
+#include <ATen/ops/_scaled_dot_product_fused_attention_overrideable_backward_native.h>
+#include <ATen/ops/_scaled_mm_native.h>
+#include <ATen/ops/_segment_reduce_backward_native.h>
+#include <ATen/ops/_shape_as_tensor_native.h>
+#include <ATen/ops/_slow_conv2d_backward_native.h>
+#include <ATen/ops/_slow_conv2d_forward_native.h>
+#include <ATen/ops/_sobol_engine_draw_native.h>
+#include <ATen/ops/_sobol_engine_ff_native.h>
+#include <ATen/ops/_sobol_engine_initialize_state_native.h>
+#include <ATen/ops/_sobol_engine_scramble_native.h>
+#include <ATen/ops/_softmax_native.h>
+#include <ATen/ops/_softmax_backward_data_native.h>
+#include <ATen/ops/_sparse_addmm_native.h>
+#include <ATen/ops/_sparse_broadcast_to_native.h>
+#include <ATen/ops/_sparse_broadcast_to_copy_native.h>
+#include <ATen/ops/_sparse_bsc_tensor_unsafe_native.h>
+#include <ATen/ops/_sparse_bsr_tensor_unsafe_native.h>
+#include <ATen/ops/_sparse_compressed_tensor_unsafe_native.h>
+#include <ATen/ops/_sparse_compressed_tensor_with_dims_native.h>
+#include <ATen/ops/_sparse_coo_tensor_unsafe_native.h>
+#include <ATen/ops/_sparse_coo_tensor_with_dims_native.h>
+#include <ATen/ops/_sparse_coo_tensor_with_dims_and_tensors_native.h>
+#include <ATen/ops/_sparse_csc_tensor_unsafe_native.h>
+#include <ATen/ops/_sparse_csr_prod_native.h>
+#include <ATen/ops/_sparse_csr_sum_native.h>
+#include <ATen/ops/_sparse_csr_tensor_unsafe_native.h>
+#include <ATen/ops/_sparse_log_softmax_native.h>
+#include <ATen/ops/_sparse_log_softmax_backward_data_native.h>
+#include <ATen/ops/_sparse_mask_projection_native.h>
+#include <ATen/ops/_sparse_mm_native.h>
+#include <ATen/ops/_sparse_mm_reduce_impl_native.h>
+#include <ATen/ops/_sparse_mm_reduce_impl_backward_native.h>
+#include <ATen/ops/_sparse_semi_structured_addmm_native.h>
+#include <ATen/ops/_sparse_semi_structured_apply_native.h>
+#include <ATen/ops/_sparse_semi_structured_apply_dense_native.h>
+#include <ATen/ops/_sparse_semi_structured_linear_native.h>
+#include <ATen/ops/_sparse_semi_structured_mm_native.h>
+#include <ATen/ops/_sparse_semi_structured_tile_native.h>
+#include <ATen/ops/_sparse_softmax_native.h>
+#include <ATen/ops/_sparse_softmax_backward_data_native.h>
+#include <ATen/ops/_sparse_sparse_matmul_native.h>
+#include <ATen/ops/_sparse_sum_native.h>
+#include <ATen/ops/_sparse_sum_backward_native.h>
+#include <ATen/ops/_spdiags_native.h>
+#include <ATen/ops/_spsolve_native.h>
+#include <ATen/ops/_stack_native.h>
+#include <ATen/ops/_standard_gamma_native.h>
+#include <ATen/ops/_standard_gamma_grad_native.h>
+#include <ATen/ops/_test_ambiguous_defaults_native.h>
+#include <ATen/ops/_test_autograd_multiple_dispatch_native.h>
+#include <ATen/ops/_test_autograd_multiple_dispatch_view_native.h>
+#include <ATen/ops/_test_autograd_multiple_dispatch_view_copy_native.h>
+#include <ATen/ops/_test_check_tensor_native.h>
+#include <ATen/ops/_test_functorch_fallback_native.h>
+#include <ATen/ops/_test_optional_filled_intlist_native.h>
+#include <ATen/ops/_test_optional_floatlist_native.h>
+#include <ATen/ops/_test_optional_intlist_native.h>
+#include <ATen/ops/_test_parallel_materialize_native.h>
+#include <ATen/ops/_test_serialization_subcmul_native.h>
+#include <ATen/ops/_test_string_default_native.h>
+#include <ATen/ops/_test_warn_in_autograd_native.h>
+#include <ATen/ops/_thnn_differentiable_gru_cell_backward_native.h>
+#include <ATen/ops/_thnn_differentiable_lstm_cell_backward_native.h>
+#include <ATen/ops/_thnn_fused_gru_cell_native.h>
+#include <ATen/ops/_thnn_fused_gru_cell_backward_native.h>
+#include <ATen/ops/_thnn_fused_lstm_cell_native.h>
+#include <ATen/ops/_thnn_fused_lstm_cell_backward_native.h>
+#include <ATen/ops/_thnn_fused_lstm_cell_backward_impl_native.h>
+#include <ATen/ops/_to_copy_native.h>
+#include <ATen/ops/_to_cpu_native.h>
+#include <ATen/ops/_to_dense_native.h>
+#include <ATen/ops/_to_sparse_native.h>
+#include <ATen/ops/_to_sparse_bsc_native.h>
+#include <ATen/ops/_to_sparse_bsr_native.h>
+#include <ATen/ops/_to_sparse_csc_native.h>
+#include <ATen/ops/_to_sparse_csr_native.h>
+#include <ATen/ops/_to_sparse_semi_structured_native.h>
+#include <ATen/ops/_transform_bias_rescale_qkv_native.h>
+#include <ATen/ops/_transformer_encoder_layer_fwd_native.h>
+#include <ATen/ops/_trilinear_native.h>
+#include <ATen/ops/_triton_multi_head_attention_native.h>
+#include <ATen/ops/_triton_scaled_dot_attention_native.h>
+#include <ATen/ops/_unique_native.h>
+#include <ATen/ops/_unique2_native.h>
+#include <ATen/ops/_unpack_dual_native.h>
+#include <ATen/ops/_unsafe_index_native.h>
+#include <ATen/ops/_unsafe_index_put_native.h>
+#include <ATen/ops/_unsafe_masked_index_native.h>
+#include <ATen/ops/_unsafe_masked_index_put_accumulate_native.h>
+#include <ATen/ops/_unsafe_view_native.h>
+#include <ATen/ops/_upsample_bicubic2d_aa_native.h>
+#include <ATen/ops/_upsample_bicubic2d_aa_backward_native.h>
+#include <ATen/ops/_upsample_bilinear2d_aa_native.h>
+#include <ATen/ops/_upsample_bilinear2d_aa_backward_native.h>
+#include <ATen/ops/_upsample_nearest_exact1d_native.h>
+#include <ATen/ops/_upsample_nearest_exact1d_backward_native.h>
+#include <ATen/ops/_upsample_nearest_exact2d_native.h>
+#include <ATen/ops/_upsample_nearest_exact2d_backward_native.h>
+#include <ATen/ops/_upsample_nearest_exact3d_native.h>
+#include <ATen/ops/_upsample_nearest_exact3d_backward_native.h>
+#include <ATen/ops/_use_cudnn_ctc_loss_native.h>
+#include <ATen/ops/_use_cudnn_rnn_flatten_weight_native.h>
+#include <ATen/ops/_validate_compressed_sparse_indices_native.h>
+#include <ATen/ops/_validate_sparse_bsc_tensor_args_native.h>
+#include <ATen/ops/_validate_sparse_bsr_tensor_args_native.h>
+#include <ATen/ops/_validate_sparse_compressed_tensor_args_native.h>
+#include <ATen/ops/_validate_sparse_coo_tensor_args_native.h>
+#include <ATen/ops/_validate_sparse_csc_tensor_args_native.h>
+#include <ATen/ops/_validate_sparse_csr_tensor_args_native.h>
+#include <ATen/ops/_values_native.h>
+#include <ATen/ops/_values_copy_native.h>
+#include <ATen/ops/_version_native.h>
+#include <ATen/ops/_weight_int4pack_mm_native.h>
+#include <ATen/ops/_weight_int4pack_mm_for_cpu_native.h>
+#include <ATen/ops/_weight_int8pack_mm_native.h>
+#include <ATen/ops/_weight_norm_native.h>
+#include <ATen/ops/_weight_norm_differentiable_backward_native.h>
+#include <ATen/ops/_weight_norm_interface_native.h>
+#include <ATen/ops/_weight_norm_interface_backward_native.h>
+#include <ATen/ops/_wrapped_linear_prepack_native.h>
+#include <ATen/ops/_wrapped_quantized_linear_prepacked_native.h>
+#include <ATen/ops/abs_native.h>
+#include <ATen/ops/absolute_native.h>
+#include <ATen/ops/acos_native.h>
+#include <ATen/ops/acosh_native.h>
+#include <ATen/ops/adaptive_avg_pool1d_native.h>
+#include <ATen/ops/adaptive_avg_pool2d_native.h>
+#include <ATen/ops/adaptive_avg_pool3d_native.h>
+#include <ATen/ops/adaptive_avg_pool3d_backward_native.h>
+#include <ATen/ops/adaptive_max_pool1d_native.h>
+#include <ATen/ops/adaptive_max_pool2d_native.h>
+#include <ATen/ops/adaptive_max_pool2d_backward_native.h>
+#include <ATen/ops/adaptive_max_pool3d_native.h>
+#include <ATen/ops/adaptive_max_pool3d_backward_native.h>
+#include <ATen/ops/add_native.h>
+#include <ATen/ops/addbmm_native.h>
+#include <ATen/ops/addcdiv_native.h>
+#include <ATen/ops/addcmul_native.h>
+#include <ATen/ops/addmm_native.h>
+#include <ATen/ops/addmv_native.h>
+#include <ATen/ops/addr_native.h>
+#include <ATen/ops/adjoint_native.h>
+#include <ATen/ops/affine_grid_generator_native.h>
+#include <ATen/ops/affine_grid_generator_backward_native.h>
+#include <ATen/ops/alias_native.h>
+#include <ATen/ops/alias_copy_native.h>
+#include <ATen/ops/align_as_native.h>
+#include <ATen/ops/align_tensors_native.h>
+#include <ATen/ops/align_to_native.h>
+#include <ATen/ops/all_native.h>
+#include <ATen/ops/allclose_native.h>
+#include <ATen/ops/alpha_dropout_native.h>
+#include <ATen/ops/amax_native.h>
+#include <ATen/ops/amin_native.h>
+#include <ATen/ops/aminmax_native.h>
+#include <ATen/ops/and_native.h>
+#include <ATen/ops/angle_native.h>
+#include <ATen/ops/any_native.h>
+#include <ATen/ops/arange_native.h>
+#include <ATen/ops/arccos_native.h>
+#include <ATen/ops/arccosh_native.h>
+#include <ATen/ops/arcsin_native.h>
+#include <ATen/ops/arcsinh_native.h>
+#include <ATen/ops/arctan_native.h>
+#include <ATen/ops/arctan2_native.h>
+#include <ATen/ops/arctanh_native.h>
+#include <ATen/ops/argmax_native.h>
+#include <ATen/ops/argmin_native.h>
+#include <ATen/ops/argsort_native.h>
+#include <ATen/ops/argwhere_native.h>
+#include <ATen/ops/as_strided_native.h>
+#include <ATen/ops/as_strided_copy_native.h>
+#include <ATen/ops/as_strided_scatter_native.h>
+#include <ATen/ops/asin_native.h>
+#include <ATen/ops/asinh_native.h>
+#include <ATen/ops/atan_native.h>
+#include <ATen/ops/atan2_native.h>
+#include <ATen/ops/atanh_native.h>
+#include <ATen/ops/atleast_1d_native.h>
+#include <ATen/ops/atleast_2d_native.h>
+#include <ATen/ops/atleast_3d_native.h>
+#include <ATen/ops/avg_pool1d_native.h>
+#include <ATen/ops/avg_pool2d_native.h>
+#include <ATen/ops/avg_pool2d_backward_native.h>
+#include <ATen/ops/avg_pool3d_native.h>
+#include <ATen/ops/avg_pool3d_backward_native.h>
+#include <ATen/ops/baddbmm_native.h>
+#include <ATen/ops/bartlett_window_native.h>
+#include <ATen/ops/batch_norm_native.h>
+#include <ATen/ops/batch_norm_backward_native.h>
+#include <ATen/ops/batch_norm_backward_elemt_native.h>
+#include <ATen/ops/batch_norm_backward_reduce_native.h>
+#include <ATen/ops/batch_norm_elemt_native.h>
+#include <ATen/ops/batch_norm_gather_stats_native.h>
+#include <ATen/ops/batch_norm_gather_stats_with_counts_native.h>
+#include <ATen/ops/batch_norm_stats_native.h>
+#include <ATen/ops/batch_norm_update_stats_native.h>
+#include <ATen/ops/bernoulli_native.h>
+#include <ATen/ops/bilinear_native.h>
+#include <ATen/ops/binary_cross_entropy_native.h>
+#include <ATen/ops/binary_cross_entropy_backward_native.h>
+#include <ATen/ops/binary_cross_entropy_with_logits_native.h>
+#include <ATen/ops/bincount_native.h>
+#include <ATen/ops/binomial_native.h>
+#include <ATen/ops/bitwise_and_native.h>
+#include <ATen/ops/bitwise_left_shift_native.h>
+#include <ATen/ops/bitwise_not_native.h>
+#include <ATen/ops/bitwise_or_native.h>
+#include <ATen/ops/bitwise_right_shift_native.h>
+#include <ATen/ops/bitwise_xor_native.h>
+#include <ATen/ops/blackman_window_native.h>
+#include <ATen/ops/block_diag_native.h>
+#include <ATen/ops/bmm_native.h>
+#include <ATen/ops/broadcast_tensors_native.h>
+#include <ATen/ops/broadcast_to_native.h>
+#include <ATen/ops/bucketize_native.h>
+#include <ATen/ops/can_cast_native.h>
+#include <ATen/ops/cartesian_prod_native.h>
+#include <ATen/ops/cat_native.h>
+#include <ATen/ops/cauchy_native.h>
+#include <ATen/ops/ccol_indices_native.h>
+#include <ATen/ops/ccol_indices_copy_native.h>
+#include <ATen/ops/cdist_native.h>
+#include <ATen/ops/ceil_native.h>
+#include <ATen/ops/celu_native.h>
+#include <ATen/ops/chain_matmul_native.h>
+#include <ATen/ops/chalf_native.h>
+#include <ATen/ops/channel_shuffle_native.h>
+#include <ATen/ops/cholesky_native.h>
+#include <ATen/ops/cholesky_inverse_native.h>
+#include <ATen/ops/cholesky_solve_native.h>
+#include <ATen/ops/choose_qparams_optimized_native.h>
+#include <ATen/ops/chunk_native.h>
+#include <ATen/ops/clamp_native.h>
+#include <ATen/ops/clamp_max_native.h>
+#include <ATen/ops/clamp_min_native.h>
+#include <ATen/ops/clip_native.h>
+#include <ATen/ops/clone_native.h>
+#include <ATen/ops/coalesce_native.h>
+#include <ATen/ops/col2im_native.h>
+#include <ATen/ops/col_indices_native.h>
+#include <ATen/ops/col_indices_copy_native.h>
+#include <ATen/ops/column_stack_native.h>
+#include <ATen/ops/combinations_native.h>
+#include <ATen/ops/complex_native.h>
+#include <ATen/ops/concat_native.h>
+#include <ATen/ops/concatenate_native.h>
+#include <ATen/ops/conj_native.h>
+#include <ATen/ops/conj_physical_native.h>
+#include <ATen/ops/constant_pad_nd_native.h>
+#include <ATen/ops/contiguous_native.h>
+#include <ATen/ops/conv1d_native.h>
+#include <ATen/ops/conv2d_native.h>
+#include <ATen/ops/conv3d_native.h>
+#include <ATen/ops/conv_depthwise3d_native.h>
+#include <ATen/ops/conv_tbc_native.h>
+#include <ATen/ops/conv_tbc_backward_native.h>
+#include <ATen/ops/conv_transpose1d_native.h>
+#include <ATen/ops/conv_transpose2d_native.h>
+#include <ATen/ops/conv_transpose3d_native.h>
+#include <ATen/ops/convolution_native.h>
+#include <ATen/ops/convolution_backward_native.h>
+#include <ATen/ops/convolution_backward_overrideable_native.h>
+#include <ATen/ops/convolution_overrideable_native.h>
+#include <ATen/ops/copy_native.h>
+#include <ATen/ops/copy_sparse_to_sparse_native.h>
+#include <ATen/ops/copysign_native.h>
+#include <ATen/ops/corrcoef_native.h>
+#include <ATen/ops/cos_native.h>
+#include <ATen/ops/cosh_native.h>
+#include <ATen/ops/cosine_embedding_loss_native.h>
+#include <ATen/ops/cosine_similarity_native.h>
+#include <ATen/ops/count_nonzero_native.h>
+#include <ATen/ops/cov_native.h>
+#include <ATen/ops/cross_native.h>
+#include <ATen/ops/cross_entropy_loss_native.h>
+#include <ATen/ops/crow_indices_native.h>
+#include <ATen/ops/crow_indices_copy_native.h>
+#include <ATen/ops/ctc_loss_native.h>
+#include <ATen/ops/cudnn_affine_grid_generator_native.h>
+#include <ATen/ops/cudnn_affine_grid_generator_backward_native.h>
+#include <ATen/ops/cudnn_batch_norm_native.h>
+#include <ATen/ops/cudnn_batch_norm_backward_native.h>
+#include <ATen/ops/cudnn_convolution_native.h>
+#include <ATen/ops/cudnn_convolution_add_relu_native.h>
+#include <ATen/ops/cudnn_convolution_relu_native.h>
+#include <ATen/ops/cudnn_convolution_transpose_native.h>
+#include <ATen/ops/cudnn_grid_sampler_native.h>
+#include <ATen/ops/cudnn_grid_sampler_backward_native.h>
+#include <ATen/ops/cudnn_is_acceptable_native.h>
+#include <ATen/ops/cummax_native.h>
+#include <ATen/ops/cummaxmin_backward_native.h>
+#include <ATen/ops/cummin_native.h>
+#include <ATen/ops/cumprod_native.h>
+#include <ATen/ops/cumprod_backward_native.h>
+#include <ATen/ops/cumsum_native.h>
+#include <ATen/ops/cumulative_trapezoid_native.h>
+#include <ATen/ops/data_native.h>
+#include <ATen/ops/deg2rad_native.h>
+#include <ATen/ops/dense_dim_native.h>
+#include <ATen/ops/dequantize_native.h>
+#include <ATen/ops/det_native.h>
+#include <ATen/ops/detach_native.h>
+#include <ATen/ops/detach_copy_native.h>
+#include <ATen/ops/diag_native.h>
+#include <ATen/ops/diag_embed_native.h>
+#include <ATen/ops/diagflat_native.h>
+#include <ATen/ops/diagonal_native.h>
+#include <ATen/ops/diagonal_backward_native.h>
+#include <ATen/ops/diagonal_copy_native.h>
+#include <ATen/ops/diagonal_scatter_native.h>
+#include <ATen/ops/diff_native.h>
+#include <ATen/ops/digamma_native.h>
+#include <ATen/ops/dist_native.h>
+#include <ATen/ops/div_native.h>
+#include <ATen/ops/divide_native.h>
+#include <ATen/ops/dot_native.h>
+#include <ATen/ops/dropout_native.h>
+#include <ATen/ops/dsplit_native.h>
+#include <ATen/ops/dstack_native.h>
+#include <ATen/ops/einsum_native.h>
+#include <ATen/ops/elu_native.h>
+#include <ATen/ops/elu_backward_native.h>
+#include <ATen/ops/embedding_native.h>
+#include <ATen/ops/embedding_backward_native.h>
+#include <ATen/ops/embedding_bag_native.h>
+#include <ATen/ops/embedding_dense_backward_native.h>
+#include <ATen/ops/embedding_renorm_native.h>
+#include <ATen/ops/embedding_sparse_backward_native.h>
+#include <ATen/ops/empty_native.h>
+#include <ATen/ops/empty_like_native.h>
+#include <ATen/ops/empty_permuted_native.h>
+#include <ATen/ops/empty_quantized_native.h>
+#include <ATen/ops/empty_strided_native.h>
+#include <ATen/ops/eq_native.h>
+#include <ATen/ops/equal_native.h>
+#include <ATen/ops/erf_native.h>
+#include <ATen/ops/erfc_native.h>
+#include <ATen/ops/erfinv_native.h>
+#include <ATen/ops/exp_native.h>
+#include <ATen/ops/exp2_native.h>
+#include <ATen/ops/expand_native.h>
+#include <ATen/ops/expand_as_native.h>
+#include <ATen/ops/expand_copy_native.h>
+#include <ATen/ops/expm1_native.h>
+#include <ATen/ops/exponential_native.h>
+#include <ATen/ops/eye_native.h>
+#include <ATen/ops/fake_quantize_per_channel_affine_native.h>
+#include <ATen/ops/fake_quantize_per_channel_affine_cachemask_native.h>
+#include <ATen/ops/fake_quantize_per_channel_affine_cachemask_backward_native.h>
+#include <ATen/ops/fake_quantize_per_tensor_affine_native.h>
+#include <ATen/ops/fake_quantize_per_tensor_affine_cachemask_native.h>
+#include <ATen/ops/fake_quantize_per_tensor_affine_cachemask_backward_native.h>
+#include <ATen/ops/fbgemm_linear_fp16_weight_native.h>
+#include <ATen/ops/fbgemm_linear_fp16_weight_fp32_activation_native.h>
+#include <ATen/ops/fbgemm_linear_int8_weight_native.h>
+#include <ATen/ops/fbgemm_linear_int8_weight_fp32_activation_native.h>
+#include <ATen/ops/fbgemm_linear_quantize_weight_native.h>
+#include <ATen/ops/fbgemm_pack_gemm_matrix_fp16_native.h>
+#include <ATen/ops/fbgemm_pack_quantized_matrix_native.h>
+#include <ATen/ops/feature_alpha_dropout_native.h>
+#include <ATen/ops/feature_dropout_native.h>
+#include <ATen/ops/fft_fft_native.h>
+#include <ATen/ops/fft_fft2_native.h>
+#include <ATen/ops/fft_fftfreq_native.h>
+#include <ATen/ops/fft_fftn_native.h>
+#include <ATen/ops/fft_fftshift_native.h>
+#include <ATen/ops/fft_hfft_native.h>
+#include <ATen/ops/fft_hfft2_native.h>
+#include <ATen/ops/fft_hfftn_native.h>
+#include <ATen/ops/fft_ifft_native.h>
+#include <ATen/ops/fft_ifft2_native.h>
+#include <ATen/ops/fft_ifftn_native.h>
+#include <ATen/ops/fft_ifftshift_native.h>
+#include <ATen/ops/fft_ihfft_native.h>
+#include <ATen/ops/fft_ihfft2_native.h>
+#include <ATen/ops/fft_ihfftn_native.h>
+#include <ATen/ops/fft_irfft_native.h>
+#include <ATen/ops/fft_irfft2_native.h>
+#include <ATen/ops/fft_irfftn_native.h>
+#include <ATen/ops/fft_rfft_native.h>
+#include <ATen/ops/fft_rfft2_native.h>
+#include <ATen/ops/fft_rfftfreq_native.h>
+#include <ATen/ops/fft_rfftn_native.h>
+#include <ATen/ops/fill_native.h>
+#include <ATen/ops/fill_diagonal_native.h>
+#include <ATen/ops/fix_native.h>
+#include <ATen/ops/flatten_native.h>
+#include <ATen/ops/flatten_dense_tensors_native.h>
+#include <ATen/ops/flip_native.h>
+#include <ATen/ops/fliplr_native.h>
+#include <ATen/ops/flipud_native.h>
+#include <ATen/ops/float_power_native.h>
+#include <ATen/ops/floor_native.h>
+#include <ATen/ops/floor_divide_native.h>
+#include <ATen/ops/fmax_native.h>
+#include <ATen/ops/fmin_native.h>
+#include <ATen/ops/fmod_native.h>
+#include <ATen/ops/frac_native.h>
+#include <ATen/ops/fractional_max_pool2d_native.h>
+#include <ATen/ops/fractional_max_pool2d_backward_native.h>
+#include <ATen/ops/fractional_max_pool3d_native.h>
+#include <ATen/ops/fractional_max_pool3d_backward_native.h>
+#include <ATen/ops/frexp_native.h>
+#include <ATen/ops/frobenius_norm_native.h>
+#include <ATen/ops/from_file_native.h>
+#include <ATen/ops/full_native.h>
+#include <ATen/ops/full_like_native.h>
+#include <ATen/ops/fused_moving_avg_obs_fake_quant_native.h>
+#include <ATen/ops/gather_native.h>
+#include <ATen/ops/gather_backward_native.h>
+#include <ATen/ops/gcd_native.h>
+#include <ATen/ops/ge_native.h>
+#include <ATen/ops/gelu_native.h>
+#include <ATen/ops/gelu_backward_native.h>
+#include <ATen/ops/geometric_native.h>
+#include <ATen/ops/geqrf_native.h>
+#include <ATen/ops/ger_native.h>
+#include <ATen/ops/glu_native.h>
+#include <ATen/ops/glu_backward_native.h>
+#include <ATen/ops/glu_backward_jvp_native.h>
+#include <ATen/ops/glu_jvp_native.h>
+#include <ATen/ops/gradient_native.h>
+#include <ATen/ops/greater_native.h>
+#include <ATen/ops/greater_equal_native.h>
+#include <ATen/ops/grid_sampler_native.h>
+#include <ATen/ops/grid_sampler_2d_native.h>
+#include <ATen/ops/grid_sampler_2d_backward_native.h>
+#include <ATen/ops/grid_sampler_3d_native.h>
+#include <ATen/ops/grid_sampler_3d_backward_native.h>
+#include <ATen/ops/group_norm_native.h>
+#include <ATen/ops/gru_native.h>
+#include <ATen/ops/gru_cell_native.h>
+#include <ATen/ops/gt_native.h>
+#include <ATen/ops/hamming_window_native.h>
+#include <ATen/ops/hann_window_native.h>
+#include <ATen/ops/hardshrink_native.h>
+#include <ATen/ops/hardshrink_backward_native.h>
+#include <ATen/ops/hardsigmoid_native.h>
+#include <ATen/ops/hardsigmoid_backward_native.h>
+#include <ATen/ops/hardswish_native.h>
+#include <ATen/ops/hardswish_backward_native.h>
+#include <ATen/ops/hardtanh_native.h>
+#include <ATen/ops/hardtanh_backward_native.h>
+#include <ATen/ops/heaviside_native.h>
+#include <ATen/ops/hinge_embedding_loss_native.h>
+#include <ATen/ops/histc_native.h>
+#include <ATen/ops/histogram_native.h>
+#include <ATen/ops/histogramdd_native.h>
+#include <ATen/ops/hsplit_native.h>
+#include <ATen/ops/hspmm_native.h>
+#include <ATen/ops/hstack_native.h>
+#include <ATen/ops/huber_loss_native.h>
+#include <ATen/ops/huber_loss_backward_native.h>
+#include <ATen/ops/hypot_native.h>
+#include <ATen/ops/i0_native.h>
+#include <ATen/ops/igamma_native.h>
+#include <ATen/ops/igammac_native.h>
+#include <ATen/ops/im2col_native.h>
+#include <ATen/ops/imag_native.h>
+#include <ATen/ops/index_native.h>
+#include <ATen/ops/index_add_native.h>
+#include <ATen/ops/index_copy_native.h>
+#include <ATen/ops/index_fill_native.h>
+#include <ATen/ops/index_put_native.h>
+#include <ATen/ops/index_reduce_native.h>
+#include <ATen/ops/index_select_native.h>
+#include <ATen/ops/index_select_backward_native.h>
+#include <ATen/ops/indices_native.h>
+#include <ATen/ops/indices_copy_native.h>
+#include <ATen/ops/infinitely_differentiable_gelu_backward_native.h>
+#include <ATen/ops/inner_native.h>
+#include <ATen/ops/instance_norm_native.h>
+#include <ATen/ops/int_repr_native.h>
+#include <ATen/ops/inverse_native.h>
+#include <ATen/ops/is_coalesced_native.h>
+#include <ATen/ops/is_complex_native.h>
+#include <ATen/ops/is_conj_native.h>
+#include <ATen/ops/is_distributed_native.h>
+#include <ATen/ops/is_floating_point_native.h>
+#include <ATen/ops/is_inference_native.h>
+#include <ATen/ops/is_leaf_native.h>
+#include <ATen/ops/is_neg_native.h>
+#include <ATen/ops/is_nonzero_native.h>
+#include <ATen/ops/is_pinned_native.h>
+#include <ATen/ops/is_same_size_native.h>
+#include <ATen/ops/is_set_to_native.h>
+#include <ATen/ops/is_signed_native.h>
+#include <ATen/ops/is_vulkan_available_native.h>
+#include <ATen/ops/isclose_native.h>
+#include <ATen/ops/isfinite_native.h>
+#include <ATen/ops/isin_native.h>
+#include <ATen/ops/isinf_native.h>
+#include <ATen/ops/isnan_native.h>
+#include <ATen/ops/isneginf_native.h>
+#include <ATen/ops/isposinf_native.h>
+#include <ATen/ops/isreal_native.h>
+#include <ATen/ops/istft_native.h>
+#include <ATen/ops/item_native.h>
+#include <ATen/ops/kaiser_window_native.h>
+#include <ATen/ops/kl_div_native.h>
+#include <ATen/ops/kron_native.h>
+#include <ATen/ops/kthvalue_native.h>
+#include <ATen/ops/l1_loss_native.h>
+#include <ATen/ops/layer_norm_native.h>
+#include <ATen/ops/lcm_native.h>
+#include <ATen/ops/ldexp_native.h>
+#include <ATen/ops/le_native.h>
+#include <ATen/ops/leaky_relu_native.h>
+#include <ATen/ops/leaky_relu_backward_native.h>
+#include <ATen/ops/lerp_native.h>
+#include <ATen/ops/less_native.h>
+#include <ATen/ops/less_equal_native.h>
+#include <ATen/ops/lgamma_native.h>
+#include <ATen/ops/lift_native.h>
+#include <ATen/ops/lift_fresh_native.h>
+#include <ATen/ops/lift_fresh_copy_native.h>
+#include <ATen/ops/linalg_cholesky_native.h>
+#include <ATen/ops/linalg_cholesky_ex_native.h>
+#include <ATen/ops/linalg_cond_native.h>
+#include <ATen/ops/linalg_cross_native.h>
+#include <ATen/ops/linalg_det_native.h>
+#include <ATen/ops/linalg_diagonal_native.h>
+#include <ATen/ops/linalg_eig_native.h>
+#include <ATen/ops/linalg_eigh_native.h>
+#include <ATen/ops/linalg_eigvals_native.h>
+#include <ATen/ops/linalg_eigvalsh_native.h>
+#include <ATen/ops/linalg_householder_product_native.h>
+#include <ATen/ops/linalg_inv_native.h>
+#include <ATen/ops/linalg_inv_ex_native.h>
+#include <ATen/ops/linalg_ldl_factor_native.h>
+#include <ATen/ops/linalg_ldl_factor_ex_native.h>
+#include <ATen/ops/linalg_ldl_solve_native.h>
+#include <ATen/ops/linalg_lstsq_native.h>
+#include <ATen/ops/linalg_lu_native.h>
+#include <ATen/ops/linalg_lu_factor_native.h>
+#include <ATen/ops/linalg_lu_factor_ex_native.h>
+#include <ATen/ops/linalg_lu_solve_native.h>
+#include <ATen/ops/linalg_matmul_native.h>
+#include <ATen/ops/linalg_matrix_exp_native.h>
+#include <ATen/ops/linalg_matrix_norm_native.h>
+#include <ATen/ops/linalg_matrix_power_native.h>
+#include <ATen/ops/linalg_matrix_rank_native.h>
+#include <ATen/ops/linalg_multi_dot_native.h>
+#include <ATen/ops/linalg_norm_native.h>
+#include <ATen/ops/linalg_pinv_native.h>
+#include <ATen/ops/linalg_qr_native.h>
+#include <ATen/ops/linalg_slogdet_native.h>
+#include <ATen/ops/linalg_solve_native.h>
+#include <ATen/ops/linalg_solve_ex_native.h>
+#include <ATen/ops/linalg_solve_triangular_native.h>
+#include <ATen/ops/linalg_svd_native.h>
+#include <ATen/ops/linalg_svdvals_native.h>
+#include <ATen/ops/linalg_tensorinv_native.h>
+#include <ATen/ops/linalg_tensorsolve_native.h>
+#include <ATen/ops/linalg_vander_native.h>
+#include <ATen/ops/linalg_vecdot_native.h>
+#include <ATen/ops/linalg_vector_norm_native.h>
+#include <ATen/ops/linear_native.h>
+#include <ATen/ops/linear_backward_native.h>
+#include <ATen/ops/linspace_native.h>
+#include <ATen/ops/log_native.h>
+#include <ATen/ops/log10_native.h>
+#include <ATen/ops/log1p_native.h>
+#include <ATen/ops/log2_native.h>
+#include <ATen/ops/log_normal_native.h>
+#include <ATen/ops/log_sigmoid_native.h>
+#include <ATen/ops/log_sigmoid_backward_native.h>
+#include <ATen/ops/log_sigmoid_forward_native.h>
+#include <ATen/ops/log_softmax_native.h>
+#include <ATen/ops/logaddexp_native.h>
+#include <ATen/ops/logaddexp2_native.h>
+#include <ATen/ops/logcumsumexp_native.h>
+#include <ATen/ops/logdet_native.h>
+#include <ATen/ops/logical_and_native.h>
+#include <ATen/ops/logical_not_native.h>
+#include <ATen/ops/logical_or_native.h>
+#include <ATen/ops/logical_xor_native.h>
+#include <ATen/ops/logit_native.h>
+#include <ATen/ops/logit_backward_native.h>
+#include <ATen/ops/logspace_native.h>
+#include <ATen/ops/logsumexp_native.h>
+#include <ATen/ops/lshift_native.h>
+#include <ATen/ops/lstm_native.h>
+#include <ATen/ops/lstm_cell_native.h>
+#include <ATen/ops/lstm_mps_backward_native.h>
+#include <ATen/ops/lt_native.h>
+#include <ATen/ops/lu_solve_native.h>
+#include <ATen/ops/lu_unpack_native.h>
+#include <ATen/ops/mH_native.h>
+#include <ATen/ops/mT_native.h>
+#include <ATen/ops/margin_ranking_loss_native.h>
+#include <ATen/ops/masked_fill_native.h>
+#include <ATen/ops/masked_scatter_native.h>
+#include <ATen/ops/masked_scatter_backward_native.h>
+#include <ATen/ops/masked_select_native.h>
+#include <ATen/ops/masked_select_backward_native.h>
+#include <ATen/ops/matmul_native.h>
+#include <ATen/ops/matmul_backward_native.h>
+#include <ATen/ops/matrix_H_native.h>
+#include <ATen/ops/matrix_exp_native.h>
+#include <ATen/ops/matrix_exp_backward_native.h>
+#include <ATen/ops/matrix_power_native.h>
+#include <ATen/ops/max_native.h>
+#include <ATen/ops/max_pool1d_native.h>
+#include <ATen/ops/max_pool1d_with_indices_native.h>
+#include <ATen/ops/max_pool2d_native.h>
+#include <ATen/ops/max_pool2d_backward_native.h>
+#include <ATen/ops/max_pool2d_with_indices_native.h>
+#include <ATen/ops/max_pool2d_with_indices_backward_native.h>
+#include <ATen/ops/max_pool3d_native.h>
+#include <ATen/ops/max_pool3d_with_indices_native.h>
+#include <ATen/ops/max_pool3d_with_indices_backward_native.h>
+#include <ATen/ops/max_unpool2d_native.h>
+#include <ATen/ops/max_unpool3d_native.h>
+#include <ATen/ops/maximum_native.h>
+#include <ATen/ops/mean_native.h>
+#include <ATen/ops/median_native.h>
+#include <ATen/ops/meshgrid_native.h>
+#include <ATen/ops/min_native.h>
+#include <ATen/ops/minimum_native.h>
+#include <ATen/ops/miopen_batch_norm_native.h>
+#include <ATen/ops/miopen_batch_norm_backward_native.h>
+#include <ATen/ops/miopen_convolution_native.h>
+#include <ATen/ops/miopen_convolution_add_relu_native.h>
+#include <ATen/ops/miopen_convolution_relu_native.h>
+#include <ATen/ops/miopen_convolution_transpose_native.h>
+#include <ATen/ops/miopen_depthwise_convolution_native.h>
+#include <ATen/ops/miopen_rnn_native.h>
+#include <ATen/ops/miopen_rnn_backward_native.h>
+#include <ATen/ops/mish_native.h>
+#include <ATen/ops/mish_backward_native.h>
+#include <ATen/ops/mkldnn_adaptive_avg_pool2d_native.h>
+#include <ATen/ops/mkldnn_adaptive_avg_pool2d_backward_native.h>
+#include <ATen/ops/mkldnn_convolution_native.h>
+#include <ATen/ops/mkldnn_linear_native.h>
+#include <ATen/ops/mkldnn_linear_backward_native.h>
+#include <ATen/ops/mkldnn_linear_backward_input_native.h>
+#include <ATen/ops/mkldnn_linear_backward_weights_native.h>
+#include <ATen/ops/mkldnn_max_pool2d_native.h>
+#include <ATen/ops/mkldnn_max_pool2d_backward_native.h>
+#include <ATen/ops/mkldnn_max_pool3d_native.h>
+#include <ATen/ops/mkldnn_max_pool3d_backward_native.h>
+#include <ATen/ops/mkldnn_reorder_conv2d_weight_native.h>
+#include <ATen/ops/mkldnn_reorder_conv3d_weight_native.h>
+#include <ATen/ops/mkldnn_rnn_layer_native.h>
+#include <ATen/ops/mkldnn_rnn_layer_backward_native.h>
+#include <ATen/ops/mm_native.h>
+#include <ATen/ops/mode_native.h>
+#include <ATen/ops/moveaxis_native.h>
+#include <ATen/ops/movedim_native.h>
+#include <ATen/ops/mps_convolution_backward_native.h>
+#include <ATen/ops/mps_convolution_transpose_backward_native.h>
+#include <ATen/ops/mse_loss_native.h>
+#include <ATen/ops/mse_loss_backward_native.h>
+#include <ATen/ops/msort_native.h>
+#include <ATen/ops/mul_native.h>
+#include <ATen/ops/multi_margin_loss_native.h>
+#include <ATen/ops/multi_margin_loss_backward_native.h>
+#include <ATen/ops/multilabel_margin_loss_native.h>
+#include <ATen/ops/multilabel_margin_loss_backward_native.h>
+#include <ATen/ops/multilabel_margin_loss_forward_native.h>
+#include <ATen/ops/multinomial_native.h>
+#include <ATen/ops/multiply_native.h>
+#include <ATen/ops/mv_native.h>
+#include <ATen/ops/mvlgamma_native.h>
+#include <ATen/ops/nan_to_num_native.h>
+#include <ATen/ops/nanmean_native.h>
+#include <ATen/ops/nanmedian_native.h>
+#include <ATen/ops/nanquantile_native.h>
+#include <ATen/ops/nansum_native.h>
+#include <ATen/ops/narrow_native.h>
+#include <ATen/ops/narrow_copy_native.h>
+#include <ATen/ops/native_batch_norm_native.h>
+#include <ATen/ops/native_batch_norm_backward_native.h>
+#include <ATen/ops/native_channel_shuffle_native.h>
+#include <ATen/ops/native_dropout_native.h>
+#include <ATen/ops/native_dropout_backward_native.h>
+#include <ATen/ops/native_group_norm_native.h>
+#include <ATen/ops/native_group_norm_backward_native.h>
+#include <ATen/ops/native_layer_norm_native.h>
+#include <ATen/ops/native_layer_norm_backward_native.h>
+#include <ATen/ops/native_norm_native.h>
+#include <ATen/ops/ne_native.h>
+#include <ATen/ops/neg_native.h>
+#include <ATen/ops/negative_native.h>
+#include <ATen/ops/nested_to_padded_tensor_native.h>
+#include <ATen/ops/new_empty_native.h>
+#include <ATen/ops/new_empty_strided_native.h>
+#include <ATen/ops/new_full_native.h>
+#include <ATen/ops/new_ones_native.h>
+#include <ATen/ops/new_zeros_native.h>
+#include <ATen/ops/nextafter_native.h>
+#include <ATen/ops/nll_loss_native.h>
+#include <ATen/ops/nll_loss2d_native.h>
+#include <ATen/ops/nll_loss2d_backward_native.h>
+#include <ATen/ops/nll_loss2d_forward_native.h>
+#include <ATen/ops/nll_loss_backward_native.h>
+#include <ATen/ops/nll_loss_forward_native.h>
+#include <ATen/ops/nll_loss_nd_native.h>
+#include <ATen/ops/nonzero_native.h>
+#include <ATen/ops/nonzero_numpy_native.h>
+#include <ATen/ops/nonzero_static_native.h>
+#include <ATen/ops/norm_native.h>
+#include <ATen/ops/norm_except_dim_native.h>
+#include <ATen/ops/normal_native.h>
+#include <ATen/ops/not_equal_native.h>
+#include <ATen/ops/nuclear_norm_native.h>
+#include <ATen/ops/numpy_T_native.h>
+#include <ATen/ops/one_hot_native.h>
+#include <ATen/ops/ones_native.h>
+#include <ATen/ops/ones_like_native.h>
+#include <ATen/ops/or_native.h>
+#include <ATen/ops/orgqr_native.h>
+#include <ATen/ops/ormqr_native.h>
+#include <ATen/ops/outer_native.h>
+#include <ATen/ops/output_nr_native.h>
+#include <ATen/ops/pad_native.h>
+#include <ATen/ops/pad_sequence_native.h>
+#include <ATen/ops/pairwise_distance_native.h>
+#include <ATen/ops/pdist_native.h>
+#include <ATen/ops/permute_native.h>
+#include <ATen/ops/permute_copy_native.h>
+#include <ATen/ops/pin_memory_native.h>
+#include <ATen/ops/pinverse_native.h>
+#include <ATen/ops/pixel_shuffle_native.h>
+#include <ATen/ops/pixel_unshuffle_native.h>
+#include <ATen/ops/poisson_native.h>
+#include <ATen/ops/poisson_nll_loss_native.h>
+#include <ATen/ops/polar_native.h>
+#include <ATen/ops/polygamma_native.h>
+#include <ATen/ops/positive_native.h>
+#include <ATen/ops/pow_native.h>
+#include <ATen/ops/prelu_native.h>
+#include <ATen/ops/prod_native.h>
+#include <ATen/ops/promote_types_native.h>
+#include <ATen/ops/put_native.h>
+#include <ATen/ops/q_per_channel_axis_native.h>
+#include <ATen/ops/q_per_channel_scales_native.h>
+#include <ATen/ops/q_per_channel_zero_points_native.h>
+#include <ATen/ops/q_scale_native.h>
+#include <ATen/ops/q_zero_point_native.h>
+#include <ATen/ops/qr_native.h>
+#include <ATen/ops/qscheme_native.h>
+#include <ATen/ops/quantile_native.h>
+#include <ATen/ops/quantize_per_channel_native.h>
+#include <ATen/ops/quantize_per_tensor_native.h>
+#include <ATen/ops/quantize_per_tensor_dynamic_native.h>
+#include <ATen/ops/quantized_batch_norm_native.h>
+#include <ATen/ops/quantized_gru_cell_native.h>
+#include <ATen/ops/quantized_lstm_cell_native.h>
+#include <ATen/ops/quantized_max_pool1d_native.h>
+#include <ATen/ops/quantized_max_pool2d_native.h>
+#include <ATen/ops/quantized_max_pool3d_native.h>
+#include <ATen/ops/quantized_rnn_relu_cell_native.h>
+#include <ATen/ops/quantized_rnn_tanh_cell_native.h>
+#include <ATen/ops/rad2deg_native.h>
+#include <ATen/ops/rand_native.h>
+#include <ATen/ops/rand_like_native.h>
+#include <ATen/ops/randint_native.h>
+#include <ATen/ops/randint_like_native.h>
+#include <ATen/ops/randn_native.h>
+#include <ATen/ops/randn_like_native.h>
+#include <ATen/ops/random_native.h>
+#include <ATen/ops/randperm_native.h>
+#include <ATen/ops/range_native.h>
+#include <ATen/ops/ravel_native.h>
+#include <ATen/ops/real_native.h>
+#include <ATen/ops/reciprocal_native.h>
+#include <ATen/ops/record_stream_native.h>
+#include <ATen/ops/refine_names_native.h>
+#include <ATen/ops/reflection_pad1d_native.h>
+#include <ATen/ops/reflection_pad1d_backward_native.h>
+#include <ATen/ops/reflection_pad2d_native.h>
+#include <ATen/ops/reflection_pad2d_backward_native.h>
+#include <ATen/ops/reflection_pad3d_native.h>
+#include <ATen/ops/reflection_pad3d_backward_native.h>
+#include <ATen/ops/relu_native.h>
+#include <ATen/ops/relu6_native.h>
+#include <ATen/ops/remainder_native.h>
+#include <ATen/ops/rename_native.h>
+#include <ATen/ops/renorm_native.h>
+#include <ATen/ops/repeat_native.h>
+#include <ATen/ops/repeat_interleave_native.h>
+#include <ATen/ops/replication_pad1d_native.h>
+#include <ATen/ops/replication_pad1d_backward_native.h>
+#include <ATen/ops/replication_pad2d_native.h>
+#include <ATen/ops/replication_pad2d_backward_native.h>
+#include <ATen/ops/replication_pad3d_native.h>
+#include <ATen/ops/replication_pad3d_backward_native.h>
+#include <ATen/ops/requires_grad_native.h>
+#include <ATen/ops/reshape_native.h>
+#include <ATen/ops/reshape_as_native.h>
+#include <ATen/ops/resize_native.h>
+#include <ATen/ops/resize_as_native.h>
+#include <ATen/ops/resize_as_sparse_native.h>
+#include <ATen/ops/resolve_conj_native.h>
+#include <ATen/ops/resolve_neg_native.h>
+#include <ATen/ops/result_type_native.h>
+#include <ATen/ops/retain_grad_native.h>
+#include <ATen/ops/retains_grad_native.h>
+#include <ATen/ops/rms_norm_native.h>
+#include <ATen/ops/rnn_relu_native.h>
+#include <ATen/ops/rnn_relu_cell_native.h>
+#include <ATen/ops/rnn_tanh_native.h>
+#include <ATen/ops/rnn_tanh_cell_native.h>
+#include <ATen/ops/roll_native.h>
+#include <ATen/ops/rot90_native.h>
+#include <ATen/ops/round_native.h>
+#include <ATen/ops/row_indices_native.h>
+#include <ATen/ops/row_indices_copy_native.h>
+#include <ATen/ops/row_stack_native.h>
+#include <ATen/ops/rrelu_native.h>
+#include <ATen/ops/rrelu_with_noise_native.h>
+#include <ATen/ops/rrelu_with_noise_backward_native.h>
+#include <ATen/ops/rshift_native.h>
+#include <ATen/ops/rsqrt_native.h>
+#include <ATen/ops/rsub_native.h>
+#include <ATen/ops/scalar_tensor_native.h>
+#include <ATen/ops/scaled_dot_product_attention_native.h>
+#include <ATen/ops/scatter_native.h>
+#include <ATen/ops/scatter_add_native.h>
+#include <ATen/ops/scatter_reduce_native.h>
+#include <ATen/ops/searchsorted_native.h>
+#include <ATen/ops/segment_reduce_native.h>
+#include <ATen/ops/select_native.h>
+#include <ATen/ops/select_backward_native.h>
+#include <ATen/ops/select_copy_native.h>
+#include <ATen/ops/select_scatter_native.h>
+#include <ATen/ops/selu_native.h>
+#include <ATen/ops/set_native.h>
+#include <ATen/ops/set_data_native.h>
+#include <ATen/ops/sgn_native.h>
+#include <ATen/ops/sigmoid_native.h>
+#include <ATen/ops/sigmoid_backward_native.h>
+#include <ATen/ops/sign_native.h>
+#include <ATen/ops/signbit_native.h>
+#include <ATen/ops/silu_native.h>
+#include <ATen/ops/silu_backward_native.h>
+#include <ATen/ops/sin_native.h>
+#include <ATen/ops/sinc_native.h>
+#include <ATen/ops/sinh_native.h>
+#include <ATen/ops/size_native.h>
+#include <ATen/ops/slice_native.h>
+#include <ATen/ops/slice_backward_native.h>
+#include <ATen/ops/slice_copy_native.h>
+#include <ATen/ops/slice_inverse_native.h>
+#include <ATen/ops/slice_scatter_native.h>
+#include <ATen/ops/slogdet_native.h>
+#include <ATen/ops/slow_conv3d_native.h>
+#include <ATen/ops/slow_conv3d_forward_native.h>
+#include <ATen/ops/slow_conv_dilated2d_native.h>
+#include <ATen/ops/slow_conv_dilated3d_native.h>
+#include <ATen/ops/slow_conv_transpose2d_native.h>
+#include <ATen/ops/slow_conv_transpose3d_native.h>
+#include <ATen/ops/smm_native.h>
+#include <ATen/ops/smooth_l1_loss_native.h>
+#include <ATen/ops/smooth_l1_loss_backward_native.h>
+#include <ATen/ops/soft_margin_loss_native.h>
+#include <ATen/ops/soft_margin_loss_backward_native.h>
+#include <ATen/ops/softmax_native.h>
+#include <ATen/ops/softplus_native.h>
+#include <ATen/ops/softplus_backward_native.h>
+#include <ATen/ops/softshrink_native.h>
+#include <ATen/ops/softshrink_backward_native.h>
+#include <ATen/ops/sort_native.h>
+#include <ATen/ops/sparse_bsc_tensor_native.h>
+#include <ATen/ops/sparse_bsr_tensor_native.h>
+#include <ATen/ops/sparse_compressed_tensor_native.h>
+#include <ATen/ops/sparse_coo_tensor_native.h>
+#include <ATen/ops/sparse_csc_tensor_native.h>
+#include <ATen/ops/sparse_csr_tensor_native.h>
+#include <ATen/ops/sparse_dim_native.h>
+#include <ATen/ops/sparse_mask_native.h>
+#include <ATen/ops/sparse_resize_native.h>
+#include <ATen/ops/sparse_resize_and_clear_native.h>
+#include <ATen/ops/sparse_sampled_addmm_native.h>
+#include <ATen/ops/special_airy_ai_native.h>
+#include <ATen/ops/special_bessel_j0_native.h>
+#include <ATen/ops/special_bessel_j1_native.h>
+#include <ATen/ops/special_bessel_y0_native.h>
+#include <ATen/ops/special_bessel_y1_native.h>
+#include <ATen/ops/special_chebyshev_polynomial_t_native.h>
+#include <ATen/ops/special_chebyshev_polynomial_u_native.h>
+#include <ATen/ops/special_chebyshev_polynomial_v_native.h>
+#include <ATen/ops/special_chebyshev_polynomial_w_native.h>
+#include <ATen/ops/special_digamma_native.h>
+#include <ATen/ops/special_entr_native.h>
+#include <ATen/ops/special_erf_native.h>
+#include <ATen/ops/special_erfc_native.h>
+#include <ATen/ops/special_erfcx_native.h>
+#include <ATen/ops/special_erfinv_native.h>
+#include <ATen/ops/special_exp2_native.h>
+#include <ATen/ops/special_expit_native.h>
+#include <ATen/ops/special_expm1_native.h>
+#include <ATen/ops/special_gammainc_native.h>
+#include <ATen/ops/special_gammaincc_native.h>
+#include <ATen/ops/special_gammaln_native.h>
+#include <ATen/ops/special_hermite_polynomial_h_native.h>
+#include <ATen/ops/special_hermite_polynomial_he_native.h>
+#include <ATen/ops/special_i0_native.h>
+#include <ATen/ops/special_i0e_native.h>
+#include <ATen/ops/special_i1_native.h>
+#include <ATen/ops/special_i1e_native.h>
+#include <ATen/ops/special_laguerre_polynomial_l_native.h>
+#include <ATen/ops/special_legendre_polynomial_p_native.h>
+#include <ATen/ops/special_log1p_native.h>
+#include <ATen/ops/special_log_ndtr_native.h>
+#include <ATen/ops/special_log_softmax_native.h>
+#include <ATen/ops/special_logit_native.h>
+#include <ATen/ops/special_logsumexp_native.h>
+#include <ATen/ops/special_modified_bessel_i0_native.h>
+#include <ATen/ops/special_modified_bessel_i1_native.h>
+#include <ATen/ops/special_modified_bessel_k0_native.h>
+#include <ATen/ops/special_modified_bessel_k1_native.h>
+#include <ATen/ops/special_multigammaln_native.h>
+#include <ATen/ops/special_ndtr_native.h>
+#include <ATen/ops/special_ndtri_native.h>
+#include <ATen/ops/special_polygamma_native.h>
+#include <ATen/ops/special_psi_native.h>
+#include <ATen/ops/special_round_native.h>
+#include <ATen/ops/special_scaled_modified_bessel_k0_native.h>
+#include <ATen/ops/special_scaled_modified_bessel_k1_native.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_t_native.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_u_native.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_v_native.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_w_native.h>
+#include <ATen/ops/special_sinc_native.h>
+#include <ATen/ops/special_softmax_native.h>
+#include <ATen/ops/special_spherical_bessel_j0_native.h>
+#include <ATen/ops/special_xlog1py_native.h>
+#include <ATen/ops/special_xlogy_native.h>
+#include <ATen/ops/special_zeta_native.h>
+#include <ATen/ops/split_native.h>
+#include <ATen/ops/split_copy_native.h>
+#include <ATen/ops/split_with_sizes_native.h>
+#include <ATen/ops/split_with_sizes_copy_native.h>
+#include <ATen/ops/sqrt_native.h>
+#include <ATen/ops/square_native.h>
+#include <ATen/ops/squeeze_native.h>
+#include <ATen/ops/squeeze_copy_native.h>
+#include <ATen/ops/sspaddmm_native.h>
+#include <ATen/ops/stack_native.h>
+#include <ATen/ops/std_native.h>
+#include <ATen/ops/std_mean_native.h>
+#include <ATen/ops/stft_native.h>
+#include <ATen/ops/stride_native.h>
+#include <ATen/ops/sub_native.h>
+#include <ATen/ops/subtract_native.h>
+#include <ATen/ops/sum_native.h>
+#include <ATen/ops/sum_to_size_native.h>
+#include <ATen/ops/svd_native.h>
+#include <ATen/ops/swapaxes_native.h>
+#include <ATen/ops/swapdims_native.h>
+#include <ATen/ops/sym_constrain_range_native.h>
+#include <ATen/ops/sym_constrain_range_for_size_native.h>
+#include <ATen/ops/sym_numel_native.h>
+#include <ATen/ops/sym_size_native.h>
+#include <ATen/ops/sym_storage_offset_native.h>
+#include <ATen/ops/sym_stride_native.h>
+#include <ATen/ops/t_native.h>
+#include <ATen/ops/t_copy_native.h>
+#include <ATen/ops/take_native.h>
+#include <ATen/ops/take_along_dim_native.h>
+#include <ATen/ops/tan_native.h>
+#include <ATen/ops/tanh_native.h>
+#include <ATen/ops/tanh_backward_native.h>
+#include <ATen/ops/tensor_split_native.h>
+#include <ATen/ops/tensordot_native.h>
+#include <ATen/ops/thnn_conv2d_native.h>
+#include <ATen/ops/threshold_native.h>
+#include <ATen/ops/threshold_backward_native.h>
+#include <ATen/ops/tile_native.h>
+#include <ATen/ops/to_native.h>
+#include <ATen/ops/to_dense_native.h>
+#include <ATen/ops/to_dense_backward_native.h>
+#include <ATen/ops/to_mkldnn_native.h>
+#include <ATen/ops/to_mkldnn_backward_native.h>
+#include <ATen/ops/to_padded_tensor_native.h>
+#include <ATen/ops/to_sparse_native.h>
+#include <ATen/ops/to_sparse_bsc_native.h>
+#include <ATen/ops/to_sparse_bsr_native.h>
+#include <ATen/ops/to_sparse_csc_native.h>
+#include <ATen/ops/to_sparse_csr_native.h>
+#include <ATen/ops/topk_native.h>
+#include <ATen/ops/trace_native.h>
+#include <ATen/ops/trace_backward_native.h>
+#include <ATen/ops/transpose_native.h>
+#include <ATen/ops/transpose_copy_native.h>
+#include <ATen/ops/trapezoid_native.h>
+#include <ATen/ops/trapz_native.h>
+#include <ATen/ops/triangular_solve_native.h>
+#include <ATen/ops/tril_native.h>
+#include <ATen/ops/tril_indices_native.h>
+#include <ATen/ops/triplet_margin_loss_native.h>
+#include <ATen/ops/triu_native.h>
+#include <ATen/ops/triu_indices_native.h>
+#include <ATen/ops/true_divide_native.h>
+#include <ATen/ops/trunc_native.h>
+#include <ATen/ops/type_as_native.h>
+#include <ATen/ops/unbind_native.h>
+#include <ATen/ops/unbind_copy_native.h>
+#include <ATen/ops/unflatten_native.h>
+#include <ATen/ops/unflatten_dense_tensors_native.h>
+#include <ATen/ops/unfold_native.h>
+#include <ATen/ops/unfold_backward_native.h>
+#include <ATen/ops/unfold_copy_native.h>
+#include <ATen/ops/uniform_native.h>
+#include <ATen/ops/unique_consecutive_native.h>
+#include <ATen/ops/unique_dim_native.h>
+#include <ATen/ops/unique_dim_consecutive_native.h>
+#include <ATen/ops/unsafe_chunk_native.h>
+#include <ATen/ops/unsafe_split_native.h>
+#include <ATen/ops/unsafe_split_with_sizes_native.h>
+#include <ATen/ops/unsqueeze_native.h>
+#include <ATen/ops/unsqueeze_copy_native.h>
+#include <ATen/ops/upsample_bicubic2d_native.h>
+#include <ATen/ops/upsample_bicubic2d_backward_native.h>
+#include <ATen/ops/upsample_bilinear2d_native.h>
+#include <ATen/ops/upsample_bilinear2d_backward_native.h>
+#include <ATen/ops/upsample_linear1d_native.h>
+#include <ATen/ops/upsample_linear1d_backward_native.h>
+#include <ATen/ops/upsample_nearest1d_native.h>
+#include <ATen/ops/upsample_nearest1d_backward_native.h>
+#include <ATen/ops/upsample_nearest2d_native.h>
+#include <ATen/ops/upsample_nearest2d_backward_native.h>
+#include <ATen/ops/upsample_nearest3d_native.h>
+#include <ATen/ops/upsample_nearest3d_backward_native.h>
+#include <ATen/ops/upsample_trilinear3d_native.h>
+#include <ATen/ops/upsample_trilinear3d_backward_native.h>
+#include <ATen/ops/value_selecting_reduction_backward_native.h>
+#include <ATen/ops/values_native.h>
+#include <ATen/ops/values_copy_native.h>
+#include <ATen/ops/vander_native.h>
+#include <ATen/ops/var_native.h>
+#include <ATen/ops/var_mean_native.h>
+#include <ATen/ops/vdot_native.h>
+#include <ATen/ops/view_native.h>
+#include <ATen/ops/view_as_native.h>
+#include <ATen/ops/view_as_complex_native.h>
+#include <ATen/ops/view_as_complex_copy_native.h>
+#include <ATen/ops/view_as_real_native.h>
+#include <ATen/ops/view_as_real_copy_native.h>
+#include <ATen/ops/view_copy_native.h>
+#include <ATen/ops/vsplit_native.h>
+#include <ATen/ops/vstack_native.h>
+#include <ATen/ops/where_native.h>
+#include <ATen/ops/xlogy_native.h>
+#include <ATen/ops/xor_native.h>
+#include <ATen/ops/zero_native.h>
+#include <ATen/ops/zeros_native.h>
+#include <ATen/ops/zeros_like_native.h>
+
+

lib/python3.12/site-packages/torch/include/ATen/NativeMetaFunctions.h

@@ -0,0 +1,1334 @@
+#pragma once
+
+// @generated by torchgen/gen.py from NativeMetaFunctions.h
+
+#include <ATen/core/Tensor.h>
+#include <ATen/core/IListRef.h>
+#include <ATen/TensorMeta.h>
+#include <ATen/TensorIterator.h>
+
+#include <ATen/ops/_adaptive_avg_pool2d_meta.h>
+#include <ATen/ops/_adaptive_avg_pool2d_backward_meta.h>
+#include <ATen/ops/_adaptive_avg_pool3d_meta.h>
+#include <ATen/ops/_adaptive_avg_pool3d_backward_meta.h>
+#include <ATen/ops/_add_batch_dim_meta.h>
+#include <ATen/ops/_add_relu_meta.h>
+#include <ATen/ops/_addmm_activation_meta.h>
+#include <ATen/ops/_aminmax_meta.h>
+#include <ATen/ops/_amp_foreach_non_finite_check_and_unscale_meta.h>
+#include <ATen/ops/_amp_update_scale_meta.h>
+#include <ATen/ops/_assert_async_meta.h>
+#include <ATen/ops/_assert_scalar_meta.h>
+#include <ATen/ops/_assert_tensor_metadata_meta.h>
+#include <ATen/ops/_autocast_to_full_precision_meta.h>
+#include <ATen/ops/_autocast_to_reduced_precision_meta.h>
+#include <ATen/ops/_backward_meta.h>
+#include <ATen/ops/_batch_norm_impl_index_meta.h>
+#include <ATen/ops/_batch_norm_impl_index_backward_meta.h>
+#include <ATen/ops/_batch_norm_no_update_meta.h>
+#include <ATen/ops/_batch_norm_with_update_meta.h>
+#include <ATen/ops/_cast_Byte_meta.h>
+#include <ATen/ops/_cast_Char_meta.h>
+#include <ATen/ops/_cast_Double_meta.h>
+#include <ATen/ops/_cast_Float_meta.h>
+#include <ATen/ops/_cast_Half_meta.h>
+#include <ATen/ops/_cast_Int_meta.h>
+#include <ATen/ops/_cast_Long_meta.h>
+#include <ATen/ops/_cast_Short_meta.h>
+#include <ATen/ops/_cdist_backward_meta.h>
+#include <ATen/ops/_cdist_forward_meta.h>
+#include <ATen/ops/_cholesky_solve_helper_meta.h>
+#include <ATen/ops/_choose_qparams_per_tensor_meta.h>
+#include <ATen/ops/_chunk_cat_meta.h>
+#include <ATen/ops/_coalesce_meta.h>
+#include <ATen/ops/_coalesced_meta.h>
+#include <ATen/ops/_compute_linear_combination_meta.h>
+#include <ATen/ops/_conj_meta.h>
+#include <ATen/ops/_conj_copy_meta.h>
+#include <ATen/ops/_conj_physical_meta.h>
+#include <ATen/ops/_conv_depthwise2d_meta.h>
+#include <ATen/ops/_convert_indices_from_coo_to_csr_meta.h>
+#include <ATen/ops/_convert_indices_from_csr_to_coo_meta.h>
+#include <ATen/ops/_convert_weight_to_int4pack_meta.h>
+#include <ATen/ops/_convert_weight_to_int4pack_for_cpu_meta.h>
+#include <ATen/ops/_convolution_meta.h>
+#include <ATen/ops/_convolution_double_backward_meta.h>
+#include <ATen/ops/_convolution_mode_meta.h>
+#include <ATen/ops/_copy_from_meta.h>
+#include <ATen/ops/_copy_from_and_resize_meta.h>
+#include <ATen/ops/_cslt_compress_meta.h>
+#include <ATen/ops/_cslt_sparse_mm_meta.h>
+#include <ATen/ops/_cslt_sparse_mm_search_meta.h>
+#include <ATen/ops/_ctc_loss_meta.h>
+#include <ATen/ops/_ctc_loss_backward_meta.h>
+#include <ATen/ops/_cudnn_ctc_loss_meta.h>
+#include <ATen/ops/_cudnn_init_dropout_state_meta.h>
+#include <ATen/ops/_cudnn_rnn_meta.h>
+#include <ATen/ops/_cudnn_rnn_backward_meta.h>
+#include <ATen/ops/_cudnn_rnn_flatten_weight_meta.h>
+#include <ATen/ops/_cufft_clear_plan_cache_meta.h>
+#include <ATen/ops/_cufft_get_plan_cache_max_size_meta.h>
+#include <ATen/ops/_cufft_get_plan_cache_size_meta.h>
+#include <ATen/ops/_cufft_set_plan_cache_max_size_meta.h>
+#include <ATen/ops/_cummax_helper_meta.h>
+#include <ATen/ops/_cummin_helper_meta.h>
+#include <ATen/ops/_debug_has_internal_overlap_meta.h>
+#include <ATen/ops/_dimI_meta.h>
+#include <ATen/ops/_dimV_meta.h>
+#include <ATen/ops/_dim_arange_meta.h>
+#include <ATen/ops/_dirichlet_grad_meta.h>
+#include <ATen/ops/_efficient_attention_backward_meta.h>
+#include <ATen/ops/_efficient_attention_forward_meta.h>
+#include <ATen/ops/_efficientzerotensor_meta.h>
+#include <ATen/ops/_embedding_bag_meta.h>
+#include <ATen/ops/_embedding_bag_backward_meta.h>
+#include <ATen/ops/_embedding_bag_dense_backward_meta.h>
+#include <ATen/ops/_embedding_bag_forward_only_meta.h>
+#include <ATen/ops/_embedding_bag_per_sample_weights_backward_meta.h>
+#include <ATen/ops/_embedding_bag_sparse_backward_meta.h>
+#include <ATen/ops/_empty_affine_quantized_meta.h>
+#include <ATen/ops/_empty_per_channel_affine_quantized_meta.h>
+#include <ATen/ops/_euclidean_dist_meta.h>
+#include <ATen/ops/_fake_quantize_learnable_per_channel_affine_meta.h>
+#include <ATen/ops/_fake_quantize_learnable_per_channel_affine_backward_meta.h>
+#include <ATen/ops/_fake_quantize_learnable_per_tensor_affine_meta.h>
+#include <ATen/ops/_fake_quantize_learnable_per_tensor_affine_backward_meta.h>
+#include <ATen/ops/_fake_quantize_per_tensor_affine_cachemask_tensor_qparams_meta.h>
+#include <ATen/ops/_fft_c2c_meta.h>
+#include <ATen/ops/_fft_c2r_meta.h>
+#include <ATen/ops/_fft_r2c_meta.h>
+#include <ATen/ops/_fill_mem_eff_dropout_mask_meta.h>
+#include <ATen/ops/_flash_attention_backward_meta.h>
+#include <ATen/ops/_flash_attention_forward_meta.h>
+#include <ATen/ops/_foobar_meta.h>
+#include <ATen/ops/_foreach_abs_meta.h>
+#include <ATen/ops/_foreach_acos_meta.h>
+#include <ATen/ops/_foreach_add_meta.h>
+#include <ATen/ops/_foreach_addcdiv_meta.h>
+#include <ATen/ops/_foreach_addcmul_meta.h>
+#include <ATen/ops/_foreach_asin_meta.h>
+#include <ATen/ops/_foreach_atan_meta.h>
+#include <ATen/ops/_foreach_ceil_meta.h>
+#include <ATen/ops/_foreach_clamp_max_meta.h>
+#include <ATen/ops/_foreach_clamp_min_meta.h>
+#include <ATen/ops/_foreach_copy_meta.h>
+#include <ATen/ops/_foreach_cos_meta.h>
+#include <ATen/ops/_foreach_cosh_meta.h>
+#include <ATen/ops/_foreach_div_meta.h>
+#include <ATen/ops/_foreach_erf_meta.h>
+#include <ATen/ops/_foreach_erfc_meta.h>
+#include <ATen/ops/_foreach_exp_meta.h>
+#include <ATen/ops/_foreach_expm1_meta.h>
+#include <ATen/ops/_foreach_floor_meta.h>
+#include <ATen/ops/_foreach_frac_meta.h>
+#include <ATen/ops/_foreach_lerp_meta.h>
+#include <ATen/ops/_foreach_lgamma_meta.h>
+#include <ATen/ops/_foreach_log_meta.h>
+#include <ATen/ops/_foreach_log10_meta.h>
+#include <ATen/ops/_foreach_log1p_meta.h>
+#include <ATen/ops/_foreach_log2_meta.h>
+#include <ATen/ops/_foreach_max_meta.h>
+#include <ATen/ops/_foreach_maximum_meta.h>
+#include <ATen/ops/_foreach_minimum_meta.h>
+#include <ATen/ops/_foreach_mul_meta.h>
+#include <ATen/ops/_foreach_neg_meta.h>
+#include <ATen/ops/_foreach_norm_meta.h>
+#include <ATen/ops/_foreach_pow_meta.h>
+#include <ATen/ops/_foreach_reciprocal_meta.h>
+#include <ATen/ops/_foreach_round_meta.h>
+#include <ATen/ops/_foreach_rsqrt_meta.h>
+#include <ATen/ops/_foreach_sigmoid_meta.h>
+#include <ATen/ops/_foreach_sign_meta.h>
+#include <ATen/ops/_foreach_sin_meta.h>
+#include <ATen/ops/_foreach_sinh_meta.h>
+#include <ATen/ops/_foreach_sqrt_meta.h>
+#include <ATen/ops/_foreach_sub_meta.h>
+#include <ATen/ops/_foreach_tan_meta.h>
+#include <ATen/ops/_foreach_tanh_meta.h>
+#include <ATen/ops/_foreach_trunc_meta.h>
+#include <ATen/ops/_foreach_zero_meta.h>
+#include <ATen/ops/_functional_assert_async_meta.h>
+#include <ATen/ops/_functional_assert_scalar_meta.h>
+#include <ATen/ops/_functional_sym_constrain_range_meta.h>
+#include <ATen/ops/_functional_sym_constrain_range_for_size_meta.h>
+#include <ATen/ops/_fused_adagrad_meta.h>
+#include <ATen/ops/_fused_adam_meta.h>
+#include <ATen/ops/_fused_adamw_meta.h>
+#include <ATen/ops/_fused_dropout_meta.h>
+#include <ATen/ops/_fused_moving_avg_obs_fq_helper_meta.h>
+#include <ATen/ops/_fused_sdp_choice_meta.h>
+#include <ATen/ops/_fused_sgd_meta.h>
+#include <ATen/ops/_fw_primal_meta.h>
+#include <ATen/ops/_fw_primal_copy_meta.h>
+#include <ATen/ops/_gather_sparse_backward_meta.h>
+#include <ATen/ops/_grid_sampler_2d_cpu_fallback_meta.h>
+#include <ATen/ops/_grid_sampler_2d_cpu_fallback_backward_meta.h>
+#include <ATen/ops/_has_compatible_shallow_copy_type_meta.h>
+#include <ATen/ops/_has_same_storage_numel_meta.h>
+#include <ATen/ops/_histogramdd_bin_edges_meta.h>
+#include <ATen/ops/_histogramdd_from_bin_cts_meta.h>
+#include <ATen/ops/_histogramdd_from_bin_tensors_meta.h>
+#include <ATen/ops/_index_put_impl_meta.h>
+#include <ATen/ops/_indices_meta.h>
+#include <ATen/ops/_indices_copy_meta.h>
+#include <ATen/ops/_int_mm_meta.h>
+#include <ATen/ops/_is_all_true_meta.h>
+#include <ATen/ops/_is_any_true_meta.h>
+#include <ATen/ops/_is_zerotensor_meta.h>
+#include <ATen/ops/_jagged_to_padded_dense_forward_meta.h>
+#include <ATen/ops/_lazy_clone_meta.h>
+#include <ATen/ops/_linalg_check_errors_meta.h>
+#include <ATen/ops/_linalg_det_meta.h>
+#include <ATen/ops/_linalg_eigh_meta.h>
+#include <ATen/ops/_linalg_eigvals_meta.h>
+#include <ATen/ops/_linalg_slogdet_meta.h>
+#include <ATen/ops/_linalg_solve_ex_meta.h>
+#include <ATen/ops/_linalg_svd_meta.h>
+#include <ATen/ops/_local_scalar_dense_meta.h>
+#include <ATen/ops/_log_softmax_meta.h>
+#include <ATen/ops/_log_softmax_backward_data_meta.h>
+#include <ATen/ops/_logcumsumexp_meta.h>
+#include <ATen/ops/_lstm_mps_meta.h>
+#include <ATen/ops/_lu_with_info_meta.h>
+#include <ATen/ops/_make_dep_token_meta.h>
+#include <ATen/ops/_make_dual_meta.h>
+#include <ATen/ops/_make_dual_copy_meta.h>
+#include <ATen/ops/_make_per_channel_quantized_tensor_meta.h>
+#include <ATen/ops/_make_per_tensor_quantized_tensor_meta.h>
+#include <ATen/ops/_masked_scale_meta.h>
+#include <ATen/ops/_masked_softmax_meta.h>
+#include <ATen/ops/_masked_softmax_backward_meta.h>
+#include <ATen/ops/_mixed_dtypes_linear_meta.h>
+#include <ATen/ops/_mkldnn_reshape_meta.h>
+#include <ATen/ops/_mkldnn_transpose_meta.h>
+#include <ATen/ops/_mps_convolution_meta.h>
+#include <ATen/ops/_mps_convolution_transpose_meta.h>
+#include <ATen/ops/_native_batch_norm_legit_meta.h>
+#include <ATen/ops/_native_batch_norm_legit_no_training_meta.h>
+#include <ATen/ops/_native_multi_head_attention_meta.h>
+#include <ATen/ops/_neg_view_meta.h>
+#include <ATen/ops/_neg_view_copy_meta.h>
+#include <ATen/ops/_nested_compute_contiguous_strides_offsets_meta.h>
+#include <ATen/ops/_nested_from_padded_meta.h>
+#include <ATen/ops/_nested_from_padded_and_nested_example_meta.h>
+#include <ATen/ops/_nested_from_padded_tensor_meta.h>
+#include <ATen/ops/_nested_get_jagged_dummy_meta.h>
+#include <ATen/ops/_nested_get_lengths_meta.h>
+#include <ATen/ops/_nested_get_max_seqlen_meta.h>
+#include <ATen/ops/_nested_get_min_seqlen_meta.h>
+#include <ATen/ops/_nested_get_offsets_meta.h>
+#include <ATen/ops/_nested_get_ragged_idx_meta.h>
+#include <ATen/ops/_nested_get_values_meta.h>
+#include <ATen/ops/_nested_get_values_copy_meta.h>
+#include <ATen/ops/_nested_select_backward_meta.h>
+#include <ATen/ops/_nested_sum_backward_meta.h>
+#include <ATen/ops/_nested_tensor_from_mask_meta.h>
+#include <ATen/ops/_nested_tensor_from_mask_left_aligned_meta.h>
+#include <ATen/ops/_nested_tensor_from_tensor_list_meta.h>
+#include <ATen/ops/_nested_tensor_size_meta.h>
+#include <ATen/ops/_nested_tensor_softmax_with_shape_meta.h>
+#include <ATen/ops/_nested_tensor_storage_offsets_meta.h>
+#include <ATen/ops/_nested_tensor_strides_meta.h>
+#include <ATen/ops/_nested_view_from_buffer_meta.h>
+#include <ATen/ops/_nested_view_from_buffer_copy_meta.h>
+#include <ATen/ops/_nested_view_from_jagged_meta.h>
+#include <ATen/ops/_nested_view_from_jagged_copy_meta.h>
+#include <ATen/ops/_new_zeros_with_same_feature_meta_meta.h>
+#include <ATen/ops/_nnpack_available_meta.h>
+#include <ATen/ops/_nnpack_spatial_convolution_meta.h>
+#include <ATen/ops/_nnz_meta.h>
+#include <ATen/ops/_pack_padded_sequence_meta.h>
+#include <ATen/ops/_pack_padded_sequence_backward_meta.h>
+#include <ATen/ops/_pad_circular_meta.h>
+#include <ATen/ops/_pad_enum_meta.h>
+#include <ATen/ops/_pad_packed_sequence_meta.h>
+#include <ATen/ops/_padded_dense_to_jagged_forward_meta.h>
+#include <ATen/ops/_pdist_backward_meta.h>
+#include <ATen/ops/_pdist_forward_meta.h>
+#include <ATen/ops/_pin_memory_meta.h>
+#include <ATen/ops/_prelu_kernel_meta.h>
+#include <ATen/ops/_prelu_kernel_backward_meta.h>
+#include <ATen/ops/_print_meta.h>
+#include <ATen/ops/_propagate_xla_data_meta.h>
+#include <ATen/ops/_remove_batch_dim_meta.h>
+#include <ATen/ops/_reshape_alias_meta.h>
+#include <ATen/ops/_reshape_alias_copy_meta.h>
+#include <ATen/ops/_reshape_copy_meta.h>
+#include <ATen/ops/_reshape_from_tensor_meta.h>
+#include <ATen/ops/_resize_output_meta.h>
+#include <ATen/ops/_rowwise_prune_meta.h>
+#include <ATen/ops/_safe_softmax_meta.h>
+#include <ATen/ops/_sample_dirichlet_meta.h>
+#include <ATen/ops/_saturate_weight_to_fp16_meta.h>
+#include <ATen/ops/_scaled_dot_product_attention_math_meta.h>
+#include <ATen/ops/_scaled_dot_product_attention_math_for_mps_meta.h>
+#include <ATen/ops/_scaled_dot_product_cudnn_attention_meta.h>
+#include <ATen/ops/_scaled_dot_product_cudnn_attention_backward_meta.h>
+#include <ATen/ops/_scaled_dot_product_efficient_attention_meta.h>
+#include <ATen/ops/_scaled_dot_product_efficient_attention_backward_meta.h>
+#include <ATen/ops/_scaled_dot_product_flash_attention_meta.h>
+#include <ATen/ops/_scaled_dot_product_flash_attention_backward_meta.h>
+#include <ATen/ops/_scaled_dot_product_flash_attention_for_cpu_meta.h>
+#include <ATen/ops/_scaled_dot_product_flash_attention_for_cpu_backward_meta.h>
+#include <ATen/ops/_scaled_dot_product_fused_attention_overrideable_meta.h>
+#include <ATen/ops/_scaled_dot_product_fused_attention_overrideable_backward_meta.h>
+#include <ATen/ops/_scaled_mm_meta.h>
+#include <ATen/ops/_segment_reduce_backward_meta.h>
+#include <ATen/ops/_shape_as_tensor_meta.h>
+#include <ATen/ops/_slow_conv2d_backward_meta.h>
+#include <ATen/ops/_slow_conv2d_forward_meta.h>
+#include <ATen/ops/_sobol_engine_draw_meta.h>
+#include <ATen/ops/_sobol_engine_ff_meta.h>
+#include <ATen/ops/_sobol_engine_initialize_state_meta.h>
+#include <ATen/ops/_sobol_engine_scramble_meta.h>
+#include <ATen/ops/_softmax_meta.h>
+#include <ATen/ops/_softmax_backward_data_meta.h>
+#include <ATen/ops/_sparse_addmm_meta.h>
+#include <ATen/ops/_sparse_broadcast_to_meta.h>
+#include <ATen/ops/_sparse_broadcast_to_copy_meta.h>
+#include <ATen/ops/_sparse_bsc_tensor_unsafe_meta.h>
+#include <ATen/ops/_sparse_bsr_tensor_unsafe_meta.h>
+#include <ATen/ops/_sparse_compressed_tensor_unsafe_meta.h>
+#include <ATen/ops/_sparse_compressed_tensor_with_dims_meta.h>
+#include <ATen/ops/_sparse_coo_tensor_unsafe_meta.h>
+#include <ATen/ops/_sparse_coo_tensor_with_dims_meta.h>
+#include <ATen/ops/_sparse_coo_tensor_with_dims_and_tensors_meta.h>
+#include <ATen/ops/_sparse_csc_tensor_unsafe_meta.h>
+#include <ATen/ops/_sparse_csr_prod_meta.h>
+#include <ATen/ops/_sparse_csr_sum_meta.h>
+#include <ATen/ops/_sparse_csr_tensor_unsafe_meta.h>
+#include <ATen/ops/_sparse_log_softmax_meta.h>
+#include <ATen/ops/_sparse_log_softmax_backward_data_meta.h>
+#include <ATen/ops/_sparse_mask_projection_meta.h>
+#include <ATen/ops/_sparse_mm_meta.h>
+#include <ATen/ops/_sparse_mm_reduce_impl_meta.h>
+#include <ATen/ops/_sparse_mm_reduce_impl_backward_meta.h>
+#include <ATen/ops/_sparse_semi_structured_addmm_meta.h>
+#include <ATen/ops/_sparse_semi_structured_apply_meta.h>
+#include <ATen/ops/_sparse_semi_structured_apply_dense_meta.h>
+#include <ATen/ops/_sparse_semi_structured_linear_meta.h>
+#include <ATen/ops/_sparse_semi_structured_mm_meta.h>
+#include <ATen/ops/_sparse_semi_structured_tile_meta.h>
+#include <ATen/ops/_sparse_softmax_meta.h>
+#include <ATen/ops/_sparse_softmax_backward_data_meta.h>
+#include <ATen/ops/_sparse_sparse_matmul_meta.h>
+#include <ATen/ops/_sparse_sum_meta.h>
+#include <ATen/ops/_sparse_sum_backward_meta.h>
+#include <ATen/ops/_spdiags_meta.h>
+#include <ATen/ops/_spsolve_meta.h>
+#include <ATen/ops/_stack_meta.h>
+#include <ATen/ops/_standard_gamma_meta.h>
+#include <ATen/ops/_standard_gamma_grad_meta.h>
+#include <ATen/ops/_test_ambiguous_defaults_meta.h>
+#include <ATen/ops/_test_autograd_multiple_dispatch_meta.h>
+#include <ATen/ops/_test_autograd_multiple_dispatch_view_meta.h>
+#include <ATen/ops/_test_autograd_multiple_dispatch_view_copy_meta.h>
+#include <ATen/ops/_test_check_tensor_meta.h>
+#include <ATen/ops/_test_functorch_fallback_meta.h>
+#include <ATen/ops/_test_optional_filled_intlist_meta.h>
+#include <ATen/ops/_test_optional_floatlist_meta.h>
+#include <ATen/ops/_test_optional_intlist_meta.h>
+#include <ATen/ops/_test_parallel_materialize_meta.h>
+#include <ATen/ops/_test_serialization_subcmul_meta.h>
+#include <ATen/ops/_test_string_default_meta.h>
+#include <ATen/ops/_test_warn_in_autograd_meta.h>
+#include <ATen/ops/_thnn_differentiable_gru_cell_backward_meta.h>
+#include <ATen/ops/_thnn_differentiable_lstm_cell_backward_meta.h>
+#include <ATen/ops/_thnn_fused_gru_cell_meta.h>
+#include <ATen/ops/_thnn_fused_gru_cell_backward_meta.h>
+#include <ATen/ops/_thnn_fused_lstm_cell_meta.h>
+#include <ATen/ops/_thnn_fused_lstm_cell_backward_meta.h>
+#include <ATen/ops/_thnn_fused_lstm_cell_backward_impl_meta.h>
+#include <ATen/ops/_to_copy_meta.h>
+#include <ATen/ops/_to_cpu_meta.h>
+#include <ATen/ops/_to_dense_meta.h>
+#include <ATen/ops/_to_sparse_meta.h>
+#include <ATen/ops/_to_sparse_bsc_meta.h>
+#include <ATen/ops/_to_sparse_bsr_meta.h>
+#include <ATen/ops/_to_sparse_csc_meta.h>
+#include <ATen/ops/_to_sparse_csr_meta.h>
+#include <ATen/ops/_to_sparse_semi_structured_meta.h>
+#include <ATen/ops/_transform_bias_rescale_qkv_meta.h>
+#include <ATen/ops/_transformer_encoder_layer_fwd_meta.h>
+#include <ATen/ops/_trilinear_meta.h>
+#include <ATen/ops/_triton_multi_head_attention_meta.h>
+#include <ATen/ops/_triton_scaled_dot_attention_meta.h>
+#include <ATen/ops/_unique_meta.h>
+#include <ATen/ops/_unique2_meta.h>
+#include <ATen/ops/_unpack_dual_meta.h>
+#include <ATen/ops/_unsafe_index_meta.h>
+#include <ATen/ops/_unsafe_index_put_meta.h>
+#include <ATen/ops/_unsafe_masked_index_meta.h>
+#include <ATen/ops/_unsafe_masked_index_put_accumulate_meta.h>
+#include <ATen/ops/_unsafe_view_meta.h>
+#include <ATen/ops/_upsample_bicubic2d_aa_meta.h>
+#include <ATen/ops/_upsample_bicubic2d_aa_backward_meta.h>
+#include <ATen/ops/_upsample_bilinear2d_aa_meta.h>
+#include <ATen/ops/_upsample_bilinear2d_aa_backward_meta.h>
+#include <ATen/ops/_upsample_nearest_exact1d_meta.h>
+#include <ATen/ops/_upsample_nearest_exact1d_backward_meta.h>
+#include <ATen/ops/_upsample_nearest_exact2d_meta.h>
+#include <ATen/ops/_upsample_nearest_exact2d_backward_meta.h>
+#include <ATen/ops/_upsample_nearest_exact3d_meta.h>
+#include <ATen/ops/_upsample_nearest_exact3d_backward_meta.h>
+#include <ATen/ops/_use_cudnn_ctc_loss_meta.h>
+#include <ATen/ops/_use_cudnn_rnn_flatten_weight_meta.h>
+#include <ATen/ops/_validate_compressed_sparse_indices_meta.h>
+#include <ATen/ops/_validate_sparse_bsc_tensor_args_meta.h>
+#include <ATen/ops/_validate_sparse_bsr_tensor_args_meta.h>
+#include <ATen/ops/_validate_sparse_compressed_tensor_args_meta.h>
+#include <ATen/ops/_validate_sparse_coo_tensor_args_meta.h>
+#include <ATen/ops/_validate_sparse_csc_tensor_args_meta.h>
+#include <ATen/ops/_validate_sparse_csr_tensor_args_meta.h>
+#include <ATen/ops/_values_meta.h>
+#include <ATen/ops/_values_copy_meta.h>
+#include <ATen/ops/_version_meta.h>
+#include <ATen/ops/_weight_int4pack_mm_meta.h>
+#include <ATen/ops/_weight_int4pack_mm_for_cpu_meta.h>
+#include <ATen/ops/_weight_int8pack_mm_meta.h>
+#include <ATen/ops/_weight_norm_meta.h>
+#include <ATen/ops/_weight_norm_differentiable_backward_meta.h>
+#include <ATen/ops/_weight_norm_interface_meta.h>
+#include <ATen/ops/_weight_norm_interface_backward_meta.h>
+#include <ATen/ops/_wrapped_linear_prepack_meta.h>
+#include <ATen/ops/_wrapped_quantized_linear_prepacked_meta.h>
+#include <ATen/ops/abs_meta.h>
+#include <ATen/ops/absolute_meta.h>
+#include <ATen/ops/acos_meta.h>
+#include <ATen/ops/acosh_meta.h>
+#include <ATen/ops/adaptive_avg_pool1d_meta.h>
+#include <ATen/ops/adaptive_avg_pool2d_meta.h>
+#include <ATen/ops/adaptive_avg_pool3d_meta.h>
+#include <ATen/ops/adaptive_avg_pool3d_backward_meta.h>
+#include <ATen/ops/adaptive_max_pool1d_meta.h>
+#include <ATen/ops/adaptive_max_pool2d_meta.h>
+#include <ATen/ops/adaptive_max_pool2d_backward_meta.h>
+#include <ATen/ops/adaptive_max_pool3d_meta.h>
+#include <ATen/ops/adaptive_max_pool3d_backward_meta.h>
+#include <ATen/ops/add_meta.h>
+#include <ATen/ops/addbmm_meta.h>
+#include <ATen/ops/addcdiv_meta.h>
+#include <ATen/ops/addcmul_meta.h>
+#include <ATen/ops/addmm_meta.h>
+#include <ATen/ops/addmv_meta.h>
+#include <ATen/ops/addr_meta.h>
+#include <ATen/ops/adjoint_meta.h>
+#include <ATen/ops/affine_grid_generator_meta.h>
+#include <ATen/ops/affine_grid_generator_backward_meta.h>
+#include <ATen/ops/alias_meta.h>
+#include <ATen/ops/alias_copy_meta.h>
+#include <ATen/ops/align_as_meta.h>
+#include <ATen/ops/align_tensors_meta.h>
+#include <ATen/ops/align_to_meta.h>
+#include <ATen/ops/all_meta.h>
+#include <ATen/ops/allclose_meta.h>
+#include <ATen/ops/alpha_dropout_meta.h>
+#include <ATen/ops/amax_meta.h>
+#include <ATen/ops/amin_meta.h>
+#include <ATen/ops/aminmax_meta.h>
+#include <ATen/ops/and_meta.h>
+#include <ATen/ops/angle_meta.h>
+#include <ATen/ops/any_meta.h>
+#include <ATen/ops/arange_meta.h>
+#include <ATen/ops/arccos_meta.h>
+#include <ATen/ops/arccosh_meta.h>
+#include <ATen/ops/arcsin_meta.h>
+#include <ATen/ops/arcsinh_meta.h>
+#include <ATen/ops/arctan_meta.h>
+#include <ATen/ops/arctan2_meta.h>
+#include <ATen/ops/arctanh_meta.h>
+#include <ATen/ops/argmax_meta.h>
+#include <ATen/ops/argmin_meta.h>
+#include <ATen/ops/argsort_meta.h>
+#include <ATen/ops/argwhere_meta.h>
+#include <ATen/ops/as_strided_meta.h>
+#include <ATen/ops/as_strided_copy_meta.h>
+#include <ATen/ops/as_strided_scatter_meta.h>
+#include <ATen/ops/asin_meta.h>
+#include <ATen/ops/asinh_meta.h>
+#include <ATen/ops/atan_meta.h>
+#include <ATen/ops/atan2_meta.h>
+#include <ATen/ops/atanh_meta.h>
+#include <ATen/ops/atleast_1d_meta.h>
+#include <ATen/ops/atleast_2d_meta.h>
+#include <ATen/ops/atleast_3d_meta.h>
+#include <ATen/ops/avg_pool1d_meta.h>
+#include <ATen/ops/avg_pool2d_meta.h>
+#include <ATen/ops/avg_pool2d_backward_meta.h>
+#include <ATen/ops/avg_pool3d_meta.h>
+#include <ATen/ops/avg_pool3d_backward_meta.h>
+#include <ATen/ops/baddbmm_meta.h>
+#include <ATen/ops/bartlett_window_meta.h>
+#include <ATen/ops/batch_norm_meta.h>
+#include <ATen/ops/batch_norm_backward_meta.h>
+#include <ATen/ops/batch_norm_backward_elemt_meta.h>
+#include <ATen/ops/batch_norm_backward_reduce_meta.h>
+#include <ATen/ops/batch_norm_elemt_meta.h>
+#include <ATen/ops/batch_norm_gather_stats_meta.h>
+#include <ATen/ops/batch_norm_gather_stats_with_counts_meta.h>
+#include <ATen/ops/batch_norm_stats_meta.h>
+#include <ATen/ops/batch_norm_update_stats_meta.h>
+#include <ATen/ops/bernoulli_meta.h>
+#include <ATen/ops/bilinear_meta.h>
+#include <ATen/ops/binary_cross_entropy_meta.h>
+#include <ATen/ops/binary_cross_entropy_backward_meta.h>
+#include <ATen/ops/binary_cross_entropy_with_logits_meta.h>
+#include <ATen/ops/bincount_meta.h>
+#include <ATen/ops/binomial_meta.h>
+#include <ATen/ops/bitwise_and_meta.h>
+#include <ATen/ops/bitwise_left_shift_meta.h>
+#include <ATen/ops/bitwise_not_meta.h>
+#include <ATen/ops/bitwise_or_meta.h>
+#include <ATen/ops/bitwise_right_shift_meta.h>
+#include <ATen/ops/bitwise_xor_meta.h>
+#include <ATen/ops/blackman_window_meta.h>
+#include <ATen/ops/block_diag_meta.h>
+#include <ATen/ops/bmm_meta.h>
+#include <ATen/ops/broadcast_tensors_meta.h>
+#include <ATen/ops/broadcast_to_meta.h>
+#include <ATen/ops/bucketize_meta.h>
+#include <ATen/ops/can_cast_meta.h>
+#include <ATen/ops/cartesian_prod_meta.h>
+#include <ATen/ops/cat_meta.h>
+#include <ATen/ops/cauchy_meta.h>
+#include <ATen/ops/ccol_indices_meta.h>
+#include <ATen/ops/ccol_indices_copy_meta.h>
+#include <ATen/ops/cdist_meta.h>
+#include <ATen/ops/ceil_meta.h>
+#include <ATen/ops/celu_meta.h>
+#include <ATen/ops/chain_matmul_meta.h>
+#include <ATen/ops/chalf_meta.h>
+#include <ATen/ops/channel_shuffle_meta.h>
+#include <ATen/ops/cholesky_meta.h>
+#include <ATen/ops/cholesky_inverse_meta.h>
+#include <ATen/ops/cholesky_solve_meta.h>
+#include <ATen/ops/choose_qparams_optimized_meta.h>
+#include <ATen/ops/chunk_meta.h>
+#include <ATen/ops/clamp_meta.h>
+#include <ATen/ops/clamp_max_meta.h>
+#include <ATen/ops/clamp_min_meta.h>
+#include <ATen/ops/clip_meta.h>
+#include <ATen/ops/clone_meta.h>
+#include <ATen/ops/coalesce_meta.h>
+#include <ATen/ops/col2im_meta.h>
+#include <ATen/ops/col_indices_meta.h>
+#include <ATen/ops/col_indices_copy_meta.h>
+#include <ATen/ops/column_stack_meta.h>
+#include <ATen/ops/combinations_meta.h>
+#include <ATen/ops/complex_meta.h>
+#include <ATen/ops/concat_meta.h>
+#include <ATen/ops/concatenate_meta.h>
+#include <ATen/ops/conj_meta.h>
+#include <ATen/ops/conj_physical_meta.h>
+#include <ATen/ops/constant_pad_nd_meta.h>
+#include <ATen/ops/contiguous_meta.h>
+#include <ATen/ops/conv1d_meta.h>
+#include <ATen/ops/conv2d_meta.h>
+#include <ATen/ops/conv3d_meta.h>
+#include <ATen/ops/conv_depthwise3d_meta.h>
+#include <ATen/ops/conv_tbc_meta.h>
+#include <ATen/ops/conv_tbc_backward_meta.h>
+#include <ATen/ops/conv_transpose1d_meta.h>
+#include <ATen/ops/conv_transpose2d_meta.h>
+#include <ATen/ops/conv_transpose3d_meta.h>
+#include <ATen/ops/convolution_meta.h>
+#include <ATen/ops/convolution_backward_meta.h>
+#include <ATen/ops/convolution_backward_overrideable_meta.h>
+#include <ATen/ops/convolution_overrideable_meta.h>
+#include <ATen/ops/copy_meta.h>
+#include <ATen/ops/copy_sparse_to_sparse_meta.h>
+#include <ATen/ops/copysign_meta.h>
+#include <ATen/ops/corrcoef_meta.h>
+#include <ATen/ops/cos_meta.h>
+#include <ATen/ops/cosh_meta.h>
+#include <ATen/ops/cosine_embedding_loss_meta.h>
+#include <ATen/ops/cosine_similarity_meta.h>
+#include <ATen/ops/count_nonzero_meta.h>
+#include <ATen/ops/cov_meta.h>
+#include <ATen/ops/cross_meta.h>
+#include <ATen/ops/cross_entropy_loss_meta.h>
+#include <ATen/ops/crow_indices_meta.h>
+#include <ATen/ops/crow_indices_copy_meta.h>
+#include <ATen/ops/ctc_loss_meta.h>
+#include <ATen/ops/cudnn_affine_grid_generator_meta.h>
+#include <ATen/ops/cudnn_affine_grid_generator_backward_meta.h>
+#include <ATen/ops/cudnn_batch_norm_meta.h>
+#include <ATen/ops/cudnn_batch_norm_backward_meta.h>
+#include <ATen/ops/cudnn_convolution_meta.h>
+#include <ATen/ops/cudnn_convolution_add_relu_meta.h>
+#include <ATen/ops/cudnn_convolution_relu_meta.h>
+#include <ATen/ops/cudnn_convolution_transpose_meta.h>
+#include <ATen/ops/cudnn_grid_sampler_meta.h>
+#include <ATen/ops/cudnn_grid_sampler_backward_meta.h>
+#include <ATen/ops/cudnn_is_acceptable_meta.h>
+#include <ATen/ops/cummax_meta.h>
+#include <ATen/ops/cummaxmin_backward_meta.h>
+#include <ATen/ops/cummin_meta.h>
+#include <ATen/ops/cumprod_meta.h>
+#include <ATen/ops/cumprod_backward_meta.h>
+#include <ATen/ops/cumsum_meta.h>
+#include <ATen/ops/cumulative_trapezoid_meta.h>
+#include <ATen/ops/data_meta.h>
+#include <ATen/ops/deg2rad_meta.h>
+#include <ATen/ops/dense_dim_meta.h>
+#include <ATen/ops/dequantize_meta.h>
+#include <ATen/ops/det_meta.h>
+#include <ATen/ops/detach_meta.h>
+#include <ATen/ops/detach_copy_meta.h>
+#include <ATen/ops/diag_meta.h>
+#include <ATen/ops/diag_embed_meta.h>
+#include <ATen/ops/diagflat_meta.h>
+#include <ATen/ops/diagonal_meta.h>
+#include <ATen/ops/diagonal_backward_meta.h>
+#include <ATen/ops/diagonal_copy_meta.h>
+#include <ATen/ops/diagonal_scatter_meta.h>
+#include <ATen/ops/diff_meta.h>
+#include <ATen/ops/digamma_meta.h>
+#include <ATen/ops/dist_meta.h>
+#include <ATen/ops/div_meta.h>
+#include <ATen/ops/divide_meta.h>
+#include <ATen/ops/dot_meta.h>
+#include <ATen/ops/dropout_meta.h>
+#include <ATen/ops/dsplit_meta.h>
+#include <ATen/ops/dstack_meta.h>
+#include <ATen/ops/einsum_meta.h>
+#include <ATen/ops/elu_meta.h>
+#include <ATen/ops/elu_backward_meta.h>
+#include <ATen/ops/embedding_meta.h>
+#include <ATen/ops/embedding_backward_meta.h>
+#include <ATen/ops/embedding_bag_meta.h>
+#include <ATen/ops/embedding_dense_backward_meta.h>
+#include <ATen/ops/embedding_renorm_meta.h>
+#include <ATen/ops/embedding_sparse_backward_meta.h>
+#include <ATen/ops/empty_meta.h>
+#include <ATen/ops/empty_like_meta.h>
+#include <ATen/ops/empty_permuted_meta.h>
+#include <ATen/ops/empty_quantized_meta.h>
+#include <ATen/ops/empty_strided_meta.h>
+#include <ATen/ops/eq_meta.h>
+#include <ATen/ops/equal_meta.h>
+#include <ATen/ops/erf_meta.h>
+#include <ATen/ops/erfc_meta.h>
+#include <ATen/ops/erfinv_meta.h>
+#include <ATen/ops/exp_meta.h>
+#include <ATen/ops/exp2_meta.h>
+#include <ATen/ops/expand_meta.h>
+#include <ATen/ops/expand_as_meta.h>
+#include <ATen/ops/expand_copy_meta.h>
+#include <ATen/ops/expm1_meta.h>
+#include <ATen/ops/exponential_meta.h>
+#include <ATen/ops/eye_meta.h>
+#include <ATen/ops/fake_quantize_per_channel_affine_meta.h>
+#include <ATen/ops/fake_quantize_per_channel_affine_cachemask_meta.h>
+#include <ATen/ops/fake_quantize_per_channel_affine_cachemask_backward_meta.h>
+#include <ATen/ops/fake_quantize_per_tensor_affine_meta.h>
+#include <ATen/ops/fake_quantize_per_tensor_affine_cachemask_meta.h>
+#include <ATen/ops/fake_quantize_per_tensor_affine_cachemask_backward_meta.h>
+#include <ATen/ops/fbgemm_linear_fp16_weight_meta.h>
+#include <ATen/ops/fbgemm_linear_fp16_weight_fp32_activation_meta.h>
+#include <ATen/ops/fbgemm_linear_int8_weight_meta.h>
+#include <ATen/ops/fbgemm_linear_int8_weight_fp32_activation_meta.h>
+#include <ATen/ops/fbgemm_linear_quantize_weight_meta.h>
+#include <ATen/ops/fbgemm_pack_gemm_matrix_fp16_meta.h>
+#include <ATen/ops/fbgemm_pack_quantized_matrix_meta.h>
+#include <ATen/ops/feature_alpha_dropout_meta.h>
+#include <ATen/ops/feature_dropout_meta.h>
+#include <ATen/ops/fft_fft_meta.h>
+#include <ATen/ops/fft_fft2_meta.h>
+#include <ATen/ops/fft_fftfreq_meta.h>
+#include <ATen/ops/fft_fftn_meta.h>
+#include <ATen/ops/fft_fftshift_meta.h>
+#include <ATen/ops/fft_hfft_meta.h>
+#include <ATen/ops/fft_hfft2_meta.h>
+#include <ATen/ops/fft_hfftn_meta.h>
+#include <ATen/ops/fft_ifft_meta.h>
+#include <ATen/ops/fft_ifft2_meta.h>
+#include <ATen/ops/fft_ifftn_meta.h>
+#include <ATen/ops/fft_ifftshift_meta.h>
+#include <ATen/ops/fft_ihfft_meta.h>
+#include <ATen/ops/fft_ihfft2_meta.h>
+#include <ATen/ops/fft_ihfftn_meta.h>
+#include <ATen/ops/fft_irfft_meta.h>
+#include <ATen/ops/fft_irfft2_meta.h>
+#include <ATen/ops/fft_irfftn_meta.h>
+#include <ATen/ops/fft_rfft_meta.h>
+#include <ATen/ops/fft_rfft2_meta.h>
+#include <ATen/ops/fft_rfftfreq_meta.h>
+#include <ATen/ops/fft_rfftn_meta.h>
+#include <ATen/ops/fill_meta.h>
+#include <ATen/ops/fill_diagonal_meta.h>
+#include <ATen/ops/fix_meta.h>
+#include <ATen/ops/flatten_meta.h>
+#include <ATen/ops/flatten_dense_tensors_meta.h>
+#include <ATen/ops/flip_meta.h>
+#include <ATen/ops/fliplr_meta.h>
+#include <ATen/ops/flipud_meta.h>
+#include <ATen/ops/float_power_meta.h>
+#include <ATen/ops/floor_meta.h>
+#include <ATen/ops/floor_divide_meta.h>
+#include <ATen/ops/fmax_meta.h>
+#include <ATen/ops/fmin_meta.h>
+#include <ATen/ops/fmod_meta.h>
+#include <ATen/ops/frac_meta.h>
+#include <ATen/ops/fractional_max_pool2d_meta.h>
+#include <ATen/ops/fractional_max_pool2d_backward_meta.h>
+#include <ATen/ops/fractional_max_pool3d_meta.h>
+#include <ATen/ops/fractional_max_pool3d_backward_meta.h>
+#include <ATen/ops/frexp_meta.h>
+#include <ATen/ops/frobenius_norm_meta.h>
+#include <ATen/ops/from_file_meta.h>
+#include <ATen/ops/full_meta.h>
+#include <ATen/ops/full_like_meta.h>
+#include <ATen/ops/fused_moving_avg_obs_fake_quant_meta.h>
+#include <ATen/ops/gather_meta.h>
+#include <ATen/ops/gather_backward_meta.h>
+#include <ATen/ops/gcd_meta.h>
+#include <ATen/ops/ge_meta.h>
+#include <ATen/ops/gelu_meta.h>
+#include <ATen/ops/gelu_backward_meta.h>
+#include <ATen/ops/geometric_meta.h>
+#include <ATen/ops/geqrf_meta.h>
+#include <ATen/ops/ger_meta.h>
+#include <ATen/ops/glu_meta.h>
+#include <ATen/ops/glu_backward_meta.h>
+#include <ATen/ops/glu_backward_jvp_meta.h>
+#include <ATen/ops/glu_jvp_meta.h>
+#include <ATen/ops/gradient_meta.h>
+#include <ATen/ops/greater_meta.h>
+#include <ATen/ops/greater_equal_meta.h>
+#include <ATen/ops/grid_sampler_meta.h>
+#include <ATen/ops/grid_sampler_2d_meta.h>
+#include <ATen/ops/grid_sampler_2d_backward_meta.h>
+#include <ATen/ops/grid_sampler_3d_meta.h>
+#include <ATen/ops/grid_sampler_3d_backward_meta.h>
+#include <ATen/ops/group_norm_meta.h>
+#include <ATen/ops/gru_meta.h>
+#include <ATen/ops/gru_cell_meta.h>
+#include <ATen/ops/gt_meta.h>
+#include <ATen/ops/hamming_window_meta.h>
+#include <ATen/ops/hann_window_meta.h>
+#include <ATen/ops/hardshrink_meta.h>
+#include <ATen/ops/hardshrink_backward_meta.h>
+#include <ATen/ops/hardsigmoid_meta.h>
+#include <ATen/ops/hardsigmoid_backward_meta.h>
+#include <ATen/ops/hardswish_meta.h>
+#include <ATen/ops/hardswish_backward_meta.h>
+#include <ATen/ops/hardtanh_meta.h>
+#include <ATen/ops/hardtanh_backward_meta.h>
+#include <ATen/ops/heaviside_meta.h>
+#include <ATen/ops/hinge_embedding_loss_meta.h>
+#include <ATen/ops/histc_meta.h>
+#include <ATen/ops/histogram_meta.h>
+#include <ATen/ops/histogramdd_meta.h>
+#include <ATen/ops/hsplit_meta.h>
+#include <ATen/ops/hspmm_meta.h>
+#include <ATen/ops/hstack_meta.h>
+#include <ATen/ops/huber_loss_meta.h>
+#include <ATen/ops/huber_loss_backward_meta.h>
+#include <ATen/ops/hypot_meta.h>
+#include <ATen/ops/i0_meta.h>
+#include <ATen/ops/igamma_meta.h>
+#include <ATen/ops/igammac_meta.h>
+#include <ATen/ops/im2col_meta.h>
+#include <ATen/ops/imag_meta.h>
+#include <ATen/ops/index_meta.h>
+#include <ATen/ops/index_add_meta.h>
+#include <ATen/ops/index_copy_meta.h>
+#include <ATen/ops/index_fill_meta.h>
+#include <ATen/ops/index_put_meta.h>
+#include <ATen/ops/index_reduce_meta.h>
+#include <ATen/ops/index_select_meta.h>
+#include <ATen/ops/index_select_backward_meta.h>
+#include <ATen/ops/indices_meta.h>
+#include <ATen/ops/indices_copy_meta.h>
+#include <ATen/ops/infinitely_differentiable_gelu_backward_meta.h>
+#include <ATen/ops/inner_meta.h>
+#include <ATen/ops/instance_norm_meta.h>
+#include <ATen/ops/int_repr_meta.h>
+#include <ATen/ops/inverse_meta.h>
+#include <ATen/ops/is_coalesced_meta.h>
+#include <ATen/ops/is_complex_meta.h>
+#include <ATen/ops/is_conj_meta.h>
+#include <ATen/ops/is_distributed_meta.h>
+#include <ATen/ops/is_floating_point_meta.h>
+#include <ATen/ops/is_inference_meta.h>
+#include <ATen/ops/is_leaf_meta.h>
+#include <ATen/ops/is_neg_meta.h>
+#include <ATen/ops/is_nonzero_meta.h>
+#include <ATen/ops/is_pinned_meta.h>
+#include <ATen/ops/is_same_size_meta.h>
+#include <ATen/ops/is_set_to_meta.h>
+#include <ATen/ops/is_signed_meta.h>
+#include <ATen/ops/is_vulkan_available_meta.h>
+#include <ATen/ops/isclose_meta.h>
+#include <ATen/ops/isfinite_meta.h>
+#include <ATen/ops/isin_meta.h>
+#include <ATen/ops/isinf_meta.h>
+#include <ATen/ops/isnan_meta.h>
+#include <ATen/ops/isneginf_meta.h>
+#include <ATen/ops/isposinf_meta.h>
+#include <ATen/ops/isreal_meta.h>
+#include <ATen/ops/istft_meta.h>
+#include <ATen/ops/item_meta.h>
+#include <ATen/ops/kaiser_window_meta.h>
+#include <ATen/ops/kl_div_meta.h>
+#include <ATen/ops/kron_meta.h>
+#include <ATen/ops/kthvalue_meta.h>
+#include <ATen/ops/l1_loss_meta.h>
+#include <ATen/ops/layer_norm_meta.h>
+#include <ATen/ops/lcm_meta.h>
+#include <ATen/ops/ldexp_meta.h>
+#include <ATen/ops/le_meta.h>
+#include <ATen/ops/leaky_relu_meta.h>
+#include <ATen/ops/leaky_relu_backward_meta.h>
+#include <ATen/ops/lerp_meta.h>
+#include <ATen/ops/less_meta.h>
+#include <ATen/ops/less_equal_meta.h>
+#include <ATen/ops/lgamma_meta.h>
+#include <ATen/ops/lift_meta.h>
+#include <ATen/ops/lift_fresh_meta.h>
+#include <ATen/ops/lift_fresh_copy_meta.h>
+#include <ATen/ops/linalg_cholesky_meta.h>
+#include <ATen/ops/linalg_cholesky_ex_meta.h>
+#include <ATen/ops/linalg_cond_meta.h>
+#include <ATen/ops/linalg_cross_meta.h>
+#include <ATen/ops/linalg_det_meta.h>
+#include <ATen/ops/linalg_diagonal_meta.h>
+#include <ATen/ops/linalg_eig_meta.h>
+#include <ATen/ops/linalg_eigh_meta.h>
+#include <ATen/ops/linalg_eigvals_meta.h>
+#include <ATen/ops/linalg_eigvalsh_meta.h>
+#include <ATen/ops/linalg_householder_product_meta.h>
+#include <ATen/ops/linalg_inv_meta.h>
+#include <ATen/ops/linalg_inv_ex_meta.h>
+#include <ATen/ops/linalg_ldl_factor_meta.h>
+#include <ATen/ops/linalg_ldl_factor_ex_meta.h>
+#include <ATen/ops/linalg_ldl_solve_meta.h>
+#include <ATen/ops/linalg_lstsq_meta.h>
+#include <ATen/ops/linalg_lu_meta.h>
+#include <ATen/ops/linalg_lu_factor_meta.h>
+#include <ATen/ops/linalg_lu_factor_ex_meta.h>
+#include <ATen/ops/linalg_lu_solve_meta.h>
+#include <ATen/ops/linalg_matmul_meta.h>
+#include <ATen/ops/linalg_matrix_exp_meta.h>
+#include <ATen/ops/linalg_matrix_norm_meta.h>
+#include <ATen/ops/linalg_matrix_power_meta.h>
+#include <ATen/ops/linalg_matrix_rank_meta.h>
+#include <ATen/ops/linalg_multi_dot_meta.h>
+#include <ATen/ops/linalg_norm_meta.h>
+#include <ATen/ops/linalg_pinv_meta.h>
+#include <ATen/ops/linalg_qr_meta.h>
+#include <ATen/ops/linalg_slogdet_meta.h>
+#include <ATen/ops/linalg_solve_meta.h>
+#include <ATen/ops/linalg_solve_ex_meta.h>
+#include <ATen/ops/linalg_solve_triangular_meta.h>
+#include <ATen/ops/linalg_svd_meta.h>
+#include <ATen/ops/linalg_svdvals_meta.h>
+#include <ATen/ops/linalg_tensorinv_meta.h>
+#include <ATen/ops/linalg_tensorsolve_meta.h>
+#include <ATen/ops/linalg_vander_meta.h>
+#include <ATen/ops/linalg_vecdot_meta.h>
+#include <ATen/ops/linalg_vector_norm_meta.h>
+#include <ATen/ops/linear_meta.h>
+#include <ATen/ops/linear_backward_meta.h>
+#include <ATen/ops/linspace_meta.h>
+#include <ATen/ops/log_meta.h>
+#include <ATen/ops/log10_meta.h>
+#include <ATen/ops/log1p_meta.h>
+#include <ATen/ops/log2_meta.h>
+#include <ATen/ops/log_normal_meta.h>
+#include <ATen/ops/log_sigmoid_meta.h>
+#include <ATen/ops/log_sigmoid_backward_meta.h>
+#include <ATen/ops/log_sigmoid_forward_meta.h>
+#include <ATen/ops/log_softmax_meta.h>
+#include <ATen/ops/logaddexp_meta.h>
+#include <ATen/ops/logaddexp2_meta.h>
+#include <ATen/ops/logcumsumexp_meta.h>
+#include <ATen/ops/logdet_meta.h>
+#include <ATen/ops/logical_and_meta.h>
+#include <ATen/ops/logical_not_meta.h>
+#include <ATen/ops/logical_or_meta.h>
+#include <ATen/ops/logical_xor_meta.h>
+#include <ATen/ops/logit_meta.h>
+#include <ATen/ops/logit_backward_meta.h>
+#include <ATen/ops/logspace_meta.h>
+#include <ATen/ops/logsumexp_meta.h>
+#include <ATen/ops/lshift_meta.h>
+#include <ATen/ops/lstm_meta.h>
+#include <ATen/ops/lstm_cell_meta.h>
+#include <ATen/ops/lstm_mps_backward_meta.h>
+#include <ATen/ops/lt_meta.h>
+#include <ATen/ops/lu_solve_meta.h>
+#include <ATen/ops/lu_unpack_meta.h>
+#include <ATen/ops/mH_meta.h>
+#include <ATen/ops/mT_meta.h>
+#include <ATen/ops/margin_ranking_loss_meta.h>
+#include <ATen/ops/masked_fill_meta.h>
+#include <ATen/ops/masked_scatter_meta.h>
+#include <ATen/ops/masked_scatter_backward_meta.h>
+#include <ATen/ops/masked_select_meta.h>
+#include <ATen/ops/masked_select_backward_meta.h>
+#include <ATen/ops/matmul_meta.h>
+#include <ATen/ops/matmul_backward_meta.h>
+#include <ATen/ops/matrix_H_meta.h>
+#include <ATen/ops/matrix_exp_meta.h>
+#include <ATen/ops/matrix_exp_backward_meta.h>
+#include <ATen/ops/matrix_power_meta.h>
+#include <ATen/ops/max_meta.h>
+#include <ATen/ops/max_pool1d_meta.h>
+#include <ATen/ops/max_pool1d_with_indices_meta.h>
+#include <ATen/ops/max_pool2d_meta.h>
+#include <ATen/ops/max_pool2d_backward_meta.h>
+#include <ATen/ops/max_pool2d_with_indices_meta.h>
+#include <ATen/ops/max_pool2d_with_indices_backward_meta.h>
+#include <ATen/ops/max_pool3d_meta.h>
+#include <ATen/ops/max_pool3d_with_indices_meta.h>
+#include <ATen/ops/max_pool3d_with_indices_backward_meta.h>
+#include <ATen/ops/max_unpool2d_meta.h>
+#include <ATen/ops/max_unpool3d_meta.h>
+#include <ATen/ops/maximum_meta.h>
+#include <ATen/ops/mean_meta.h>
+#include <ATen/ops/median_meta.h>
+#include <ATen/ops/meshgrid_meta.h>
+#include <ATen/ops/min_meta.h>
+#include <ATen/ops/minimum_meta.h>
+#include <ATen/ops/miopen_batch_norm_meta.h>
+#include <ATen/ops/miopen_batch_norm_backward_meta.h>
+#include <ATen/ops/miopen_convolution_meta.h>
+#include <ATen/ops/miopen_convolution_add_relu_meta.h>
+#include <ATen/ops/miopen_convolution_relu_meta.h>
+#include <ATen/ops/miopen_convolution_transpose_meta.h>
+#include <ATen/ops/miopen_depthwise_convolution_meta.h>
+#include <ATen/ops/miopen_rnn_meta.h>
+#include <ATen/ops/miopen_rnn_backward_meta.h>
+#include <ATen/ops/mish_meta.h>
+#include <ATen/ops/mish_backward_meta.h>
+#include <ATen/ops/mkldnn_adaptive_avg_pool2d_meta.h>
+#include <ATen/ops/mkldnn_adaptive_avg_pool2d_backward_meta.h>
+#include <ATen/ops/mkldnn_convolution_meta.h>
+#include <ATen/ops/mkldnn_linear_meta.h>
+#include <ATen/ops/mkldnn_linear_backward_meta.h>
+#include <ATen/ops/mkldnn_linear_backward_input_meta.h>
+#include <ATen/ops/mkldnn_linear_backward_weights_meta.h>
+#include <ATen/ops/mkldnn_max_pool2d_meta.h>
+#include <ATen/ops/mkldnn_max_pool2d_backward_meta.h>
+#include <ATen/ops/mkldnn_max_pool3d_meta.h>
+#include <ATen/ops/mkldnn_max_pool3d_backward_meta.h>
+#include <ATen/ops/mkldnn_reorder_conv2d_weight_meta.h>
+#include <ATen/ops/mkldnn_reorder_conv3d_weight_meta.h>
+#include <ATen/ops/mkldnn_rnn_layer_meta.h>
+#include <ATen/ops/mkldnn_rnn_layer_backward_meta.h>
+#include <ATen/ops/mm_meta.h>
+#include <ATen/ops/mode_meta.h>
+#include <ATen/ops/moveaxis_meta.h>
+#include <ATen/ops/movedim_meta.h>
+#include <ATen/ops/mps_convolution_backward_meta.h>
+#include <ATen/ops/mps_convolution_transpose_backward_meta.h>
+#include <ATen/ops/mse_loss_meta.h>
+#include <ATen/ops/mse_loss_backward_meta.h>
+#include <ATen/ops/msort_meta.h>
+#include <ATen/ops/mul_meta.h>
+#include <ATen/ops/multi_margin_loss_meta.h>
+#include <ATen/ops/multi_margin_loss_backward_meta.h>
+#include <ATen/ops/multilabel_margin_loss_meta.h>
+#include <ATen/ops/multilabel_margin_loss_backward_meta.h>
+#include <ATen/ops/multilabel_margin_loss_forward_meta.h>
+#include <ATen/ops/multinomial_meta.h>
+#include <ATen/ops/multiply_meta.h>
+#include <ATen/ops/mv_meta.h>
+#include <ATen/ops/mvlgamma_meta.h>
+#include <ATen/ops/nan_to_num_meta.h>
+#include <ATen/ops/nanmean_meta.h>
+#include <ATen/ops/nanmedian_meta.h>
+#include <ATen/ops/nanquantile_meta.h>
+#include <ATen/ops/nansum_meta.h>
+#include <ATen/ops/narrow_meta.h>
+#include <ATen/ops/narrow_copy_meta.h>
+#include <ATen/ops/native_batch_norm_meta.h>
+#include <ATen/ops/native_batch_norm_backward_meta.h>
+#include <ATen/ops/native_channel_shuffle_meta.h>
+#include <ATen/ops/native_dropout_meta.h>
+#include <ATen/ops/native_dropout_backward_meta.h>
+#include <ATen/ops/native_group_norm_meta.h>
+#include <ATen/ops/native_group_norm_backward_meta.h>
+#include <ATen/ops/native_layer_norm_meta.h>
+#include <ATen/ops/native_layer_norm_backward_meta.h>
+#include <ATen/ops/native_norm_meta.h>
+#include <ATen/ops/ne_meta.h>
+#include <ATen/ops/neg_meta.h>
+#include <ATen/ops/negative_meta.h>
+#include <ATen/ops/nested_to_padded_tensor_meta.h>
+#include <ATen/ops/new_empty_meta.h>
+#include <ATen/ops/new_empty_strided_meta.h>
+#include <ATen/ops/new_full_meta.h>
+#include <ATen/ops/new_ones_meta.h>
+#include <ATen/ops/new_zeros_meta.h>
+#include <ATen/ops/nextafter_meta.h>
+#include <ATen/ops/nll_loss_meta.h>
+#include <ATen/ops/nll_loss2d_meta.h>
+#include <ATen/ops/nll_loss2d_backward_meta.h>
+#include <ATen/ops/nll_loss2d_forward_meta.h>
+#include <ATen/ops/nll_loss_backward_meta.h>
+#include <ATen/ops/nll_loss_forward_meta.h>
+#include <ATen/ops/nll_loss_nd_meta.h>
+#include <ATen/ops/nonzero_meta.h>
+#include <ATen/ops/nonzero_numpy_meta.h>
+#include <ATen/ops/nonzero_static_meta.h>
+#include <ATen/ops/norm_meta.h>
+#include <ATen/ops/norm_except_dim_meta.h>
+#include <ATen/ops/normal_meta.h>
+#include <ATen/ops/not_equal_meta.h>
+#include <ATen/ops/nuclear_norm_meta.h>
+#include <ATen/ops/numpy_T_meta.h>
+#include <ATen/ops/one_hot_meta.h>
+#include <ATen/ops/ones_meta.h>
+#include <ATen/ops/ones_like_meta.h>
+#include <ATen/ops/or_meta.h>
+#include <ATen/ops/orgqr_meta.h>
+#include <ATen/ops/ormqr_meta.h>
+#include <ATen/ops/outer_meta.h>
+#include <ATen/ops/output_nr_meta.h>
+#include <ATen/ops/pad_meta.h>
+#include <ATen/ops/pad_sequence_meta.h>
+#include <ATen/ops/pairwise_distance_meta.h>
+#include <ATen/ops/pdist_meta.h>
+#include <ATen/ops/permute_meta.h>
+#include <ATen/ops/permute_copy_meta.h>
+#include <ATen/ops/pin_memory_meta.h>
+#include <ATen/ops/pinverse_meta.h>
+#include <ATen/ops/pixel_shuffle_meta.h>
+#include <ATen/ops/pixel_unshuffle_meta.h>
+#include <ATen/ops/poisson_meta.h>
+#include <ATen/ops/poisson_nll_loss_meta.h>
+#include <ATen/ops/polar_meta.h>
+#include <ATen/ops/polygamma_meta.h>
+#include <ATen/ops/positive_meta.h>
+#include <ATen/ops/pow_meta.h>
+#include <ATen/ops/prelu_meta.h>
+#include <ATen/ops/prod_meta.h>
+#include <ATen/ops/promote_types_meta.h>
+#include <ATen/ops/put_meta.h>
+#include <ATen/ops/q_per_channel_axis_meta.h>
+#include <ATen/ops/q_per_channel_scales_meta.h>
+#include <ATen/ops/q_per_channel_zero_points_meta.h>
+#include <ATen/ops/q_scale_meta.h>
+#include <ATen/ops/q_zero_point_meta.h>
+#include <ATen/ops/qr_meta.h>
+#include <ATen/ops/qscheme_meta.h>
+#include <ATen/ops/quantile_meta.h>
+#include <ATen/ops/quantize_per_channel_meta.h>
+#include <ATen/ops/quantize_per_tensor_meta.h>
+#include <ATen/ops/quantize_per_tensor_dynamic_meta.h>
+#include <ATen/ops/quantized_batch_norm_meta.h>
+#include <ATen/ops/quantized_gru_cell_meta.h>
+#include <ATen/ops/quantized_lstm_cell_meta.h>
+#include <ATen/ops/quantized_max_pool1d_meta.h>
+#include <ATen/ops/quantized_max_pool2d_meta.h>
+#include <ATen/ops/quantized_max_pool3d_meta.h>
+#include <ATen/ops/quantized_rnn_relu_cell_meta.h>
+#include <ATen/ops/quantized_rnn_tanh_cell_meta.h>
+#include <ATen/ops/rad2deg_meta.h>
+#include <ATen/ops/rand_meta.h>
+#include <ATen/ops/rand_like_meta.h>
+#include <ATen/ops/randint_meta.h>
+#include <ATen/ops/randint_like_meta.h>
+#include <ATen/ops/randn_meta.h>
+#include <ATen/ops/randn_like_meta.h>
+#include <ATen/ops/random_meta.h>
+#include <ATen/ops/randperm_meta.h>
+#include <ATen/ops/range_meta.h>
+#include <ATen/ops/ravel_meta.h>
+#include <ATen/ops/real_meta.h>
+#include <ATen/ops/reciprocal_meta.h>
+#include <ATen/ops/record_stream_meta.h>
+#include <ATen/ops/refine_names_meta.h>
+#include <ATen/ops/reflection_pad1d_meta.h>
+#include <ATen/ops/reflection_pad1d_backward_meta.h>
+#include <ATen/ops/reflection_pad2d_meta.h>
+#include <ATen/ops/reflection_pad2d_backward_meta.h>
+#include <ATen/ops/reflection_pad3d_meta.h>
+#include <ATen/ops/reflection_pad3d_backward_meta.h>
+#include <ATen/ops/relu_meta.h>
+#include <ATen/ops/relu6_meta.h>
+#include <ATen/ops/remainder_meta.h>
+#include <ATen/ops/rename_meta.h>
+#include <ATen/ops/renorm_meta.h>
+#include <ATen/ops/repeat_meta.h>
+#include <ATen/ops/repeat_interleave_meta.h>
+#include <ATen/ops/replication_pad1d_meta.h>
+#include <ATen/ops/replication_pad1d_backward_meta.h>
+#include <ATen/ops/replication_pad2d_meta.h>
+#include <ATen/ops/replication_pad2d_backward_meta.h>
+#include <ATen/ops/replication_pad3d_meta.h>
+#include <ATen/ops/replication_pad3d_backward_meta.h>
+#include <ATen/ops/requires_grad_meta.h>
+#include <ATen/ops/reshape_meta.h>
+#include <ATen/ops/reshape_as_meta.h>
+#include <ATen/ops/resize_meta.h>
+#include <ATen/ops/resize_as_meta.h>
+#include <ATen/ops/resize_as_sparse_meta.h>
+#include <ATen/ops/resolve_conj_meta.h>
+#include <ATen/ops/resolve_neg_meta.h>
+#include <ATen/ops/result_type_meta.h>
+#include <ATen/ops/retain_grad_meta.h>
+#include <ATen/ops/retains_grad_meta.h>
+#include <ATen/ops/rms_norm_meta.h>
+#include <ATen/ops/rnn_relu_meta.h>
+#include <ATen/ops/rnn_relu_cell_meta.h>
+#include <ATen/ops/rnn_tanh_meta.h>
+#include <ATen/ops/rnn_tanh_cell_meta.h>
+#include <ATen/ops/roll_meta.h>
+#include <ATen/ops/rot90_meta.h>
+#include <ATen/ops/round_meta.h>
+#include <ATen/ops/row_indices_meta.h>
+#include <ATen/ops/row_indices_copy_meta.h>
+#include <ATen/ops/row_stack_meta.h>
+#include <ATen/ops/rrelu_meta.h>
+#include <ATen/ops/rrelu_with_noise_meta.h>
+#include <ATen/ops/rrelu_with_noise_backward_meta.h>
+#include <ATen/ops/rshift_meta.h>
+#include <ATen/ops/rsqrt_meta.h>
+#include <ATen/ops/rsub_meta.h>
+#include <ATen/ops/scalar_tensor_meta.h>
+#include <ATen/ops/scaled_dot_product_attention_meta.h>
+#include <ATen/ops/scatter_meta.h>
+#include <ATen/ops/scatter_add_meta.h>
+#include <ATen/ops/scatter_reduce_meta.h>
+#include <ATen/ops/searchsorted_meta.h>
+#include <ATen/ops/segment_reduce_meta.h>
+#include <ATen/ops/select_meta.h>
+#include <ATen/ops/select_backward_meta.h>
+#include <ATen/ops/select_copy_meta.h>
+#include <ATen/ops/select_scatter_meta.h>
+#include <ATen/ops/selu_meta.h>
+#include <ATen/ops/set_meta.h>
+#include <ATen/ops/set_data_meta.h>
+#include <ATen/ops/sgn_meta.h>
+#include <ATen/ops/sigmoid_meta.h>
+#include <ATen/ops/sigmoid_backward_meta.h>
+#include <ATen/ops/sign_meta.h>
+#include <ATen/ops/signbit_meta.h>
+#include <ATen/ops/silu_meta.h>
+#include <ATen/ops/silu_backward_meta.h>
+#include <ATen/ops/sin_meta.h>
+#include <ATen/ops/sinc_meta.h>
+#include <ATen/ops/sinh_meta.h>
+#include <ATen/ops/size_meta.h>
+#include <ATen/ops/slice_meta.h>
+#include <ATen/ops/slice_backward_meta.h>
+#include <ATen/ops/slice_copy_meta.h>
+#include <ATen/ops/slice_inverse_meta.h>
+#include <ATen/ops/slice_scatter_meta.h>
+#include <ATen/ops/slogdet_meta.h>
+#include <ATen/ops/slow_conv3d_meta.h>
+#include <ATen/ops/slow_conv3d_forward_meta.h>
+#include <ATen/ops/slow_conv_dilated2d_meta.h>
+#include <ATen/ops/slow_conv_dilated3d_meta.h>
+#include <ATen/ops/slow_conv_transpose2d_meta.h>
+#include <ATen/ops/slow_conv_transpose3d_meta.h>
+#include <ATen/ops/smm_meta.h>
+#include <ATen/ops/smooth_l1_loss_meta.h>
+#include <ATen/ops/smooth_l1_loss_backward_meta.h>
+#include <ATen/ops/soft_margin_loss_meta.h>
+#include <ATen/ops/soft_margin_loss_backward_meta.h>
+#include <ATen/ops/softmax_meta.h>
+#include <ATen/ops/softplus_meta.h>
+#include <ATen/ops/softplus_backward_meta.h>
+#include <ATen/ops/softshrink_meta.h>
+#include <ATen/ops/softshrink_backward_meta.h>
+#include <ATen/ops/sort_meta.h>
+#include <ATen/ops/sparse_bsc_tensor_meta.h>
+#include <ATen/ops/sparse_bsr_tensor_meta.h>
+#include <ATen/ops/sparse_compressed_tensor_meta.h>
+#include <ATen/ops/sparse_coo_tensor_meta.h>
+#include <ATen/ops/sparse_csc_tensor_meta.h>
+#include <ATen/ops/sparse_csr_tensor_meta.h>
+#include <ATen/ops/sparse_dim_meta.h>
+#include <ATen/ops/sparse_mask_meta.h>
+#include <ATen/ops/sparse_resize_meta.h>
+#include <ATen/ops/sparse_resize_and_clear_meta.h>
+#include <ATen/ops/sparse_sampled_addmm_meta.h>
+#include <ATen/ops/special_airy_ai_meta.h>
+#include <ATen/ops/special_bessel_j0_meta.h>
+#include <ATen/ops/special_bessel_j1_meta.h>
+#include <ATen/ops/special_bessel_y0_meta.h>
+#include <ATen/ops/special_bessel_y1_meta.h>
+#include <ATen/ops/special_chebyshev_polynomial_t_meta.h>
+#include <ATen/ops/special_chebyshev_polynomial_u_meta.h>
+#include <ATen/ops/special_chebyshev_polynomial_v_meta.h>
+#include <ATen/ops/special_chebyshev_polynomial_w_meta.h>
+#include <ATen/ops/special_digamma_meta.h>
+#include <ATen/ops/special_entr_meta.h>
+#include <ATen/ops/special_erf_meta.h>
+#include <ATen/ops/special_erfc_meta.h>
+#include <ATen/ops/special_erfcx_meta.h>
+#include <ATen/ops/special_erfinv_meta.h>
+#include <ATen/ops/special_exp2_meta.h>
+#include <ATen/ops/special_expit_meta.h>
+#include <ATen/ops/special_expm1_meta.h>
+#include <ATen/ops/special_gammainc_meta.h>
+#include <ATen/ops/special_gammaincc_meta.h>
+#include <ATen/ops/special_gammaln_meta.h>
+#include <ATen/ops/special_hermite_polynomial_h_meta.h>
+#include <ATen/ops/special_hermite_polynomial_he_meta.h>
+#include <ATen/ops/special_i0_meta.h>
+#include <ATen/ops/special_i0e_meta.h>
+#include <ATen/ops/special_i1_meta.h>
+#include <ATen/ops/special_i1e_meta.h>
+#include <ATen/ops/special_laguerre_polynomial_l_meta.h>
+#include <ATen/ops/special_legendre_polynomial_p_meta.h>
+#include <ATen/ops/special_log1p_meta.h>
+#include <ATen/ops/special_log_ndtr_meta.h>
+#include <ATen/ops/special_log_softmax_meta.h>
+#include <ATen/ops/special_logit_meta.h>
+#include <ATen/ops/special_logsumexp_meta.h>
+#include <ATen/ops/special_modified_bessel_i0_meta.h>
+#include <ATen/ops/special_modified_bessel_i1_meta.h>
+#include <ATen/ops/special_modified_bessel_k0_meta.h>
+#include <ATen/ops/special_modified_bessel_k1_meta.h>
+#include <ATen/ops/special_multigammaln_meta.h>
+#include <ATen/ops/special_ndtr_meta.h>
+#include <ATen/ops/special_ndtri_meta.h>
+#include <ATen/ops/special_polygamma_meta.h>
+#include <ATen/ops/special_psi_meta.h>
+#include <ATen/ops/special_round_meta.h>
+#include <ATen/ops/special_scaled_modified_bessel_k0_meta.h>
+#include <ATen/ops/special_scaled_modified_bessel_k1_meta.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_t_meta.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_u_meta.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_v_meta.h>
+#include <ATen/ops/special_shifted_chebyshev_polynomial_w_meta.h>
+#include <ATen/ops/special_sinc_meta.h>
+#include <ATen/ops/special_softmax_meta.h>
+#include <ATen/ops/special_spherical_bessel_j0_meta.h>
+#include <ATen/ops/special_xlog1py_meta.h>
+#include <ATen/ops/special_xlogy_meta.h>
+#include <ATen/ops/special_zeta_meta.h>
+#include <ATen/ops/split_meta.h>
+#include <ATen/ops/split_copy_meta.h>
+#include <ATen/ops/split_with_sizes_meta.h>
+#include <ATen/ops/split_with_sizes_copy_meta.h>
+#include <ATen/ops/sqrt_meta.h>
+#include <ATen/ops/square_meta.h>
+#include <ATen/ops/squeeze_meta.h>
+#include <ATen/ops/squeeze_copy_meta.h>
+#include <ATen/ops/sspaddmm_meta.h>
+#include <ATen/ops/stack_meta.h>
+#include <ATen/ops/std_meta.h>
+#include <ATen/ops/std_mean_meta.h>
+#include <ATen/ops/stft_meta.h>
+#include <ATen/ops/stride_meta.h>
+#include <ATen/ops/sub_meta.h>
+#include <ATen/ops/subtract_meta.h>
+#include <ATen/ops/sum_meta.h>
+#include <ATen/ops/sum_to_size_meta.h>
+#include <ATen/ops/svd_meta.h>
+#include <ATen/ops/swapaxes_meta.h>
+#include <ATen/ops/swapdims_meta.h>
+#include <ATen/ops/sym_constrain_range_meta.h>
+#include <ATen/ops/sym_constrain_range_for_size_meta.h>
+#include <ATen/ops/sym_numel_meta.h>
+#include <ATen/ops/sym_size_meta.h>
+#include <ATen/ops/sym_storage_offset_meta.h>
+#include <ATen/ops/sym_stride_meta.h>
+#include <ATen/ops/t_meta.h>
+#include <ATen/ops/t_copy_meta.h>
+#include <ATen/ops/take_meta.h>
+#include <ATen/ops/take_along_dim_meta.h>
+#include <ATen/ops/tan_meta.h>
+#include <ATen/ops/tanh_meta.h>
+#include <ATen/ops/tanh_backward_meta.h>
+#include <ATen/ops/tensor_split_meta.h>
+#include <ATen/ops/tensordot_meta.h>
+#include <ATen/ops/thnn_conv2d_meta.h>
+#include <ATen/ops/threshold_meta.h>
+#include <ATen/ops/threshold_backward_meta.h>
+#include <ATen/ops/tile_meta.h>
+#include <ATen/ops/to_meta.h>
+#include <ATen/ops/to_dense_meta.h>
+#include <ATen/ops/to_dense_backward_meta.h>
+#include <ATen/ops/to_mkldnn_meta.h>
+#include <ATen/ops/to_mkldnn_backward_meta.h>
+#include <ATen/ops/to_padded_tensor_meta.h>
+#include <ATen/ops/to_sparse_meta.h>
+#include <ATen/ops/to_sparse_bsc_meta.h>
+#include <ATen/ops/to_sparse_bsr_meta.h>
+#include <ATen/ops/to_sparse_csc_meta.h>
+#include <ATen/ops/to_sparse_csr_meta.h>
+#include <ATen/ops/topk_meta.h>
+#include <ATen/ops/trace_meta.h>
+#include <ATen/ops/trace_backward_meta.h>
+#include <ATen/ops/transpose_meta.h>
+#include <ATen/ops/transpose_copy_meta.h>
+#include <ATen/ops/trapezoid_meta.h>
+#include <ATen/ops/trapz_meta.h>
+#include <ATen/ops/triangular_solve_meta.h>
+#include <ATen/ops/tril_meta.h>
+#include <ATen/ops/tril_indices_meta.h>
+#include <ATen/ops/triplet_margin_loss_meta.h>
+#include <ATen/ops/triu_meta.h>
+#include <ATen/ops/triu_indices_meta.h>
+#include <ATen/ops/true_divide_meta.h>
+#include <ATen/ops/trunc_meta.h>
+#include <ATen/ops/type_as_meta.h>
+#include <ATen/ops/unbind_meta.h>
+#include <ATen/ops/unbind_copy_meta.h>
+#include <ATen/ops/unflatten_meta.h>
+#include <ATen/ops/unflatten_dense_tensors_meta.h>
+#include <ATen/ops/unfold_meta.h>
+#include <ATen/ops/unfold_backward_meta.h>
+#include <ATen/ops/unfold_copy_meta.h>
+#include <ATen/ops/uniform_meta.h>
+#include <ATen/ops/unique_consecutive_meta.h>
+#include <ATen/ops/unique_dim_meta.h>
+#include <ATen/ops/unique_dim_consecutive_meta.h>
+#include <ATen/ops/unsafe_chunk_meta.h>
+#include <ATen/ops/unsafe_split_meta.h>
+#include <ATen/ops/unsafe_split_with_sizes_meta.h>
+#include <ATen/ops/unsqueeze_meta.h>
+#include <ATen/ops/unsqueeze_copy_meta.h>
+#include <ATen/ops/upsample_bicubic2d_meta.h>
+#include <ATen/ops/upsample_bicubic2d_backward_meta.h>
+#include <ATen/ops/upsample_bilinear2d_meta.h>
+#include <ATen/ops/upsample_bilinear2d_backward_meta.h>
+#include <ATen/ops/upsample_linear1d_meta.h>
+#include <ATen/ops/upsample_linear1d_backward_meta.h>
+#include <ATen/ops/upsample_nearest1d_meta.h>
+#include <ATen/ops/upsample_nearest1d_backward_meta.h>
+#include <ATen/ops/upsample_nearest2d_meta.h>
+#include <ATen/ops/upsample_nearest2d_backward_meta.h>
+#include <ATen/ops/upsample_nearest3d_meta.h>
+#include <ATen/ops/upsample_nearest3d_backward_meta.h>
+#include <ATen/ops/upsample_trilinear3d_meta.h>
+#include <ATen/ops/upsample_trilinear3d_backward_meta.h>
+#include <ATen/ops/value_selecting_reduction_backward_meta.h>
+#include <ATen/ops/values_meta.h>
+#include <ATen/ops/values_copy_meta.h>
+#include <ATen/ops/vander_meta.h>
+#include <ATen/ops/var_meta.h>
+#include <ATen/ops/var_mean_meta.h>
+#include <ATen/ops/vdot_meta.h>
+#include <ATen/ops/view_meta.h>
+#include <ATen/ops/view_as_meta.h>
+#include <ATen/ops/view_as_complex_meta.h>
+#include <ATen/ops/view_as_complex_copy_meta.h>
+#include <ATen/ops/view_as_real_meta.h>
+#include <ATen/ops/view_as_real_copy_meta.h>
+#include <ATen/ops/view_copy_meta.h>
+#include <ATen/ops/vsplit_meta.h>
+#include <ATen/ops/vstack_meta.h>
+#include <ATen/ops/where_meta.h>
+#include <ATen/ops/xlogy_meta.h>
+#include <ATen/ops/xor_meta.h>
+#include <ATen/ops/zero_meta.h>
+#include <ATen/ops/zeros_meta.h>
+#include <ATen/ops/zeros_like_meta.h>
+
+namespace at {
+
+namespace meta {
+
+
+
+} // namespace meta
+} // namespace at

lib/python3.12/site-packages/torch/include/ATen/NestedTensorImpl.h

@@ -0,0 +1,286 @@
+#pragma once
+#include <ATen/MemoryOverlap.h>
+#include <ATen/Tensor.h>
+#include <c10/core/DispatchKey.h>
+#include <c10/core/DispatchKeySet.h>
+#include <c10/core/MemoryFormat.h>
+#include <c10/core/TensorImpl.h>
+#include <c10/util/ArrayRef.h>
+#include <c10/util/Exception.h>
+#include <c10/util/Metaprogramming.h>
+#include <c10/util/irange.h>
+
+namespace at::native {
+struct NestedTensorImpl;
+inline bool nested_tensor_impl_is_contiguous(const NestedTensorImpl* nt);
+int64_t get_numel_from_nested_size_tensor(const at::Tensor& tensor);
+at::Tensor construct_nested_strides(const at::Tensor& nested_size);
+at::Tensor construct_offsets(const at::Tensor& nested_size);
+
+struct TORCH_API NestedTensorImpl : public c10::TensorImpl {
+  explicit NestedTensorImpl(
+      Storage storage,
+      c10::DispatchKeySet key_set,
+      const caffe2::TypeMeta data_type,
+      at::Tensor nested_sizes,
+      at::Tensor nested_strides,
+      at::Tensor storage_offsets);
+
+  explicit NestedTensorImpl(
+      const at::Tensor& buffer,
+      at::Tensor nested_sizes,
+      at::Tensor nested_strides,
+      at::Tensor storage_offsets);
+  // assume contiguous, `nested_strides` and `offsets`
+  // can be infered from `nested_sizes`
+  explicit NestedTensorImpl(
+      const at::Tensor& buffer,
+      const at::Tensor& nested_sizes);
+
+  // This constructor is used creating view tensors from nested tensors
+  explicit NestedTensorImpl(
+      c10::TensorImpl::ImplType impl_type,
+      const at::Tensor& base_tensor,
+      at::Tensor nested_sizes,
+      at::Tensor nested_strides,
+      at::Tensor storage_offsets);
+
+  // TODO: don't expose private implementation details like this; in
+  // particular, resizing this tensor will mess up our dim() and
+  // callers cannot fix it.
+  const Tensor& get_nested_sizes() const {
+    return nested_sizes_;
+  }
+  // TODO: don't expose private implementation details like this
+  const Tensor& get_nested_strides() const {
+    return nested_strides_;
+  }
+  const Tensor& get_storage_offsets() const {
+    return storage_offsets_;
+  }
+  // Returns nullopt if the ith dimension is irregular. The ith dimension
+  // of a NestedTensor is regular if the unbound tensors match in
+  // size at the (i-1)th dimension.
+  std::optional<int64_t> opt_size(int64_t d) const;
+
+  int64_t size(int64_t d) const {
+    std::optional<int64_t> optional_size = this->opt_size(d);
+    TORCH_CHECK(
+        optional_size.has_value(),
+        "Given dimension ",
+        d,
+        " is irregular and does not have a size.");
+    return *optional_size;
+  }
+  /**
+   * Return a view of the nested tensor as a 1 dimensional contiguous tensor.
+   *
+   * The buffer tensor created by this function shares the same storage_impl as
+   * the original nested tensor, and therefore can be seen as a view.
+   *
+   * @return A newly constructed view tensor
+   */
+  at::Tensor get_buffer() const {
+    TORCH_CHECK(
+        nested_tensor_impl_is_contiguous(this),
+        "NestedTensor must be contiguous to get buffer.");
+    return get_unsafe_storage_as_tensor();
+  }
+  /**
+   * If possible use get_buffer() instead. This function returns the storage
+   * as a tensor directly, which is not safe to use in general. If using this
+   * function, The caller must ensure to account for nested_sizes,
+   * nested_strides and storage_offsets.
+   *
+   * @return A newly constructed view tensor
+   */
+  at::Tensor get_unsafe_storage_as_tensor() const {
+    auto buffer_key_set_ = generate_buffer_key_set();
+    const auto buffer_size = get_buffer_size();
+    auto buffer_tensor_impl = c10::make_intrusive<TensorImpl>(
+        c10::TensorImpl::VIEW, Storage(storage_), buffer_key_set_, data_type_);
+    buffer_tensor_impl->set_sizes_contiguous(
+        c10::makeArrayRef(static_cast<int64_t>(buffer_size)));
+    return Tensor(buffer_tensor_impl);
+  }
+
+  size_t get_buffer_size() const {
+    return storage_.nbytes() / data_type_.itemsize();
+  }
+
+ protected:
+  const char* tensorimpl_type_name() const override;
+
+  // TODO: numel_custom and is_contiguous_custom can be profitably overridden
+  // with real implementations
+  int64_t numel_custom() const override;
+  c10::SymInt sym_numel_custom() const override;
+  bool is_contiguous_custom(MemoryFormat) const override;
+  int64_t size_custom(int64_t d) const override {
+    return this->size(d);
+  }
+  c10::SymInt sym_size_custom(int64_t d) const override {
+    return c10::SymInt{this->size(d)};
+  }
+  IntArrayRef sizes_custom() const override;
+  c10::SymIntArrayRef sym_sizes_custom() const override;
+  IntArrayRef strides_custom() const override;
+  c10::SymIntArrayRef sym_strides_custom() const override;
+
+  // this one is real
+  int64_t dim_custom() const override;
+
+  c10::intrusive_ptr<TensorImpl> shallow_copy_and_detach(
+      const c10::VariableVersion& version_counter,
+      bool allow_tensor_metadata_change) const override;
+
+  c10::intrusive_ptr<TensorImpl> shallow_copy_and_detach(
+      c10::VariableVersion&& version_counter,
+      bool allow_tensor_metadata_change) const override;
+
+  void shallow_copy_from(const c10::intrusive_ptr<TensorImpl>& impl) override {
+    copy_tensor_metadata(
+        /*src_impl=*/impl.get(),
+        /*dest_impl=*/this,
+        /*version_counter=*/version_counter(),
+        /*allow_tensor_metadata_change=*/allow_tensor_metadata_change());
+  }
+
+ private:
+  // Must be called after any changes to our dim() to sync the state
+  // to TensorImpl.
+  void refresh_dim();
+
+  // NOLINTNEXTLINE(cppcoreguidelines-avoid-const-or-ref-data-members)
+  const at::Tensor nested_sizes_, nested_strides_;
+  // The starting positions of the underlying tensors in contiguous buffer
+  // i.e. the buffer memory offsets to get the underlying tensors
+  // The reason to keep this metadata is that, without strong enough constraint
+  // it cannot be derived from `nested_sizes_`
+  // and `nested_strides_`:
+  // 1. when buffer has blanks, e.g. [tensor1, blank, tensor2]
+  //    this can happen e.g. after slicing a nested tensor
+  // 2. when multiple tensors share a same memory
+  // 3. when the nesting ordering is changed, e.g. [tensor1, tensor3, tensor2]
+  // Some strong enough constraints are:
+  // 1. every underlying tensor is contiguous in memory
+  //    && nesting in ascending order
+  // NOLINTNEXTLINE(cppcoreguidelines-avoid-const-or-ref-data-members)
+  const at::Tensor storage_offsets_;
+  // NOTE: -1 here means the size is missing
+  // Optional to allow it to be computed lazily from nested.
+  // TODO: maybe we can remove this metadata since
+  //       we can compute it from `nested_sizes_`
+  mutable std::optional<std::vector<int64_t>> opt_sizes_;
+
+  template <typename VariableVersion>
+  c10::intrusive_ptr<TensorImpl> shallow_copy_and_detach_core(
+      VariableVersion&& version_counter,
+      bool allow_tensor_metadata_change) const;
+
+  /**
+   * Generates a non-nested key_set from a nested tensor.
+   *
+   * For many nested tensor kernel implementations a buffer tensor
+   * is generated and redispatched to a non-nested kernel this function
+   * generates the key set used by that buffer tensor
+   *
+   * @return Appropriate key set for non-nested tensor
+   */
+  inline c10::DispatchKeySet generate_buffer_key_set() const {
+    auto buffer_key_set = this->key_set();
+    const bool Autograd = buffer_key_set.has_any(c10::autograd_dispatch_keyset);
+    // Remove nested tensor specific keys
+    buffer_key_set = buffer_key_set -
+        c10::DispatchKeySet{
+            c10::DispatchKey::NestedTensor,
+            c10::DispatchKey::AutogradNestedTensor};
+
+    // Add dense tensor specific keys
+    buffer_key_set =
+        buffer_key_set | c10::DispatchKeySet{c10::DispatchKey::Dense};
+    buffer_key_set = Autograd
+        ? c10::DispatchKeySet{c10::DispatchKey::Autograd} | buffer_key_set
+        : buffer_key_set;
+
+    return buffer_key_set;
+  }
+};
+
+inline NestedTensorImpl* get_nested_tensor_impl_or_null(
+    const at::Tensor& tensor) {
+  if (tensor.is_nested()) {
+    return static_cast<NestedTensorImpl*>(tensor.unsafeGetTensorImpl());
+  }
+  return nullptr;
+}
+
+inline NestedTensorImpl* get_nested_tensor_impl(const at::Tensor& tensor) {
+  TORCH_CHECK(
+      tensor.is_nested(), "get_nested_tensor_impl requires a NestedTensor.");
+  return static_cast<NestedTensorImpl*>(tensor.unsafeGetTensorImpl());
+}
+
+inline bool nested_tensor_impl_is_contiguous(const NestedTensorImpl* nt) {
+  int64_t ntensors = nt->size(0);
+  if (ntensors == 0) {
+    return true;
+  }
+  const Tensor &sizemat = nt->get_nested_sizes(),
+               &stridemat = nt->get_nested_strides();
+  const int64_t* offsets_ptr =
+      nt->get_storage_offsets().const_data_ptr<int64_t>();
+  int64_t orig_dim = sizemat.size(1);
+  // nesting scalars
+  if (orig_dim == 0) {
+    // each scalar must be contiguous
+    // if there is blank memory between underlying scalars
+    for (int64_t i = 0; i < ntensors; i++) {
+      if (offsets_ptr[i] != i) {
+        return false;
+      }
+    }
+  }
+  // nesting tensors
+  else {
+    // if any underlying tensor is non-contiguous
+    const int64_t *sizemat_ptr = sizemat.const_data_ptr<int64_t>(),
+                  *stridemat_ptr = stridemat.const_data_ptr<int64_t>();
+    for (int64_t i = 0; i < ntensors; i++) {
+      if (stridemat_ptr[orig_dim - 1] != 1) {
+        return false;
+      }
+      int64_t product = sizemat_ptr[orig_dim - 1];
+      for (int64_t j = orig_dim - 2; j >= 0; j--) {
+        if (stridemat_ptr[j] != product) {
+          return false;
+        }
+        product *= sizemat_ptr[j];
+      }
+      sizemat_ptr += orig_dim;
+      stridemat_ptr += orig_dim;
+    }
+    // if there is blank memory between underlying tensors
+    if (offsets_ptr[0] != 0) {
+      return false;
+    }
+    sizemat_ptr = sizemat.const_data_ptr<int64_t>();
+    stridemat_ptr = stridemat.const_data_ptr<int64_t>();
+    for (int64_t i = 1; i < ntensors; i++) {
+      if (offsets_ptr[i] !=
+          offsets_ptr[i - 1] + *sizemat_ptr * *stridemat_ptr) {
+        return false;
+      }
+      sizemat_ptr += orig_dim;
+      stridemat_ptr += orig_dim;
+    }
+  }
+  // everything is fine
+  return true;
+}
+
+inline const at::Tensor& get_nested_sizes(const at::Tensor& tensor) {
+  return get_nested_tensor_impl(tensor)->get_nested_sizes();
+}
+
+} // namespace at::native

lib/python3.12/site-packages/torch/include/ATen/NumericUtils.h

@@ -0,0 +1,203 @@
+#pragma once
+
+#ifdef __HIPCC__
+#include <hip/hip_runtime.h>
+#endif
+
+#include <c10/macros/Macros.h>
+#include <c10/util/BFloat16.h>
+#include <c10/util/Float8_e4m3fn.h>
+#include <c10/util/Float8_e4m3fnuz.h>
+#include <c10/util/Float8_e5m2.h>
+#include <c10/util/Float8_e5m2fnuz.h>
+#include <c10/util/Half.h>
+#include <c10/util/complex.h>
+
+#include <cmath>
+#include <type_traits>
+
+namespace at {
+
+// std::isnan isn't performant to use on integral types; it will
+// (uselessly) convert to floating point and then do the test.
+// This function is.
+
+template <typename T, std::enable_if_t<std::is_integral_v<T>, int> = 0>
+inline C10_HOST_DEVICE bool _isnan(T /*val*/) {
+  return false;
+}
+
+template <typename T, std::enable_if_t<std::is_floating_point_v<T>, int> = 0>
+inline C10_HOST_DEVICE bool _isnan(T val) {
+#if defined(__CUDACC__) || defined(__HIPCC__)
+  return ::isnan(val);
+#else
+  return std::isnan(val);
+#endif
+}
+
+template <typename T, std::enable_if_t<c10::is_complex<T>::value, int> = 0>
+inline C10_HOST_DEVICE bool _isnan(T val) {
+  return std::isnan(val.real()) || std::isnan(val.imag());
+}
+
+template <typename T, std::enable_if_t<std::is_same_v<T, at::Half>, int> = 0>
+inline C10_HOST_DEVICE bool _isnan(T val) {
+  return at::_isnan(static_cast<float>(val));
+}
+
+template <
+    typename T,
+    std::enable_if_t<std::is_same_v<T, at::BFloat16>, int> = 0>
+inline C10_HOST_DEVICE bool _isnan(at::BFloat16 val) {
+  return at::_isnan(static_cast<float>(val));
+}
+
+inline C10_HOST_DEVICE bool _isnan(at::BFloat16 val) {
+  return at::_isnan(static_cast<float>(val));
+}
+
+template <
+    typename T,
+    std::enable_if_t<std::is_same_v<T, at::Float8_e5m2>, int> = 0>
+inline C10_HOST_DEVICE bool _isnan(T val) {
+  return val.isnan();
+}
+
+template <
+    typename T,
+    std::enable_if_t<std::is_same_v<T, at::Float8_e4m3fn>, int> = 0>
+inline C10_HOST_DEVICE bool _isnan(T val) {
+  return val.isnan();
+}
+
+template <
+    typename T,
+    std::enable_if_t<std::is_same_v<T, at::Float8_e5m2fnuz>, int> = 0>
+inline C10_HOST_DEVICE bool _isnan(T val) {
+  return val.isnan();
+}
+
+template <
+    typename T,
+    std::enable_if_t<std::is_same_v<T, at::Float8_e4m3fnuz>, int> = 0>
+inline C10_HOST_DEVICE bool _isnan(T val) {
+  return val.isnan();
+}
+
+// std::isinf isn't performant to use on integral types; it will
+// (uselessly) convert to floating point and then do the test.
+// This function is.
+
+template <typename T, std::enable_if_t<std::is_integral_v<T>, int> = 0>
+inline C10_HOST_DEVICE bool _isinf(T /*val*/) {
+  return false;
+}
+
+template <typename T, std::enable_if_t<std::is_floating_point_v<T>, int> = 0>
+inline C10_HOST_DEVICE bool _isinf(T val) {
+#if defined(__CUDACC__) || defined(__HIPCC__)
+  return ::isinf(val);
+#else
+  return std::isinf(val);
+#endif
+}
+
+inline C10_HOST_DEVICE bool _isinf(at::Half val) {
+  return at::_isinf(static_cast<float>(val));
+}
+
+inline C10_HOST_DEVICE bool _isinf(at::BFloat16 val) {
+  return at::_isinf(static_cast<float>(val));
+}
+
+inline C10_HOST_DEVICE bool _isinf(at::Float8_e5m2 val) {
+  return val.isinf();
+}
+
+inline C10_HOST_DEVICE bool _isinf(at::Float8_e4m3fn val [[maybe_unused]]) {
+  return false;
+}
+
+inline C10_HOST_DEVICE bool _isinf(at::Float8_e5m2fnuz val [[maybe_unused]]) {
+  return false;
+}
+
+inline C10_HOST_DEVICE bool _isinf(at::Float8_e4m3fnuz val [[maybe_unused]]) {
+  return false;
+}
+
+template <typename T>
+C10_HOST_DEVICE inline T exp(T x) {
+  static_assert(
+      !std::is_same_v<T, double>,
+      "this template must be used with float or less precise type");
+#if defined(__CUDA_ARCH__) || defined(__HIP_ARCH__)
+  // use __expf fast approximation for peak bandwidth
+  return __expf(x);
+#else
+  return ::exp(x);
+#endif
+}
+
+template <>
+C10_HOST_DEVICE inline double exp<double>(double x) {
+  return ::exp(x);
+}
+
+template <typename T>
+C10_HOST_DEVICE inline T log(T x) {
+  static_assert(
+      !std::is_same_v<T, double>,
+      "this template must be used with float or less precise type");
+#if defined(__CUDA_ARCH__) || defined(__HIP_ARCH__)
+  // use __logf fast approximation for peak bandwidth
+  return __logf(x);
+#else
+  return ::log(x);
+#endif
+}
+
+template <>
+C10_HOST_DEVICE inline double log<double>(double x) {
+  return ::log(x);
+}
+
+template <typename T>
+C10_HOST_DEVICE inline T log1p(T x) {
+  static_assert(
+      !std::is_same_v<T, double>,
+      "this template must be used with float or less precise type");
+#if defined(__CUDA_ARCH__) || defined(__HIP_ARCH__)
+  // use __logf fast approximation for peak bandwidth
+  // NOTE: There is no __log1pf so unfortunately we lose precision.
+  return __logf(1.0f + x);
+#else
+  return ::log1p(x);
+#endif
+}
+
+template <>
+C10_HOST_DEVICE inline double log1p<double>(double x) {
+  return ::log1p(x);
+}
+
+template <typename T>
+C10_HOST_DEVICE inline T tan(T x) {
+  static_assert(
+      !std::is_same_v<T, double>,
+      "this template must be used with float or less precise type");
+#if defined(__CUDA_ARCH__) || defined(__HIP_ARCH__)
+  // use __tanf fast approximation for peak bandwidth
+  return __tanf(x);
+#else
+  return ::tan(x);
+#endif
+}
+
+template <>
+C10_HOST_DEVICE inline double tan<double>(double x) {
+  return ::tan(x);
+}
+
+} // namespace at

lib/python3.12/site-packages/torch/include/ATen/OpaqueTensorImpl.h

@@ -0,0 +1,187 @@
+#pragma once
+
+#include <c10/core/MemoryFormat.h>
+#include <c10/core/SymIntArrayRef.h>
+#include <c10/core/TensorImpl.h>
+#include <c10/util/Exception.h>
+
+namespace at {
+
+// An "Opaque" TensorImpl -- there are no strides and (for now)
+// even data() is not supported (thus no pointer arithmetic).
+
+// NOTE: We could allow data() in the future, but would have to ensure pointer
+// arithmetic code is properly guarded.
+//
+// NOTE: This does not support resize_ (and other metadata-changing ops) because
+// of `shallow_copy_and_detach`. We would need to define an interface to
+// "shallow copy" in order to add support.
+
+template <typename OpaqueHandle>
+struct TORCH_API OpaqueTensorImpl : public TensorImpl {
+  // public constructor for now...
+  OpaqueTensorImpl(
+      at::DispatchKeySet key_set,
+      const caffe2::TypeMeta data_type,
+      c10::Device device,
+      OpaqueHandle opaque_handle,
+      c10::IntArrayRef sizes,
+      bool is_non_overlapping_and_dense = true)
+      : TensorImpl(key_set, data_type, device),
+        opaque_handle_(std::move(opaque_handle)) {
+    set_storage_access_should_throw();
+    set_custom_sizes_strides(SizesStridesPolicy::CustomStrides);
+    sizes_and_strides_.set_sizes(sizes);
+    refresh_numel();
+    // NOLINTNEXTLINE(cppcoreguidelines-prefer-member-initializer)
+    is_non_overlapping_and_dense_ = is_non_overlapping_and_dense;
+  }
+
+  // Destructor doesn't call release_resources because it's
+  // unnecessary; don't forget to change that if needed!
+  void release_resources() override {
+    TensorImpl::release_resources();
+    opaque_handle_ = {};
+  }
+
+  void set_size(int64_t dim, int64_t new_size) override {
+    TORCH_CHECK(false, "opaque tensors do not have set_size");
+  }
+
+  void set_stride(int64_t dim, int64_t new_stride) override {
+    TORCH_CHECK(false, "opaque tensors do not have set_stride");
+  }
+
+  void set_storage_offset(int64_t storage_offset) override {
+    TORCH_CHECK(false, "opaque tensors do not have set_storage_offset");
+  }
+
+#ifdef DEBUG
+  bool has_storage() const override {
+    TORCH_INTERNAL_ASSERT_DEBUG_ONLY(
+        !storage_, "OpaqueTensorImpl assumes that storage_ is never set");
+    return false;
+  }
+#endif
+
+  /**
+   * Return a TensorImpl that is a shallow-copy of this TensorImpl.
+   *
+   * For usage of `version_counter` and `allow_tensor_metadata_change`,
+   * see NOTE [ TensorImpl Shallow-Copying ].
+   */
+  c10::intrusive_ptr<TensorImpl> shallow_copy_and_detach(
+      const c10::VariableVersion& version_counter,
+      bool allow_tensor_metadata_change) const override {
+    auto impl = c10::make_intrusive<OpaqueTensorImpl<OpaqueHandle>>(
+        key_set(),
+        dtype(),
+        device(),
+        opaque_handle_,
+        sizes_and_strides_.sizes_arrayref());
+    copy_tensor_metadata(
+        /*src_opaque_impl=*/this,
+        /*dest_opaque_impl=*/impl.get(),
+        /*version_counter=*/version_counter,
+        /*allow_tensor_metadata_change=*/allow_tensor_metadata_change);
+    impl->refresh_numel();
+    return impl;
+  }
+
+  /**
+   * Return a TensorImpl that is a shallow-copy of this TensorImpl.
+   *
+   * For usage of `version_counter` and `allow_tensor_metadata_change`,
+   * see NOTE [ TensorImpl Shallow-Copying ].
+   */
+  c10::intrusive_ptr<TensorImpl> shallow_copy_and_detach(
+      c10::VariableVersion&& version_counter,
+      bool allow_tensor_metadata_change) const override {
+    auto impl = c10::make_intrusive<OpaqueTensorImpl<OpaqueHandle>>(
+        key_set(),
+        dtype(),
+        device(),
+        opaque_handle_,
+        sizes_and_strides_.sizes_arrayref());
+    copy_tensor_metadata(
+        /*src_opaque_impl=*/this,
+        /*dest_opaque_impl=*/impl.get(),
+        /*version_counter=*/std::move(version_counter),
+        /*allow_tensor_metadata_change=*/allow_tensor_metadata_change);
+    impl->refresh_numel();
+    return impl;
+  }
+
+  /**
+   * Shallow-copies data from another TensorImpl into this TensorImpl.
+   *
+   * For why this function doesn't check this TensorImpl's
+   * `allow_tensor_metadata_change_`, see NOTE [ TensorImpl Shallow-Copying ].
+   */
+  void shallow_copy_from(const c10::intrusive_ptr<TensorImpl>& impl) override {
+    AT_ASSERT(has_compatible_shallow_copy_type(impl->key_set()));
+    auto opaque_impl =
+        static_cast<const OpaqueTensorImpl<OpaqueHandle>*>(impl.get());
+    copy_tensor_metadata(
+        /*src_impl=*/opaque_impl,
+        /*dest_impl=*/this,
+        /*version_counter=*/version_counter(),
+        /*allow_tensor_metadata_change=*/allow_tensor_metadata_change());
+    refresh_numel();
+  }
+
+  const OpaqueHandle& opaque_handle() const {
+    return opaque_handle_;
+  }
+
+  OpaqueHandle& unsafe_opaque_handle() {
+    return opaque_handle_;
+  }
+
+ protected:
+  /**
+   * Copy the tensor metadata fields (e.g. sizes / strides / storage pointer /
+   * storage_offset) from one TensorImpl to another TensorImpl.
+   *
+   * For usage of `version_counter` and `allow_tensor_metadata_change`, see NOTE
+   * [ TensorImpl Shallow-Copying ].
+   */
+  static void copy_tensor_metadata(
+      const OpaqueTensorImpl<OpaqueHandle>* src_opaque_impl,
+      OpaqueTensorImpl<OpaqueHandle>* dest_opaque_impl,
+      const c10::VariableVersion& version_counter,
+      bool allow_tensor_metadata_change) {
+    TensorImpl::copy_tensor_metadata(
+        src_opaque_impl,
+        dest_opaque_impl,
+        version_counter,
+        allow_tensor_metadata_change);
+
+    // OpaqueTensorImpl-specific fields.
+    dest_opaque_impl->opaque_handle_ = src_opaque_impl->opaque_handle_;
+  }
+
+  static void copy_tensor_metadata(
+      const OpaqueTensorImpl<OpaqueHandle>* src_opaque_impl,
+      OpaqueTensorImpl<OpaqueHandle>* dest_opaque_impl,
+      c10::VariableVersion&& version_counter,
+      bool allow_tensor_metadata_change) {
+    TensorImpl::copy_tensor_metadata(
+        src_opaque_impl,
+        dest_opaque_impl,
+        std::move(version_counter),
+        allow_tensor_metadata_change);
+
+    // OpaqueTensorImpl-specific fields.
+    dest_opaque_impl->opaque_handle_ = src_opaque_impl->opaque_handle_;
+  }
+
+ private:
+  const char* tensorimpl_type_name() const override {
+    return "OpaqueTensorImpl";
+  }
+
+  OpaqueHandle opaque_handle_;
+};
+
+} // namespace at