Revert "Add randint_like tensor overload for high (pytorch#154899)"

This reverts commit 72fe1d5.

Reverted pytorch#154899 on behalf of https://github.com/seemethere due to Failing internal tests see https://fburl.com/diff/bai044ob ([comment](pytorch#154899 (comment)))

Author: pytorchmergebot

aten/src/ATen/native/TensorFactories.cpp

@@ -1212,28 +1212,6 @@ Tensor randint_like(
   return result.random_(0, high, std::nullopt);
 }
 
-Tensor randint_like(
-    const Tensor& self,
-    const Tensor& high,
-    std::optional<ScalarType> dtype,
-    std::optional<Layout> layout,
-    std::optional<Device> device,
-    std::optional<bool> pin_memory,
-    std::optional<c10::MemoryFormat> optional_memory_format) {
-  TORCH_CHECK(
-      high.numel() == 1 && high.ndimension() == 0 && high.device().is_cpu(),
-      "high must be a scalar tensor and on CPU");
-  int64_t high_scalar = high.item<int64_t>();
-  return at::native::randint_like(
-      self,
-      high_scalar,
-      dtype,
-      layout,
-      device,
-      pin_memory,
-      optional_memory_format);
-}
-
 Tensor randint_like(
     const Tensor& self,
     int64_t low,

aten/src/ATen/native/native_functions.yaml

@@ -4781,14 +4781,6 @@
     CompositeExplicitAutograd: randint_like
   autogen: randint_like.out
 
-- func: randint_like.Tensor(Tensor self, Tensor high, *, ScalarType? dtype=None, Layout? layout=None, Device? device=None, bool? pin_memory=None, MemoryFormat? memory_format=None) -> Tensor
-  tags: nondeterministic_seeded
-  dispatch:
-    # NB: Although this composite mutates on the inside, it is
-    # non-differentiable so NonFunctional doesn't apply
-    CompositeExplicitAutograd: randint_like
-  autogen: randint_like.Tensor_out
-
 - func: randint_like.low_dtype(Tensor self, SymInt low, SymInt high, *, ScalarType? dtype=None, Layout? layout=None, Device? device=None, bool? pin_memory=None, MemoryFormat? memory_format=None) -> Tensor
   tags: nondeterministic_seeded
   dispatch:

test/dynamo/test_misc.py

@@ -12603,17 +12603,6 @@ def f():
         res = opt_f()
         self.assertEqual(ref, res)
 
-    def test_randint_no_graphbreak(self):
-        @torch.compile(backend="aot_eager", fullgraph=True)
-        def f(actions, n_act, epsilon=0.1):
-            actions_random = torch.randint_like(actions, n_act)
-
-            return actions_random
-
-        x = torch.ones([1], dtype=torch.int64)
-        y = torch.tensor(5)
-        f(x, y)
-
 
 devices = ("cuda", "hpu")
 instantiate_device_type_tests(MiscTestsDevice, globals(), only_for=devices)

test/expect/HasDecompTest.test_has_decomposition.expect

@@ -1094,8 +1094,6 @@ aten::randint.low_generator_out
 aten::randint.low_out
 aten::randint.out
 aten::randint_like
-aten::randint_like.Tensor
-aten::randint_like.Tensor_out
 aten::randint_like.low_dtype
 aten::randint_like.low_dtype_out
 aten::randint_like.out

torch/_meta_registrations.py

@@ -3505,11 +3505,6 @@ def meta_addbmm(self, batch1, batch2, *, beta=1, alpha=1):
     return self.new_empty(self.size())
 
 
-@register_meta([aten.randint_like.Tensor])
-def meta_randint_like(self, high, **kwargs):
-    return self.new_empty(self.size())
-
-
 @register_meta([aten._fused_adam_.default, aten._fused_adamw_.default])
 def meta__fused_adam_(
     self,

torch/_subclasses/fake_impls.py

@@ -68,8 +68,6 @@ def is_noncontiguous_supported(device):
     aten.randn_like.default,
     aten.randn_like.out,
     aten.randint_like.default,
-    aten.randint_like.Tensor,
-    aten.randint_like.Tensor_out,
     aten.randint_like.out,
     aten.randint_like.low_dtype,
     aten.randint_like.low_dtype_out,