mirror of
https://github.com/pytorch/pytorch.git
synced 2025-10-20 21:14:14 +08:00
c8c5187e85
Addresses numeric differences between eager and compile in https://github.com/pytorch/pytorch/issues/141753 Pull Request resolved: https://github.com/pytorch/pytorch/pull/164144 Approved by: https://github.com/bobrenjc93
553 lines
23 KiB
Python
553 lines
23 KiB
Python
# Owner(s): ["module: tests"]
|
|
"""
|
|
Fuzzer-discovered eager/compile divergence test cases.
|
|
|
|
All tests are marked as xfail since they represent known compilation bugs.
|
|
|
|
IF YOU ARE HERE YOU LIKELY DIDN'T DO ANYTHING WRONG. In fact, you probably did something right!
|
|
All of these tests are associated with bugs the fuzzer found. If one of these tests starts failing due to your PR,
|
|
it actually means your PR fixed the bug! Feel free to delete the test and close out the issue linked from the test.
|
|
"""
|
|
|
|
import pytest
|
|
|
|
import torch
|
|
from torch.testing._internal.common_utils import run_tests, TestCase
|
|
|
|
|
|
class TestFuzzerCompileIssues(TestCase):
|
|
"""Test cases for fuzzer-discovered eager/compile divergence issues."""
|
|
|
|
def setUp(self):
|
|
"""Configure common test settings."""
|
|
torch._dynamo.config.capture_scalar_outputs = True
|
|
torch._dynamo.config.capture_dynamic_output_shape_ops = True
|
|
torch._inductor.config.emulate_precision_casts = True
|
|
|
|
@pytest.mark.xfail(reason="Issue #164484")
|
|
def test_fuzzer_issue_164484(self):
|
|
torch.manual_seed(9157)
|
|
|
|
def foo(arg0, arg1, arg2, arg3):
|
|
var_node_2 = torch.full((14, 16), 1.158473253250122, dtype=torch.float32)
|
|
var_node_1 = torch.nn.functional.relu(var_node_2)
|
|
var_node_6 = torch.full((14, 1), -0.94140625, dtype=torch.bfloat16)
|
|
var_node_7 = arg0 # size=(1, 16), stride=(16, 1), dtype=bfloat16
|
|
var_node_5 = torch.matmul(
|
|
var_node_6.to(torch.bfloat16), var_node_7.to(torch.bfloat16)
|
|
)
|
|
var_node_9 = torch.full((16,), 0.76953125, dtype=torch.bfloat16)
|
|
var_node_8 = torch.reshape(var_node_9, [16])
|
|
var_node_11 = torch.full((16,), 2.4375, dtype=torch.bfloat16)
|
|
var_node_10 = torch.reshape(var_node_11, [16])
|
|
var_node_4 = torch.cat([var_node_5, var_node_8, var_node_10], dim=1)
|
|
var_node_12 = arg1 # size=(14, 48), stride=(48, 1), dtype=bfloat16
|
|
var_node_3 = torch.sub(var_node_4, var_node_12)
|
|
var_node_0 = torch.add(var_node_1, var_node_3)
|
|
var_node_14 = torch.full((14, 48), 1.4375, dtype=torch.bfloat16)
|
|
var_node_13 = torch.nn.functional.layer_norm(var_node_14, [48])
|
|
result = torch.add(var_node_0, var_node_13)
|
|
output = result + arg2 + arg3
|
|
return output
|
|
|
|
arg0 = torch.rand(
|
|
[1, 16], dtype=torch.bfloat16, device="cuda", requires_grad=True
|
|
)
|
|
arg1 = torch.rand(
|
|
[14, 48], dtype=torch.bfloat16, device="cuda", requires_grad=True
|
|
)
|
|
arg2 = torch.tensor(
|
|
0.0, dtype=torch.bfloat16, device="cuda", requires_grad=True
|
|
)
|
|
arg3 = torch.tensor(
|
|
0.0, dtype=torch.bfloat16, device="cuda", requires_grad=True
|
|
)
|
|
|
|
out_eager = foo(arg0, arg1, arg2, arg3)
|
|
out_eager.sum().backward()
|
|
print("Eager Success! ✅")
|
|
compiled_foo = torch.compile(foo, fullgraph=True, dynamic=True)
|
|
out_compiled = compiled_foo(arg0, arg1, arg2, arg3)
|
|
out_compiled.sum().backward()
|
|
print("Compile Success! ✅")
|
|
|
|
@pytest.mark.xfail(reason="Issue #164186")
|
|
def test_fuzzer_issue_164186(self):
|
|
torch.manual_seed(0)
|
|
|
|
def foo(arg0):
|
|
t0 = arg0 # size=(714, 33), stride=(33, 1), dtype=float16, device=cuda
|
|
t1 = t0.clone()
|
|
t1.zero_()
|
|
t2 = t1.contiguous().view((34, 9, 77))
|
|
t3 = t2.clone()
|
|
t3.zero_()
|
|
output = t3
|
|
return output
|
|
|
|
arg0 = torch.rand(
|
|
[714, 33], dtype=torch.float16, device="cuda", requires_grad=True
|
|
)
|
|
|
|
out_eager = foo(arg0)
|
|
out_eager.sum().backward()
|
|
print("Eager Success! ✅")
|
|
compiled_foo = torch.compile(foo, fullgraph=True, dynamic=True)
|
|
out_compiled = compiled_foo(arg0)
|
|
out_compiled.sum().backward()
|
|
print("Compile Success! ✅")
|
|
|
|
@pytest.mark.xfail(reason="Issue #164185")
|
|
def test_fuzzer_issue_164185(self):
|
|
torch.manual_seed(0)
|
|
|
|
def foo(arg0, arg1, arg2):
|
|
t0 = arg0 # size=(349200, 5), stride=(5, 1), dtype=bfloat16, device=cuda
|
|
t1 = t0.mean(
|
|
dim=1
|
|
) # size=(349200,), stride=(1,), dtype=bfloat16, device=cuda
|
|
t2 = arg1 # size=(), stride=(), dtype=int64, device=cuda
|
|
t3 = arg2 # size=(50000, 349200), stride=(50000, 1), dtype=bfloat16, device=cuda
|
|
t4 = torch.nn.functional.embedding(
|
|
torch.clamp(t2, 0, t3.size(0) - 1).to(torch.long), t3
|
|
)
|
|
t5 = torch.pow(torch.pow(torch.pow(torch.pow(t1, t4), t4), t1), t1)
|
|
t6 = t5.contiguous().view((75, 97, 48))
|
|
output = t6
|
|
return output
|
|
|
|
arg0 = torch.rand(
|
|
[349200, 5], dtype=torch.bfloat16, device="cuda", requires_grad=True
|
|
)
|
|
arg1 = torch.randint(0, 50000, [], dtype=torch.int64, device="cuda")
|
|
arg2 = torch.rand(
|
|
[50000, 349200], dtype=torch.bfloat16, device="cuda", requires_grad=True
|
|
)
|
|
|
|
out_eager = foo(arg0, arg1, arg2)
|
|
out_eager.sum().backward()
|
|
print("Eager Success! ✅")
|
|
compiled_foo = torch.compile(foo, fullgraph=True, dynamic=True)
|
|
out_compiled = compiled_foo(arg0, arg1, arg2)
|
|
out_compiled.sum().backward()
|
|
print("Compile Success! ✅")
|
|
|
|
@pytest.mark.xfail(reason="Issue #164157")
|
|
def test_fuzzer_issue_164157(self):
|
|
torch.manual_seed(0)
|
|
|
|
def foo(arg0, arg1, arg2, arg3, arg4, arg5):
|
|
t0 = arg0 # size=(47,), stride=(1,), dtype=int64, device=cuda
|
|
t1 = torch.tanh(t0) # size=(47,), stride=(1,), dtype=int64, device=cuda
|
|
t2 = arg1 # size=(), stride=(), dtype=int64, device=cuda
|
|
t3 = arg2 # size=(), stride=(), dtype=int64, device=cuda
|
|
t4 = t2 * t3 # size=(), stride=(), dtype=int64, device=cuda
|
|
t5 = t1.clone()
|
|
t5.fill_(t4.item())
|
|
t6 = (
|
|
arg3 # size=(256, 88, 1), stride=(88, 1, 1), dtype=float16, device=cuda
|
|
)
|
|
t7 = (
|
|
arg4 # size=(256, 88, 1), stride=(88, 1, 1), dtype=float16, device=cuda
|
|
)
|
|
t8 = (
|
|
arg5 # size=(256, 88, 1), stride=(88, 1, 1), dtype=float16, device=cuda
|
|
)
|
|
t9 = torch.cat([t6, t6, t7, t8], dim=2)
|
|
t10 = t9.std(dim=2)
|
|
t11 = torch.nn.functional.embedding(
|
|
torch.clamp(t5, 0, t10.size(0) - 1), t10
|
|
)
|
|
output = t11
|
|
return output
|
|
|
|
arg0 = torch.randint(0, 100, [47], dtype=torch.int64, device="cuda")
|
|
arg1 = torch.randint(0, 10, [], dtype=torch.int64, device="cuda")
|
|
arg2 = torch.randint(0, 10, [], dtype=torch.int64, device="cuda")
|
|
arg3 = torch.rand(
|
|
[256, 88, 1], dtype=torch.float16, device="cuda", requires_grad=True
|
|
)
|
|
arg4 = torch.rand(
|
|
[256, 88, 1], dtype=torch.float16, device="cuda", requires_grad=True
|
|
)
|
|
arg5 = torch.rand(
|
|
[256, 88, 1], dtype=torch.float16, device="cuda", requires_grad=True
|
|
)
|
|
|
|
out_eager = foo(arg0, arg1, arg2, arg3, arg4, arg5)
|
|
out_eager.sum().backward()
|
|
print("Eager Success! ✅")
|
|
compiled_foo = torch.compile(foo, fullgraph=True, dynamic=True)
|
|
out_compiled = compiled_foo(arg0, arg1, arg2, arg3, arg4, arg5)
|
|
out_compiled.sum().backward()
|
|
print("Compile Success! ✅")
|
|
|
|
@pytest.mark.xfail(reason="Issue #164428")
|
|
def test_fuzzer_issue_164428_already_exists(self):
|
|
torch.manual_seed(6804)
|
|
|
|
def foo(arg0, arg1, arg2):
|
|
var_node_4 = (
|
|
arg0 # size=(7, 1, 32), stride=(1, 1, 0), dtype=float64, device=cuda
|
|
)
|
|
var_node_5 = torch.full((7, 1, 32), -1.195053522845565, dtype=torch.float64)
|
|
var_node_3 = torch.div(var_node_4, var_node_5)
|
|
var_node_2 = torch.flatten(var_node_3)
|
|
var_node_8 = torch.full((2,), -0.8316502130341195, dtype=torch.float64)
|
|
var_node_9 = arg1 # size=(2, 224), stride=(224, 1), dtype=float64
|
|
var_node_7 = torch.matmul(
|
|
var_node_8.to(torch.float64), var_node_9.to(torch.float64)
|
|
)
|
|
var_node_10 = arg2 # size=(224,), stride=(1,), dtype=float64
|
|
var_node_6 = torch.sub(var_node_7, var_node_10)
|
|
var_node_1 = torch.sub(var_node_2, var_node_6)
|
|
output = var_node_1
|
|
return output
|
|
|
|
arg0 = torch.rand(
|
|
[7, 1, 32], dtype=torch.float64, device="cuda", requires_grad=True
|
|
)
|
|
arg1 = torch.rand(
|
|
[2, 224], dtype=torch.float64, device="cuda", requires_grad=True
|
|
)
|
|
arg2 = torch.rand([224], dtype=torch.float64, device="cuda", requires_grad=True)
|
|
|
|
out_eager = foo(arg0, arg1, arg2)
|
|
out_eager.sum().backward()
|
|
print("Eager Success! ✅")
|
|
compiled_foo = torch.compile(foo, fullgraph=True, dynamic=True)
|
|
out_compiled = compiled_foo(arg0, arg1, arg2)
|
|
out_compiled.sum().backward()
|
|
print("Compile Success! ✅")
|
|
|
|
@pytest.mark.xfail(reason="Issue #164086")
|
|
def test_fuzzer_issue_164086(self):
|
|
torch.manual_seed(0)
|
|
|
|
def foo(arg0, arg1, arg2, arg3, arg4, arg5):
|
|
t0 = arg0 # size=(42, 56), stride=(42, 1), dtype=int64, device=cuda
|
|
t1 = torch.tanh(
|
|
t0
|
|
) # size=(42, 56), stride=(42, 1), dtype=int64, device=cuda
|
|
t2 = t1.clone()
|
|
t2.zero_() # size=(42, 56), stride=(42, 1), dtype=int64, device=cuda
|
|
t3 = (
|
|
arg1 # size=(50000, 128), stride=(50000, 1), dtype=float16, device=cuda
|
|
)
|
|
t4 = arg2 # size=(46, 128), stride=(46, 1), dtype=float16, device=cuda
|
|
t5 = torch.nn.functional.linear(
|
|
t3, t4
|
|
) # size=(50000, 46), stride=(50000, 1), dtype=float16, device=cuda
|
|
t6 = arg3 # size=(50000, 4, 46), stride=(184, 46, 1), dtype=float16, device=cuda
|
|
t7 = t6.max(
|
|
dim=1
|
|
).values # size=(50000, 46), stride=(50000, 1), dtype=float16, device=cuda
|
|
t8 = arg4 # size=(25786, 46), stride=(46, 1), dtype=float16, device=cuda
|
|
t9 = arg5 # size=(24214, 46), stride=(46, 1), dtype=float16, device=cuda
|
|
t10 = torch.cat(
|
|
[t8, t9], dim=0
|
|
) # size=(50000, 46), stride=(50000, 1), dtype=float16, device=cuda
|
|
t11 = torch.pow(
|
|
torch.pow(torch.pow(torch.pow(t5, t7), t10), t5), t7
|
|
) # size=(50000, 46), stride=(50000, 1), dtype=float16, device=cuda
|
|
t12 = torch.nn.functional.embedding(
|
|
torch.clamp(t2, 0, t11.size(0) - 1).to(torch.long), t11
|
|
) # size=(42, 56, 46), stride=(2576, 46, 1), dtype=float16, device=cuda
|
|
output = t12
|
|
return output
|
|
|
|
arg0 = torch.randint(0, 1000, [42, 56], dtype=torch.int64, device="cuda")
|
|
arg1 = torch.rand(
|
|
[50000, 128], dtype=torch.float16, device="cuda", requires_grad=True
|
|
)
|
|
arg2 = torch.rand(
|
|
[46, 128], dtype=torch.float16, device="cuda", requires_grad=True
|
|
)
|
|
arg3 = torch.rand(
|
|
[50000, 4, 46], dtype=torch.float16, device="cuda", requires_grad=True
|
|
)
|
|
arg4 = torch.rand(
|
|
[25786, 46], dtype=torch.float16, device="cuda", requires_grad=True
|
|
)
|
|
arg5 = torch.rand(
|
|
[24214, 46], dtype=torch.float16, device="cuda", requires_grad=True
|
|
)
|
|
|
|
out_eager = foo(arg0, arg1, arg2, arg3, arg4, arg5)
|
|
out_eager.sum().backward()
|
|
print("Eager Success! ✅")
|
|
compiled_foo = torch.compile(foo, fullgraph=True, dynamic=True)
|
|
out_compiled = compiled_foo(arg0, arg1, arg2, arg3, arg4, arg5)
|
|
out_compiled.sum().backward()
|
|
print("Compile Success! ✅")
|
|
|
|
@pytest.mark.xfail(reason="Issue #163877")
|
|
def test_fuzzer_issue_163877(self):
|
|
torch.manual_seed(0)
|
|
|
|
def foo(arg0, arg1):
|
|
t0 = arg0 # size=(401120, 3), stride=(3, 1), dtype=float32, device=cuda
|
|
t1 = t0.clone()
|
|
t1.zero_() # size=(401120, 3), stride=(3, 1), dtype=float32, device=cuda
|
|
t2 = t1.reshape(
|
|
(109, 115, 96)
|
|
) # size=(109, 115, 96), stride=(11040, 96, 1), dtype=float32, device=cuda
|
|
t3 = arg1 # size=(), stride=(), dtype=float32, device=cuda
|
|
t4 = t3.contiguous() # size=(), stride=(), dtype=float32, device=cuda
|
|
t5 = torch.nn.functional.relu(
|
|
t4
|
|
) # size=(), stride=(), dtype=float32, device=cuda
|
|
t6 = t2.clone()
|
|
t6.fill_(
|
|
t5.item()
|
|
) # size=(109, 115, 96), stride=(11040, 96, 1), dtype=float32, device=cuda
|
|
output = t6
|
|
return output
|
|
|
|
arg0 = torch.rand(
|
|
[401120, 3], dtype=torch.float32, device="cuda", requires_grad=True
|
|
)
|
|
arg1 = torch.rand([], dtype=torch.float32, device="cuda", requires_grad=True)
|
|
|
|
out_eager = foo(arg0, arg1)
|
|
out_eager.sum().backward()
|
|
print("Eager Success! ✅")
|
|
compiled_foo = torch.compile(foo, fullgraph=True, dynamic=True)
|
|
out_compiled = compiled_foo(arg0, arg1)
|
|
out_compiled.sum().backward()
|
|
print("Compile Success! ✅")
|
|
|
|
@pytest.mark.xfail(reason="Issue #163971")
|
|
def test_fuzzer_issue_163971(self):
|
|
torch.manual_seed(0)
|
|
|
|
def foo(arg0):
|
|
t0 = arg0 # size=(), stride=(), dtype=bfloat16, device=cuda
|
|
t1 = torch.softmax(
|
|
t0, dim=0
|
|
) # size=(), stride=(), dtype=bfloat16, device=cuda
|
|
t2 = torch.nn.functional.gelu(
|
|
t1
|
|
) # size=(), stride=(), dtype=bfloat16, device=cuda
|
|
t3 = torch.softmax(
|
|
t2, dim=0
|
|
) # size=(), stride=(), dtype=bfloat16, device=cuda
|
|
output = t3
|
|
return output
|
|
|
|
arg0 = torch.rand([], dtype=torch.bfloat16, device="cuda", requires_grad=True)
|
|
|
|
out_eager = foo(arg0)
|
|
out_eager.sum().backward()
|
|
print("Eager Success! ✅")
|
|
compiled_foo = torch.compile(foo, fullgraph=True, dynamic=True)
|
|
out_compiled = compiled_foo(arg0)
|
|
out_compiled.sum().backward()
|
|
print("Compile Success! ✅")
|
|
|
|
@pytest.mark.xfail(reason="Issue #164059")
|
|
def test_fuzzer_issue_164059(self):
|
|
torch.manual_seed(0)
|
|
|
|
def foo(arg0, arg1, arg2):
|
|
t0 = arg0 # size=(16, 38073, 1), stride=(38073, 1, 1), dtype=float32, device=cuda
|
|
t1 = t0.clone()
|
|
t1.zero_() # size=(16, 38073, 1), stride=(38073, 1, 1), dtype=float32, device=cuda
|
|
t2 = t1.contiguous().view(
|
|
(49, 112, 111)
|
|
) # size=(49, 112, 111), stride=(5488, 112, 1), dtype=float32, device=cuda
|
|
t3 = arg1 # size=(1,), stride=(1,), dtype=int64, device=cuda
|
|
t4 = arg2 # size=(1,), stride=(1,), dtype=int64, device=cuda
|
|
t5 = t3 + t3 + t4 # size=(1,), stride=(1,), dtype=int64, device=cuda
|
|
t6 = torch.exp(t5) # size=(1,), stride=(1,), dtype=int64, device=cuda
|
|
t7 = torch.nn.functional.layer_norm(
|
|
t2, (111,)
|
|
) # size=(49, 112, 111), stride=(12432, 111, 1), dtype=float32, device=cuda
|
|
output = t7
|
|
return output
|
|
|
|
arg0 = torch.rand(
|
|
[16, 38073, 1], dtype=torch.float32, device="cuda", requires_grad=True
|
|
)
|
|
arg1 = torch.randint(0, 1000, [1], dtype=torch.int64, device="cuda")
|
|
arg2 = torch.randint(0, 1000, [1], dtype=torch.int64, device="cuda")
|
|
|
|
out_eager = foo(arg0, arg1, arg2)
|
|
out_eager.sum().backward()
|
|
print("Eager Success! ✅")
|
|
compiled_foo = torch.compile(foo, fullgraph=True, dynamic=True)
|
|
out_compiled = compiled_foo(arg0, arg1, arg2)
|
|
out_compiled.sum().backward()
|
|
print("Compile Success! ✅")
|
|
|
|
@pytest.mark.xfail(reason="Issue #164088")
|
|
def test_fuzzer_issue_164088(self):
|
|
torch.manual_seed(0)
|
|
|
|
def foo(arg0, arg1, arg2, arg3, arg4):
|
|
t0 = arg0 # size=(23, 4), stride=(4, 1), dtype=bfloat16, device=cuda
|
|
t1 = t0.clone()
|
|
t1.zero_() # size=(23, 4), stride=(4, 1), dtype=bfloat16, device=cuda
|
|
t2 = t1.contiguous().view(
|
|
(92,)
|
|
) # size=(92,), stride=(1,), dtype=bfloat16, device=cuda
|
|
t3 = arg1 # size=(5, 4, 5), stride=(20, 5, 1), dtype=bfloat16, device=cuda
|
|
t4 = t3.min() # size=(), stride=(), dtype=bfloat16, device=cuda
|
|
t5 = arg2 # size=(), stride=(), dtype=bfloat16, device=cuda
|
|
t6 = torch.nn.functional.silu(
|
|
t5
|
|
) # size=(), stride=(), dtype=bfloat16, device=cuda
|
|
t7 = arg3 # size=(3, 2, 3), stride=(6, 3, 1), dtype=bfloat16, device=cuda
|
|
t8 = t7.min() # size=(), stride=(), dtype=bfloat16, device=cuda
|
|
t9 = arg4 # size=(), stride=(), dtype=bfloat16, device=cuda
|
|
t10 = ((t8) / t9) / t9 # size=(), stride=(), dtype=bfloat16, device=cuda
|
|
t11 = (
|
|
t4 + t4 + t6 + t10 + t8
|
|
) # size=(), stride=(), dtype=bfloat16, device=cuda
|
|
t12 = t2.clone()
|
|
t12.fill_(
|
|
t11.item()
|
|
) # size=(92,), stride=(1,), dtype=bfloat16, device=cuda
|
|
output = t12
|
|
return output
|
|
|
|
arg0 = torch.rand(
|
|
[23, 4], dtype=torch.bfloat16, device="cuda", requires_grad=True
|
|
)
|
|
arg1 = torch.rand(
|
|
[5, 4, 5], dtype=torch.bfloat16, device="cuda", requires_grad=True
|
|
)
|
|
arg2 = torch.rand([], dtype=torch.bfloat16, device="cuda", requires_grad=True)
|
|
arg3 = torch.rand(
|
|
[3, 2, 3], dtype=torch.bfloat16, device="cuda", requires_grad=True
|
|
)
|
|
arg4 = torch.rand([], dtype=torch.bfloat16, device="cuda", requires_grad=True)
|
|
|
|
out_eager = foo(arg0, arg1, arg2, arg3, arg4)
|
|
out_eager.sum().backward()
|
|
print("Eager Success! ✅")
|
|
compiled_foo = torch.compile(foo, fullgraph=True, dynamic=True)
|
|
out_compiled = compiled_foo(arg0, arg1, arg2, arg3, arg4)
|
|
out_compiled.sum().backward()
|
|
print("Compile Success! ✅")
|
|
|
|
@pytest.mark.xfail(reason="Issue #163894")
|
|
def test_fuzzer_issue_163894(self):
|
|
torch.manual_seed(9)
|
|
|
|
def foo(arg0):
|
|
var_node_1 = arg0 # size=(1, 2), stride=(2, 1), dtype=int64, device=cuda
|
|
var_node_5 = torch.full(
|
|
(1, 2), -66, dtype=torch.int32
|
|
) # size=(1, 2), stride=(2, 1), dtype=int32, device=cuda
|
|
var_node_6 = torch.full(
|
|
(1, 2), 77, dtype=torch.int64
|
|
) # size=(1, 2), stride=(2, 1), dtype=int64, device=cuda
|
|
var_node_4 = torch.ops.aten.add(
|
|
var_node_5, var_node_6
|
|
) # size=(1, 2), stride=(2, 1), dtype=int32, device=cuda
|
|
var_node_7 = torch.full(
|
|
(1, 2), -64, dtype=torch.int32
|
|
) # size=(1, 2), stride=(2, 1), dtype=int32, device=cuda
|
|
var_node_3 = torch.ops.aten.mul(
|
|
var_node_4, var_node_7
|
|
) # size=(1, 2), stride=(2, 1), dtype=int32, device=cuda
|
|
var_node_9 = torch.full(
|
|
(3, 4), False, dtype=torch.bool
|
|
) # size=(3, 4), stride=(4, 1), dtype=bool, device=cuda
|
|
var_node_8 = torch.nonzero(
|
|
var_node_9
|
|
) # size=(0, 2), stride=(2, 1), dtype=int64, device=cuda
|
|
if var_node_8.numel() == 0:
|
|
var_node_8 = torch.zeros((1, 2), dtype=torch.int64, device="cuda")
|
|
var_node_2 = torch.ops.aten.add(var_node_3, var_node_8)
|
|
output = var_node_2.float()
|
|
return output
|
|
|
|
arg0 = torch.randint(0, 10, [1, 2], dtype=torch.int64, device="cuda")
|
|
|
|
out_eager = foo(arg0)
|
|
out_eager.sum().backward()
|
|
print("Eager Success! ✅")
|
|
compiled_foo = torch.compile(foo, fullgraph=True, dynamic=True)
|
|
out_compiled = compiled_foo(arg0)
|
|
out_compiled.sum().backward()
|
|
print("Compile Success! ✅")
|
|
|
|
@pytest.mark.xfail(reason="Issue #164486")
|
|
def test_fuzzer_issue_164486(self):
|
|
torch.manual_seed(238)
|
|
|
|
def foo(arg0):
|
|
var_node_2 = torch.full(
|
|
(), 1, dtype=torch.int16
|
|
) # size=(), stride=(), dtype=int16, device=cuda
|
|
var_node_3 = arg0 # size=(), stride=(), dtype=int16, device=cuda
|
|
var_node_1 = torch.add(
|
|
var_node_2, var_node_3
|
|
) # size=(), stride=(), dtype=int16, device=cuda
|
|
var_node_5 = torch.full(
|
|
(1,), 3, dtype=torch.int16
|
|
) # size=(1,), stride=(1,), dtype=int16, device=cuda
|
|
var_node_4 = torch.squeeze(
|
|
var_node_5
|
|
) # size=(), stride=(), dtype=int16, device=cuda
|
|
var_node_0 = torch.div(
|
|
var_node_1, var_node_4
|
|
) # size=(), stride=(), dtype=int16, device=cuda
|
|
result = var_node_0.float()
|
|
return result
|
|
|
|
arg0 = torch.randint(0, 10, [], dtype=torch.int16, device="cuda")
|
|
|
|
out_eager = foo(arg0)
|
|
out_eager.sum().backward()
|
|
print("Eager Success! ✅")
|
|
compiled_foo = torch.compile(foo, fullgraph=True, dynamic=True)
|
|
out_compiled = compiled_foo(arg0)
|
|
out_compiled.sum().backward()
|
|
print("Compile Success! ✅")
|
|
|
|
@pytest.mark.xfail(reason="Issue #163674")
|
|
def test_fuzzer_issue_163674(self):
|
|
torch.manual_seed(0)
|
|
|
|
def foo(arg0, arg1, arg2):
|
|
t0 = arg0 # size=(79488, 1, 3, 1), stride=(3, 3, 1, 1), dtype=float16, device=cuda
|
|
t1 = t0.clone()
|
|
t1.zero_() # size=(79488, 1, 3, 1), stride=(3, 3, 1, 1), dtype=float16, device=cuda
|
|
t2 = arg1 # size=(79488, 1, 3, 1), stride=(3, 3, 1, 1), dtype=float32, device=cuda
|
|
t3 = arg2 # size=(), stride=(), dtype=float32, device=cuda
|
|
t4 = t2.clone()
|
|
t4.fill_(
|
|
t3.item()
|
|
) # size=(79488, 1, 3, 1), stride=(3, 3, 1, 1), dtype=float32, device=cuda
|
|
t5 = torch.pow(
|
|
t1, t4
|
|
) # size=(79488, 1, 3, 1), stride=(3, 3, 1, 1), dtype=float32, device=cuda
|
|
t6 = t5.reshape(
|
|
(96, 69, 36)
|
|
) # size=(96, 69, 36), stride=(2484, 36, 1), dtype=float32, device=cuda
|
|
output = t6
|
|
return output
|
|
|
|
arg0 = torch.rand(
|
|
[79488, 1, 3, 1], dtype=torch.float16, device="cuda", requires_grad=True
|
|
)
|
|
arg1 = torch.rand(
|
|
[79488, 1, 3, 1], dtype=torch.float32, device="cuda", requires_grad=True
|
|
)
|
|
arg2 = torch.rand([], dtype=torch.float32, device="cuda", requires_grad=True)
|
|
|
|
out_eager = foo(arg0, arg1, arg2)
|
|
out_eager.sum().backward()
|
|
print("Eager Success! ✅")
|
|
compiled_foo = torch.compile(foo, fullgraph=True, dynamic=True)
|
|
out_compiled = compiled_foo(arg0, arg1, arg2)
|
|
out_compiled.sum().backward()
|
|
print("Compile Success! ✅")
|
|
|
|
|
|
if __name__ == "__main__":
|
|
run_tests()
|