[inductor] adapte windows file path (pytorch#130713)

This PR is depends on pytorch#130132 can be landed successful.
The detailed log: pytorch#124245 (comment)

After the file path was adapted for Windows, the first Windows inductor case was run successful.

```python
import torch

def foo(x, y):
    a = torch.sin(x)
    b = torch.cos(x)
    return a + b
opt_foo1 = torch.compile(foo)
print(opt_foo1(torch.randn(10, 10), torch.randn(10, 10)))
```

Result:
![image](https://github.com/user-attachments/assets/4944df47-e74d-476b-8eb5-1d1fd5abeb41)

Co-authored-by: Jiong Gong <jiong.gong@intel.com>
Pull Request resolved: pytorch#130713
Approved by: https://github.com/jgong5, https://github.com/jansel, https://github.com/desertfire

Author: xuhancn

test/functorch/test_eager_transforms.py

@@ -5085,9 +5085,8 @@ def wrapper(*args, **kwargs):
 @markDynamoStrictTest
 class TestCompileTransforms(TestCase):
     @skipIfRocm(msg="test leaks memory on ROCm")
-    # torch.compile is not supported on Windows
     # Triton only supports GPU with SM70 or later.
-    @expectedFailureIf(IS_WINDOWS or (TEST_CUDA and not SM70OrLater))
+    @expectedFailureIf(TEST_CUDA and not SM70OrLater)
     def test_compile_vmap_hessian(self, device):
         # The model and inputs are a smaller version
         # of code at benchmark repo:

torch/_inductor/codecache.py

@@ -57,27 +57,26 @@
     rocm_compile_command,
     rocm_compiler,
 )
-from .cpp_builder import (
-    _get_python_include_dirs,
-    get_cpp_compiler,
-    homebrew_libomp,
-    is_apple_clang,
-    is_clang,
-    is_conda_llvm_openmp_installed,
-)
 
 """
 codecache.py, cpp_builder.py and cpu_vec_isa.py import rule:
 https://github.com/pytorch/pytorch/issues/124245#issuecomment-2197778902
 """
 from torch._inductor.cpp_builder import (
+    _get_python_include_dirs,
     _set_gpu_runtime_env,
     _transform_cuda_paths,
     CppBuilder,
     CppOptions,
     CppTorchCudaOptions,
     get_compiler_version_info,
+    get_cpp_compiler,
     get_name_and_dir_from_output_file_path,
+    homebrew_libomp,
+    is_apple_clang,
+    is_clang,
+    is_conda_llvm_openmp_installed,
+    normalize_path_separator,
 )
 from torch._inductor.cpu_vec_isa import invalid_vec_isa, pick_vec_isa, VecISA
 from torch._inductor.runtime.compile_tasks import (
@@ -1922,7 +1921,7 @@ def cpp_prefix_path() -> str:
             content,
             "h",
         )
-    return filename
+    return normalize_path_separator(filename)
 
 
 def cpp_prefix() -> str:
@@ -2109,7 +2108,7 @@ def load_async(cls, source_code: str, cuda=False, submit_fn=None, extra_flags=()
                 fb_output_path,
             )
 
-            binary_path = (
+            binary_path = normalize_path_separator(
                 fb_output_path
                 if config.is_fbcode()
                 else cpp_builder.get_target_file_path()

torch/_inductor/cpp_builder.py

@@ -244,6 +244,12 @@ def run_command_line(cmd_line, cwd=None):
     return status
 
 
+def normalize_path_separator(orig_path: str) -> str:
+    if _IS_WINDOWS:
+        return orig_path.replace(os.sep, "/")
+    return orig_path
+
+
 class BuildOptionsBase:
     """
     This is the Base class for store cxx build options, as a template.
@@ -1219,7 +1225,7 @@ def format_build_command(
                     f"{compiler} {include_dirs_args} {definations_args} {cflags_args} {sources} "
                     f"{passthougn_args} /LD /Fe{target_file} /link {libraries_dirs_args} {libraries_args} {ldflags_args} "
                 )
-                cmd = cmd.replace("\\", "/")
+                cmd = normalize_path_separator(cmd)
             else:
                 compile_only_arg = "-c" if self._compile_only else ""
                 cmd = re.sub(
@@ -1247,7 +1253,7 @@ def format_build_command(
         return command_line
 
     def get_target_file_path(self):
-        return self._target_file
+        return normalize_path_separator(self._target_file)
 
     def build(self) -> Tuple[int, str]:
         """

torch/_inductor/cpu_vec_isa.py

@@ -89,7 +89,11 @@ def __hash__(self) -> int:
 
     def check_build(self, code: str) -> bool:
         from torch._inductor.codecache import get_lock_dir, LOCK_TIMEOUT, write
-        from torch._inductor.cpp_builder import CppBuilder, CppTorchOptions
+        from torch._inductor.cpp_builder import (
+            CppBuilder,
+            CppTorchOptions,
+            normalize_path_separator,
+        )
 
         key, input_path = write(
             code,
@@ -111,7 +115,9 @@ def check_build(self, code: str) -> bool:
             )
             try:
                 # Check if the output file exist, and compile when not.
-                output_path = x86_isa_help_builder.get_target_file_path()
+                output_path = normalize_path_separator(
+                    x86_isa_help_builder.get_target_file_path()
+                )
                 if not os.path.isfile(output_path):
                     status, target_file = x86_isa_help_builder.build()