Add allocator check for available free memory on device.

c10/xpu/XPUCachingAllocator.cpp

@@ -900,6 +900,16 @@ class DeviceCachingAllocator {
     }
     bool active_pool =
         p.pool->owner_PrivatePool && p.pool->owner_PrivatePool->allocator();
+
+    const auto& raw_device = c10::xpu::get_raw_device(p.device());
+    if (!active_pool && raw_device.has(sycl::aspect::ext_intel_free_memory)) {
+      const size_t device_free =
+          raw_device.get_info<sycl::ext::intel::info::device::free_memory>();
+      if (size > device_free) {
+        return false;
+      }
+    }
+
     if (set_fraction &&
         stats.reserved_bytes[static_cast<size_t>(StatType::AGGREGATE)].current +
                 size >

test/test_xpu.py

@@ -566,6 +566,26 @@ def test_raises_oom(self):
         with self.assertRaises(torch.OutOfMemoryError):
             torch.empty(1024 * 1024 * 1024 * 1024, device="xpu")
 
+    @serialTest()
+    @unittest.skipIf(
+        int(torch.version.xpu) < 20250000,
+        "Test requires SYCL compiler version 2025.0.0 or newer.",
+    )
+    def test_allocation_raises_oom(self):
+        gc.collect()
+        torch.xpu.empty_cache()
+        free_bytes, _ = torch.xpu.mem_get_info()
+        alloc_bytes = free_bytes * 5 // 9
+        tensor = None
+        try:
+            tensor = torch.empty(alloc_bytes, dtype=torch.bool, device="xpu")
+            with self.assertRaises(torch.OutOfMemoryError):
+                tensor.clone()
+        finally:
+            del tensor
+            gc.collect()
+            torch.xpu.empty_cache()
+
     @serialTest()
     def test_1mb_allocation_uses_small_block(self):
         gc.collect()