empty caching allocator before test_avg_pool2d large subtest (pytorch#63528)

Summary:
Otherwise, unrecoverable OOM occurs on MI25.  Fixes broken ROCm CI test1.

Pull Request resolved: pytorch#63528

Reviewed By: malfet, zhouzhuojie

Differential Revision: D30459151

Pulled By: walterddr

fbshipit-source-id: 63e205c4f486fcbdd514cfb0ed8e38584f894585

Author: jeffdaily

test/test_nn.py

@@ -13936,6 +13936,9 @@ def helper(n, c, h, w, kernel_size, stride=None,
         helper(4, 8, 8, 8, 3, count_include_pad=False, padding=2, stride=2)
         helper(4, 8, 8, 8, 3, divisor_override=42)
         helper(4, 8, 8, 8, 7)
+        # ROCm 16GB MI25 hits OOM error. Clear caching allocator prior to running large subtest.
+        if TEST_WITH_ROCM and 'cuda' in device:
+            torch.cuda.empty_cache()
         helper(200, 512, 28, 28, 2)
         helper(4, 8, 7, 7, 3, stride=1)
         helper(4, 8, 7, 7, 3, padding=2, stride=1)