Enable PE + TE (#45546) (#45591)

Krovatkin · web-flow · commit 07e66d7ca5c7 · 2020-09-30T15:56:05.000-07:00
Summary: This PR enables PE + TE for 1.7 Pull Request resolved: #45546 Reviewed By: ZolotukhinM Differential Revision: D24006940 Pulled By: Krovatkin fbshipit-source-id: a3326077d34a023941acdb06c4907c96e7ba0115
diff --git a/test/test_jit_cuda_fuser.py b/test/test_jit_cuda_fuser.py
@@ -3,7 +3,7 @@
 
 import torch
 
-from torch.testing._internal.common_utils import run_tests, ProfilingMode, GRAPH_EXECUTOR, skipIfRocm
+from torch.testing._internal.common_utils import run_tests, ProfilingMode, GRAPH_EXECUTOR, skipIfRocm, TEST_WITH_ROCM
 from torch.testing._internal.codegen.random_topo_test import runDefaultTestWithSeed
 
 from test_jit import JitTestCase, RUN_CUDA
@@ -804,4 +804,5 @@ def test_register_fuser(self):
 
 
 if __name__ == '__main__':
-    run_tests()
+    if not TEST_WITH_ROCM:
+        run_tests()
diff --git a/torch/csrc/jit/passes/tensorexpr_fuser.cpp b/torch/csrc/jit/passes/tensorexpr_fuser.cpp
@@ -192,7 +192,7 @@ bool isSupported(Node* node) {
 
 } // namespace tensorexpr
 
-static bool texpr_fuser_enabled_ = false;
+static bool texpr_fuser_enabled_ = true;
 
 void setTensorExprFuserEnabled(bool val) {
   texpr_fuser_enabled_ = val;
diff --git a/torch/csrc/jit/runtime/profiling_graph_executor_impl.cpp b/torch/csrc/jit/runtime/profiling_graph_executor_impl.cpp
@@ -45,7 +45,7 @@ static std::atomic<bool> executor_mode{true};
 static std::atomic<bool> profiling_mode{false};
 #else
 static std::atomic<bool> executor_mode{true};
-static std::atomic<bool> profiling_mode{false};
+static std::atomic<bool> profiling_mode{true};
 #endif
 
 static std::atomic<size_t> num_profiled_runs{1};
