pytorch · bddppq · Jun 7, 2019 · Jun 7, 2019 · bddppq · Jun 7, 2019
diff --git a/aten/src/ATen/native/mkldnn/SoftMax.cpp b/aten/src/ATen/native/mkldnn/SoftMax.cpp
@@ -0,0 +1,79 @@
+#include <ATen/ATen.h>
+#include <ATen/Config.h>
+#include <ATen/NativeFunctions.h>
+
+#if !AT_MKLDNN_ENABLED()
+
+namespace at {
+namespace native {
+
+Tensor mkldnn_softmax(
+    const Tensor& self,
+    const int64_t dim,
+    const bool half_to_float) {
+  AT_ERROR("mkldnn_softmax: ATen not compiled with MKLDNN support");
+}
+
+} // namespace native
+} // namespace at
+
+#else // AT_MKLDNN_EBABLED
+
+#include <ATen/native/mkldnn/MKLDNNCommon.h>
+
+namespace at {
+namespace native {
+
+namespace {
+// TODO: move this to ideep
+struct ideep_softmax_forward
+    : public ideep::softmax_forward,
+      public ideep::utils::computation_cache<ideep_softmax_forward> {
+  template <typename... Ts>
+  ideep_softmax_forward(
+      const ideep::tensor::descriptor& src_desc,
+      const ideep::tensor::descriptor& dst_desc,
+      Ts&&... args) {
+    init(src_desc, dst_desc, std::forward<Ts>(args)...);
+  }
+
+  template <class alloc>
+  static void compute(
+      const ideep::tensor& src,
+      ideep::tensor& dst,
+      int softmax_axis) {
+    if (dst.get_descriptor() != src.get_descriptor()) {
+      dst.reinit<alloc, ideep_softmax_forward>(src.get_descriptor());
+    }
+    ideep::key_t key;
+    ideep::utils::create_key(
+        key,
+        src.get_data_type(),
+        src.get_dims(),
+        src.get_internal_format(),
+        softmax_axis);
+    fetch_or_create_m(
+        comp, key, src.get_descriptor(), dst.get_descriptor(), softmax_axis);
+    comp.execute(src, dst);
+  }
+};
+} // namespace
+
+Tensor mkldnn_softmax(
+    const Tensor& self,
+    const int64_t dim,
+    const bool half_to_float) {
+  AT_ASSERTM(
+      !half_to_float,
+      "softmax with half to float conversion is not supported on Mkldnn");
+  const int64_t wrapped_dim = maybe_wrap_dim(dim, self.dim());
+  ideep::tensor& x = itensor_from_mkldnn(self);
+  ideep::tensor y;
+  ideep_softmax_forward::compute<AllocForMKLDNN>(x, y, wrapped_dim);
+  return new_with_itensor_mkldnn(std::move(y), self.options());
+}
+
+} // namespace native
+} // namespace at
+
+#endif // AT_MKLDNN_EBABLED
diff --git a/aten/src/ATen/native/native_functions.yaml b/aten/src/ATen/native/native_functions.yaml
@@ -1741,6 +1741,7 @@
   dispatch:
     CPU: softmax_cpu
     CUDA: softmax_cuda
+    MkldnnCPU: mkldnn_softmax
 
 - func: _softmax_backward_data(Tensor grad_output, Tensor output, int dim, Tensor self) -> Tensor
   dispatch:

diff --git a/test/test_mkldnn.py b/test/test_mkldnn.py
@@ -260,6 +260,14 @@ def test_linear(self):
             self._test_serialization(mkldnn_linear, (x.to_mkldnn(),))
             self._test_tracing(mkldnn_linear, (x.to_mkldnn(),))
 
+    def test_softmax(self):
+        x = torch.randn(3, 4, 5, dtype=torch.float32) * 10
+        for dim in range(x.ndim):
+            softmax = torch.nn.Softmax(dim=dim)
+            self.assertEqual(
+                softmax(x),
+                softmax(x.to_mkldnn()).to_dense())
+
     def test_sigmoid(self):
         x = torch.randn(4, 5, dtype=torch.float32) * 10
         mkldnn_x = x.to_mkldnn()