CPU: Implement scalar broadcast

src/ngraph/runtime/cpu/emitter.cpp

@@ -20,6 +20,7 @@
 
 #include "ngraph/descriptor/layout/dense_tensor_view_layout.hpp"
 #include "ngraph/node.hpp"
+#include "ngraph/ops/broadcast.hpp"
 #include "ngraph/ops/concatenate.hpp"
 #include "ngraph/ops/constant.hpp"
 #include "ngraph/ops/get_tuple_element.hpp"
@@ -771,3 +772,47 @@ void Emitter::EMITTER_DECL(EmitParameterizedConstantUInt64)
 
     TU += "};\n    }\n";
 }
+
+void Emitter::EMITTER_DECL(EmitBroadcast)
+{
+    auto broadcast = static_cast<const op::Broadcast*>(n);
+
+    auto arg_tensor_type =
+        dynamic_pointer_cast<const TensorViewType>(n->get_arguments().at(0)->get_value_type());
+    assert(arg_tensor_type);
+
+    auto result_tensor_type = dynamic_pointer_cast<const TensorViewType>(n->get_value_type());
+    assert(result_tensor_type);
+
+    auto arg_shape = arg_tensor_type->get_shape();
+    auto result_shape = result_tensor_type->get_shape();
+    auto& result_element_type = result_tensor_type->get_element_type();
+
+    if (broadcast->get_broadcast_axes().empty())
+    {
+        TU +=
+            "    {\n"
+            "        call_frame->get_parameterized_tensor_view<" +
+            element_type_names[TI(result_element_type)] + ">(" + to_string(outputs[0].get_index()) +
+            ")->get_vector() =\n"
+            "        call_frame->get_parameterized_tensor_view<" +
+            element_type_names[TI(result_element_type)] + ">(" + to_string(inputs[0].get_index()) +
+            ")->get_vector();\n"
+            "    }\n";
+    }
+    else if (arg_shape.size() == 0)
+    {
+        TU += "    {\n"
+              "        auto arg0 = call_frame->get_tensor_view_data<" + element_type_names[TI(result_element_type)] + ">(" + to_string(inputs[0].get_index()) + ");\n"
+              "        auto out  = call_frame->get_tensor_view_data<" + element_type_names[TI(result_element_type)] + ">(" + to_string(outputs[0].get_index()) + ");\n"
+              "        EigenArray1d<" + element_type_names[TI(result_element_type)] + ">(out, "
+                       EIGEN_VECTOR_FORMAT(outputs[0].get_layout<DenseTensorViewLayout>()->get_size()) ") =\n"
+              "        EigenArray1d<" + element_type_names[TI(result_element_type)] + ">(arg0, "
+                       EIGEN_VECTOR_FORMAT(inputs[0].get_layout<DenseTensorViewLayout>()->get_size()) ")(0, 0);\n"
+              "    }\n";
+    }
+    else
+    {
+        throw ngraph_error("Broadcast not implemented for given inputs");
+    }
+}

src/ngraph/runtime/cpu/emitter.hpp

@@ -73,6 +73,7 @@ namespace ngraph
                 void EMITTER_DECL(EmitParameterizedConstantUInt8);
                 void EMITTER_DECL(EmitParameterizedConstantUInt32);
                 void EMITTER_DECL(EmitParameterizedConstantUInt64);
+                void EMITTER_DECL(EmitBroadcast);
             };
         }
     }

src/ngraph/runtime/cpu/external_function.cpp

@@ -62,6 +62,7 @@
 // TODO: Decide if we want to ship this or
 // just enable it for developer build-test cycles
 //#define NGCPU_PCH
+//#define NGCPU_DEBUGINFO
 
 using namespace std;
 using namespace ngraph::runtime::cpu;
@@ -107,7 +108,7 @@ static const OpMap dispatcher{
      &Emitter::EmitParameterizedConstantUInt32},
     {TI(ngraph::op::ParameterizedConstant<ngraph::element::UInt64>),
      &Emitter::EmitParameterizedConstantUInt64},
-
+    {TI(ngraph::op::Broadcast), &Emitter::EmitBroadcast},
 };
 
 #undef TI
@@ -252,6 +253,10 @@ extern "C" void __entrypoint(ngraph::runtime::cpu::CallFrame* call_frame,
     estate.enable_pch();
 #endif
 
+#if defined(NGCPU_DEBUGINFO)
+    estate.enable_debuginfo();
+#endif
+
     auto llvm_module = estate.compile(TU, "__ngcpu_codegen.cpp");
     assert(llvm_module);
     estate.add_module(llvm_module);

test/cpu.cpp

@@ -1128,8 +1128,9 @@ TEST(execute, function_call)
     (*cf)({x, z, y}, {result});
     ASSERT_EQ((vector<float>{100, 144, 196, 256}), result->get_vector());
 }
+*/
 
-TEST(execute, broadcast_scalar_vector)
+TEST(cpu, broadcast_scalar_vector)
 {
     auto shape_a = Shape{};
     auto A = make_shared<op::Parameter>(element::Float32::element_type(), shape_a);
@@ -1138,7 +1139,7 @@ TEST(execute, broadcast_scalar_vector)
     auto f = make_shared<Function>(
         make_shared<op::Broadcast>(A, shape_r, AxisSet{0}), rt, op::Parameters{A});
 
-    auto manager = runtime::Manager::get("NGVM");
+    auto manager = runtime::Manager::get("CPU");
     auto external = manager->compile(f);
     auto backend = manager->allocate_backend();
     auto cf = backend->make_call_frame(external);
@@ -1152,7 +1153,7 @@ TEST(execute, broadcast_scalar_vector)
     ASSERT_EQ((vector<float>{6, 6, 6, 6}), result->get_vector());
 }
 
-TEST(execute, broadcast_scalar_matrix)
+TEST(cpu, broadcast_scalar_matrix)
 {
     auto shape_a = Shape{};
     auto A = make_shared<op::Parameter>(element::Float32::element_type(), shape_a);
@@ -1161,7 +1162,7 @@ TEST(execute, broadcast_scalar_matrix)
     auto f = make_shared<Function>(
         make_shared<op::Broadcast>(A, shape_r, AxisSet{0, 1}), rt, op::Parameters{A});
 
-    auto manager = runtime::Manager::get("NGVM");
+    auto manager = runtime::Manager::get("CPU");
     auto external = manager->compile(f);
     auto backend = manager->allocate_backend();
     auto cf = backend->make_call_frame(external);
@@ -1175,7 +1176,7 @@ TEST(execute, broadcast_scalar_matrix)
     ASSERT_EQ((vector<float>{6, 6, 6, 6}), result->get_vector());
 }
 
-TEST(execute, broadcast_scalar_tensor)
+TEST(cpu, broadcast_scalar_tensor)
 {
     auto shape_a = Shape{};
     auto A = make_shared<op::Parameter>(element::Float32::element_type(), shape_a);
@@ -1184,7 +1185,7 @@ TEST(execute, broadcast_scalar_tensor)
     auto f = make_shared<Function>(
         make_shared<op::Broadcast>(A, shape_r, AxisSet{0, 1, 2}), rt, op::Parameters{A});
 
-    auto manager = runtime::Manager::get("NGVM");
+    auto manager = runtime::Manager::get("CPU");
     auto external = manager->compile(f);
     auto backend = manager->allocate_backend();
     auto cf = backend->make_call_frame(external);
@@ -1198,7 +1199,7 @@ TEST(execute, broadcast_scalar_tensor)
     ASSERT_EQ((vector<float>{6, 6, 6, 6, 6, 6, 6, 6}), result->get_vector());
 }
 
-TEST(execute, broadcast_trivial)
+TEST(cpu, broadcast_trivial)
 {
     auto shape = Shape{2, 2, 2};
     auto A = make_shared<op::Parameter>(element::Float32::element_type(), shape);
@@ -1206,7 +1207,7 @@ TEST(execute, broadcast_trivial)
     auto f = make_shared<Function>(
         make_shared<op::Broadcast>(A, shape, AxisSet{}), rt, op::Parameters{A});
 
-    auto manager = runtime::Manager::get("NGVM");
+    auto manager = runtime::Manager::get("CPU");
     auto external = manager->compile(f);
     auto backend = manager->allocate_backend();
     auto cf = backend->make_call_frame(external);
@@ -1220,6 +1221,7 @@ TEST(execute, broadcast_trivial)
     ASSERT_EQ((vector<float>{2, 4, 6, 8, 16, 32, 64, 128}), result->get_vector());
 }
 
+/*
 TEST(execute, broadcast_vector_colwise)
 {
     auto shape_a = Shape{3};