SoftSign ONNX parser (#1046)

Add onnx parser and unit tests for Softsign

src/onnx/parse_softsign.cpp

+#include <migraphx/onnx/op_parser.hpp>
+#include <migraphx/onnx/checks.hpp>
+#include <migraphx/instruction.hpp>
+#include <migraphx/make_op.hpp>
+
+namespace migraphx {
+inline namespace MIGRAPHX_INLINE_NS {
+namespace onnx {
+
+struct parse_softsign : op_parser<parse_softsign>
+{
+    std::vector<op_desc> operators() const { return {{"Softsign"}}; }
+
+    instruction_ref parse(const op_desc& /*opd*/,
+                          const onnx_parser& /*parser*/,
+                          const onnx_parser::node_info& info,
+                          std::vector<instruction_ref> args) const
+    {
+        // Apply pointwise formula: y = x / (1 + |x|)
+        auto mb_ones = info.add_instruction(
+            migraphx::make_op("multibroadcast", {{"out_lens", args[0]->get_shape().lens()}}),
+            info.add_literal(migraphx::literal{migraphx::shape{args[0]->get_shape().type()}, {1}}));
+        auto abs = info.add_instruction(migraphx::make_op("abs"), args[0]);
+        auto add = info.add_instruction(migraphx::make_op("add"), abs, mb_ones);
+        return info.add_instruction(migraphx::make_op("div"), args[0], add);
+    }
+};
+
+} // namespace onnx
+} // namespace MIGRAPHX_INLINE_NS
+} // namespace migraphx

test/onnx/gen_onnx.py

@@ -4380,6 +4380,15 @@ def softmax_nonstd_input_test():
 
 
 @onnx_test
+def softsign_test():
+    x = helper.make_tensor_value_info('x', TensorProto.FLOAT, [5])
+    y = helper.make_tensor_value_info('y', TensorProto.FLOAT, [5])
+
+    node = onnx.helper.make_node('Softsign', inputs=['x'], outputs=['y'])
+
+    return ([node], [x], [y])
+
+
 def softplus_test():
     x = helper.make_tensor_value_info('x', TensorProto.FLOAT, [5])
     y = helper.make_tensor_value_info('y', TensorProto.FLOAT, [5])
@@ -4390,6 +4399,15 @@ def softplus_test():
 
 
 @onnx_test
+def softsign_nd_test():
+    x = helper.make_tensor_value_info('x', TensorProto.FLOAT16, [3, 4, 5])
+    y = helper.make_tensor_value_info('y', TensorProto.FLOAT16, [3, 4, 5])
+
+    node = onnx.helper.make_node('Softsign', inputs=['x'], outputs=['y'])
+
+    return ([node], [x], [y])
+
+
 def softplus_nd_test():
     x = helper.make_tensor_value_info('x', TensorProto.FLOAT16, [3, 4, 5])
     y = helper.make_tensor_value_info('y', TensorProto.FLOAT16, [3, 4, 5])

test/onnx/onnx_test.cpp

@@ -4135,6 +4135,46 @@ TEST_CASE(softplus_nd_test)
     EXPECT(p == prog);
 }
 
+TEST_CASE(softsign_test)
+{
+    migraphx::program p;
+    auto* mm = p.get_main_module();
+
+    std::vector<std::size_t> input_lens{5};
+    auto input_type = migraphx::shape::float_type;
+
+    auto x = mm->add_parameter("x", migraphx::shape{input_type, input_lens});
+    auto mb_ones =
+        mm->add_instruction(migraphx::make_op("multibroadcast", {{"out_lens", input_lens}}),
+                            mm->add_literal(migraphx::literal{migraphx::shape{input_type}, {1}}));
+    auto abs = mm->add_instruction(migraphx::make_op("abs"), x);
+    auto add = mm->add_instruction(migraphx::make_op("add"), abs, mb_ones);
+    mm->add_instruction(migraphx::make_op("div"), x, add);
+
+    auto prog = optimize_onnx("softsign_test.onnx");
+    EXPECT(p == prog);
+}
+
+TEST_CASE(softsign_nd_test)
+{
+    migraphx::program p;
+    auto* mm = p.get_main_module();
+
+    std::vector<std::size_t> input_lens{3, 4, 5};
+    auto input_type = migraphx::shape::half_type;
+
+    auto x = mm->add_parameter("x", migraphx::shape{input_type, input_lens});
+    auto mb_ones =
+        mm->add_instruction(migraphx::make_op("multibroadcast", {{"out_lens", input_lens}}),
+                            mm->add_literal(migraphx::literal{migraphx::shape{input_type}, {1}}));
+    auto abs = mm->add_instruction(migraphx::make_op("abs"), x);
+    auto add = mm->add_instruction(migraphx::make_op("add"), abs, mb_ones);
+    mm->add_instruction(migraphx::make_op("div"), x, add);
+
+    auto prog = optimize_onnx("softsign_nd_test.onnx");
+    EXPECT(p == prog);
+}
+
 TEST_CASE(split_minus_axis_test)
 {
     migraphx::program p;

test/onnx/softsign_nd_test.onnx

+softsign_nd_test:V
+
+
x
y"Softsignsoftsign_nd_testZ
+
x
+
+
+
+
+b
+
y
+
+
+
+
+B
\ No newline at end of file

test/onnx/softsign_test.onnx

+
softsign_test:C
+
+
x
y"Softsign
softsign_testZ
+
x
+
+

+b
+
y
+
+

+B
\ No newline at end of file

test/onnx/verify_onnx.cpp

@@ -609,6 +609,27 @@ TEST_CASE(softplus_test)
     EXPECT(migraphx::verify_range(result_vector, gold));
 }
 
+TEST_CASE(softsign_test)
+{
+    migraphx::program p = migraphx::parse_onnx("softsign_test.onnx");
+    p.compile(migraphx::ref::target{});
+
+    migraphx::shape s{migraphx::shape::float_type, {5}};
+    std::vector<float> data = {0, 1, 2, 3, 4};
+
+    migraphx::parameter_map pp;
+    pp["x"] = migraphx::argument(s, data.data());
+
+    auto result = p.eval(pp).back();
+    std::vector<float> result_vector;
+    result.visit([&](auto output) { result_vector.assign(output.begin(), output.end()); });
+    std::vector<float> gold(5);
+    std::transform(
+        data.begin(), data.end(), gold.begin(), [](auto x) { return x / (1.0 + std::abs(x)); });
+
+    EXPECT(migraphx::verify_range(result_vector, gold));
+}
+
 TEST_CASE(upsample_test)
 {
     migraphx::program p = migraphx::parse_onnx("upsample_test.onnx");

test/py/onnx_backend_test.py

@@ -282,8 +282,6 @@ def create_backend_test(testname=None, target_device=None):
         backend_test.exclude(r'test_size_cpu')
         backend_test.exclude(r'test_size_example_cpu')
         backend_test.exclude(r'test_softmax_cross_entropy_*')
-        backend_test.exclude(r'test_softsign_cpu')
-        backend_test.exclude(r'test_softsign_example_cpu')
         backend_test.exclude(r'test_Embedding_cpu')
 
         # real model tests