Add HardSwish op ONNX parser (#1066)

Add HardSwish to HardSigmoid parser

HardSwish formula is y = x * HardSigmoid<alpha=1/6, beta=0.5>(x)
HardSigmoid parser sets alpha to 1/6 and adds the mul instruction if op name is HardSwish

Resolves #1062

src/onnx/parse_hardsigmoid.cpp

@@ -10,20 +10,27 @@ namespace onnx {
 
 struct parse_hardsigmoid : op_parser<parse_hardsigmoid>
 {
-    std::vector<op_desc> operators() const { return {{"HardSigmoid"}}; }
+    std::vector<op_desc> operators() const { return {{"HardSigmoid"}, {"HardSwish"}}; }
 
-    instruction_ref parse(const op_desc& /*opd*/,
+    instruction_ref parse(const op_desc& opd,
                           const onnx_parser& /*parser*/,
                           const onnx_parser::node_info& info,
                           std::vector<instruction_ref> args) const
     {
         float alpha = 0.2;
         float beta  = 0.5;
-        if(contains(info.attributes, "alpha"))
-            alpha = info.attributes.at("alpha").f();
+        if(opd.onnx_name == "HardSwish")
+        {
+            alpha = 1.0 / 6.0;
+        }
+        else
+        {
+            if(contains(info.attributes, "alpha"))
+                alpha = info.attributes.at("alpha").f();
 
-        if(contains(info.attributes, "beta"))
-            beta = info.attributes.at("beta").f();
+            if(contains(info.attributes, "beta"))
+                beta = info.attributes.at("beta").f();
+        }
 
         auto input_lens = args[0]->get_shape().lens();
         auto input_type = args[0]->get_shape().type();
@@ -40,9 +47,13 @@ struct parse_hardsigmoid : op_parser<parse_hardsigmoid>
             migraphx::make_op("multibroadcast", {{"out_lens", input_lens}}),
             info.add_literal(migraphx::literal{migraphx::shape{input_type}, {1}}));
 
-        auto mul = info.add_instruction(migraphx::make_op("mul"), mb_alpha, args[0]);
-        auto add = info.add_instruction(migraphx::make_op("add"), mb_beta, mul);
-        return info.add_instruction(migraphx::make_op("clip"), add, mb_zero, mb_one);
+        auto mul         = info.add_instruction(migraphx::make_op("mul"), mb_alpha, args[0]);
+        auto add         = info.add_instruction(migraphx::make_op("add"), mb_beta, mul);
+        auto hardsigmoid = info.add_instruction(migraphx::make_op("clip"), add, mb_zero, mb_one);
+        if(opd.onnx_name == "HardSwish")
+            return info.add_instruction(migraphx::make_op("mul"), args[0], hardsigmoid);
+
+        return hardsigmoid;
     }
 };
 

test/onnx/gen_onnx.py

@@ -1694,6 +1694,16 @@ def hardsigmoid_verify_test():
     return ([node], [x], [y])
 
 
+@onnx_test
+def hardswish_test():
+    x = helper.make_tensor_value_info('x', TensorProto.FLOAT, [2, 5])
+    y = helper.make_tensor_value_info('y', TensorProto.FLOAT, [2, 5])
+
+    node = onnx.helper.make_node('HardSwish', inputs=['x'], outputs=['y'])
+
+    return ([node], [x], [y])
+
+
 @onnx_test
 def if_else_test():
     x = onnx.helper.make_tensor_value_info('x', onnx.TensorProto.FLOAT, [2, 3])

test/onnx/hardswish_test.onnx

+hardswish_test:M
+
+
x
y"	HardSwishhardswish_testZ
+
x
+

+
+b
+
y
+

+
+B
\ No newline at end of file

test/onnx/onnx_test.cpp

@@ -1680,6 +1680,41 @@ TEST_CASE(hardsigmoid_half_test)
     EXPECT(p == prog);
 }
 
+TEST_CASE(hardswish_test)
+{
+    migraphx::program p;
+    auto* mm = p.get_main_module();
+    std::vector<std::size_t> input_lens{2, 5};
+    auto input_type = migraphx::shape::float_type;
+    migraphx::shape s{input_type, input_lens};
+    auto x = mm->add_parameter("x", s);
+
+    float alpha = 1.0 / 6.0;
+    float beta  = 0.5;
+
+    auto mb_alpha = mm->add_instruction(
+        migraphx::make_op("multibroadcast", {{"out_lens", input_lens}}),
+        mm->add_literal(migraphx::literal{migraphx::shape{input_type}, {alpha}}));
+    auto mb_beta = mm->add_instruction(
+        migraphx::make_op("multibroadcast", {{"out_lens", input_lens}}),
+        mm->add_literal(migraphx::literal{migraphx::shape{input_type}, {beta}}));
+    auto mb_zero =
+        mm->add_instruction(migraphx::make_op("multibroadcast", {{"out_lens", input_lens}}),
+                            mm->add_literal(migraphx::literal{migraphx::shape{input_type}, {0}}));
+    auto mb_one =
+        mm->add_instruction(migraphx::make_op("multibroadcast", {{"out_lens", input_lens}}),
+                            mm->add_literal(migraphx::literal{migraphx::shape{input_type}, {1}}));
+
+    auto mul         = mm->add_instruction(migraphx::make_op("mul"), mb_alpha, x);
+    auto add         = mm->add_instruction(migraphx::make_op("add"), mb_beta, mul);
+    auto hardsigmoid = mm->add_instruction(migraphx::make_op("clip"), add, mb_zero, mb_one);
+    mm->add_instruction(migraphx::make_op("mul"), x, hardsigmoid);
+
+    auto prog = optimize_onnx("hardswish_test.onnx");
+
+    EXPECT(p == prog);
+}
+
 TEST_CASE(if_else_test)
 {
     migraphx::program p;

test/py/onnx_backend_test.py

@@ -119,6 +119,7 @@ def create_backend_test(testname=None, target_device=None):
         backend_test.include(r'.*test_globalmaxpool.*')
         backend_test.include(r'.*test_greater.*')
         backend_test.include(r'.*test_hardsigmoid.*')
+        backend_test.include(r'.*test_hardswish.*')
         backend_test.include(r'.*test_identity.*')
         backend_test.include(r'.*test_if.*')
         backend_test.include(r'.*test_LeakyReLU*')