add sqdiff and tests

src/include/migraphx/op/sqdiff.hpp

+#ifndef MIGRAPHX_GUARD_OPERATORS_SQDIFF_HPP
+#define MIGRAPHX_GUARD_OPERATORS_SQDIFF_HPP
+
+#include <migraphx/op/binary.hpp>
+
+namespace migraphx {
+inline namespace MIGRAPHX_INLINE_NS {
+namespace op {
+
+struct sqdiff : binary<sqdiff>
+{
+    auto apply() const
+    {
+        return [](auto x, auto y) { return (x - y) * (x - y); };
+    }
+};
+
+} // namespace op
+} // namespace MIGRAPHX_INLINE_NS
+} // namespace migraphx
+
+#endif

src/include/migraphx/operators.hpp

@@ -58,6 +58,7 @@
 #include <migraphx/op/sin.hpp>
 #include <migraphx/op/slice.hpp>
 #include <migraphx/op/softmax.hpp>
+#include <migraphx/op/sqdiff.hpp>
 #include <migraphx/op/squeeze.hpp>
 #include <migraphx/op/sub.hpp>
 #include <migraphx/op/tanh.hpp>

src/targets/gpu/CMakeLists.txt

@@ -40,6 +40,7 @@ add_library(migraphx_device
     device/clip.cpp
     device/reduce_sum.cpp
     device/reduce_mean.cpp
+    device/sqdiff.cpp
 )
 set_target_properties(migraphx_device PROPERTIES EXPORT_NAME device)
 rocm_clang_tidy_check(migraphx_device)

src/targets/gpu/device/sqdiff.cpp

+#include <migraphx/gpu/device/sqdiff.hpp>
+#include <migraphx/gpu/device/nary.hpp>
+
+namespace migraphx {
+inline namespace MIGRAPHX_INLINE_NS {
+namespace gpu {
+namespace device {
+
+void sqdiff(hipStream_t stream, const argument& result, const argument& arg1, const argument& arg2)
+{
+    nary(stream, result, arg2, arg1)([](auto x, auto y) { return (x - y) * (x - y); });
+}
+
+} // namespace device
+} // namespace gpu
+} // namespace MIGRAPHX_INLINE_NS
+} // namespace migraphx

src/targets/gpu/include/migraphx/gpu/device/sqdiff.hpp

+
+#ifndef MIGRAPHX_GUARD_RTGLIB_DEVICE_SQDIFF_HPP
+#define MIGRAPHX_GUARD_RTGLIB_DEVICE_SQDIFF_HPP
+
+#include <migraphx/argument.hpp>
+#include <migraphx/config.hpp>
+#include <hip/hip_runtime_api.h>
+
+namespace migraphx {
+inline namespace MIGRAPHX_INLINE_NS {
+namespace gpu {
+namespace device {
+
+void sqdiff(hipStream_t stream, const argument& result, const argument& arg1, const argument& arg2);
+
+} // namespace device
+} // namespace gpu
+} // namespace MIGRAPHX_INLINE_NS
+} // namespace migraphx
+
+#endif

src/targets/gpu/include/migraphx/gpu/sqdiff.hpp

+#ifndef MIGRAPHX_GUARD_RTGLIB_SQDIFF_HPP
+#define MIGRAPHX_GUARD_RTGLIB_SQDIFF_HPP
+
+#include <migraphx/gpu/oper.hpp>
+#include <migraphx/gpu/device/sqdiff.hpp>
+
+namespace migraphx {
+inline namespace MIGRAPHX_INLINE_NS {
+namespace gpu {
+
+struct hip_sqdiff : binary_device<hip_sqdiff, device::sqdiff>
+{
+};
+
+} // namespace gpu
+} // namespace MIGRAPHX_INLINE_NS
+} // namespace migraphx
+
+#endif

src/targets/gpu/lowering.cpp

@@ -52,6 +52,7 @@
 #include <migraphx/gpu/clip.hpp>
 #include <migraphx/gpu/reduce_sum.hpp>
 #include <migraphx/gpu/reduce_mean.hpp>
+#include <migraphx/gpu/sqdiff.hpp>
 #include <utility>
 #include <functional>
 #include <algorithm>
@@ -101,6 +102,7 @@ struct miopen_apply
         add_generic_op<hip_mul>("mul");
         add_generic_op<hip_max>("max");
         add_generic_op<hip_min>("min");
+        add_generic_op<hip_sqdiff>("sqdiff");
 
         add_extend_op<miopen_gemm, op::dot>("dot");
         add_extend_op<miopen_contiguous, op::contiguous>("contiguous");

src/tf/tf.cpp

@@ -157,6 +157,7 @@ struct tf_parser
 
         add_binary_op("Add", op::add{});
         add_binary_op("Mul", op::mul{});
+        add_binary_op("SquaredDifference", op::sqdiff{});
         add_binary_op("Sub", op::sub{});
 
         add_mem_op("AvgPool", &tf_parser::parse_pooling);

test/cpu_ops_test.cpp

@@ -1853,4 +1853,19 @@ TEST_CASE(reduce_mean_int)
     EXPECT(results_vector == gold);
 }
 
+TEST_CASE(sqdiff_test)
+{
+    migraphx::program p;
+    migraphx::shape s{migraphx::shape::float_type, {3}};
+    auto l1 = p.add_literal(migraphx::literal{s, {-1, 0, 1}});
+    auto l2 = p.add_literal(migraphx::literal{s, {1, 2, 3}});
+    p.add_instruction(migraphx::op::sqdiff{}, l1, l2);
+    p.compile(migraphx::cpu::target{});
+    auto result = p.eval({});
+    std::vector<float> results_vector(3);
+    result.visit([&](auto output) { results_vector.assign(output.begin(), output.end()); });
+    std::vector<float> gold = {4, 4, 4};
+    EXPECT(migraphx::verify_range(results_vector, gold));
+}
+
 int main(int argc, const char* argv[]) { test::run(argc, argv); }

test/tf/sqdiff_test.pb

+
+:
+
0Placeholder*
+dtype0
*
+shape:

+:
+
1Placeholder*
+dtype0
*
+shape:

+*
+sqdiffSquaredDifference
0
1*
+
T0
"
\ No newline at end of file

test/tf/tf_test.cpp

@@ -339,6 +339,17 @@ TEST_CASE(softmax_test)
     EXPECT(p == prog);
 }
 
+TEST_CASE(sqdiff_test)
+{
+    migraphx::program p;
+    auto l0 = p.add_parameter("0", migraphx::shape{migraphx::shape::float_type, {1, 2, 2, 3}});
+    auto l1 = p.add_parameter("1", migraphx::shape{migraphx::shape::float_type, {1, 2, 2, 3}});
+    p.add_instruction(migraphx::op::sqdiff{}, l0, l1);
+    auto prog = optimize_tf("sqdiff_test.pb", false);
+
+    EXPECT(p == prog);
+}
+
 TEST_CASE(squeeze_test)
 {
     migraphx::program p;