Merge branch 'develop' into pow_operator

src/include/migraphx/op/sqrt.hpp

+#ifndef MIGRAPHX_GUARD_OPERATORS_SQRT_HPP
+#define MIGRAPHX_GUARD_OPERATORS_SQRT_HPP
+
+#include <migraphx/op/unary.hpp>
+#include <migraphx/config.hpp>
+
+namespace migraphx {
+inline namespace MIGRAPHX_INLINE_NS {
+namespace op {
+
+struct sqrt : unary<sqrt>
+{
+    auto apply() const
+    {
+        return [](auto x) { return std::sqrt(x); };
+    }
+};
+
+} // namespace op
+} // namespace MIGRAPHX_INLINE_NS
+} // namespace migraphx
+
+#endif

src/include/migraphx/operators.hpp

@@ -59,6 +59,7 @@
 #include <migraphx/op/sin.hpp>
 #include <migraphx/op/slice.hpp>
 #include <migraphx/op/softmax.hpp>
+#include <migraphx/op/sqrt.hpp>
 #include <migraphx/op/sqdiff.hpp>
 #include <migraphx/op/squeeze.hpp>
 #include <migraphx/op/sub.hpp>

src/onnx/onnx.cpp

@@ -54,6 +54,7 @@ struct onnx_parser
         add_generic_op("Asin", op::asin{});
         add_generic_op("Acos", op::acos{});
         add_generic_op("Atan", op::atan{});
+        add_generic_op("Sqrt", op::sqrt{});
 
         add_binary_op("Add", op::add{});
         add_binary_op("Div", op::div{});

src/targets/gpu/CMakeLists.txt

@@ -40,6 +40,7 @@ add_library(migraphx_device
     device/div.cpp
     device/clip.cpp
     device/reduce_sum.cpp
+    device/sqrt.cpp
     device/reduce_mean.cpp
     device/pow.cpp
     device/sqdiff.cpp

src/targets/gpu/device/sqrt.cpp

+#include <migraphx/gpu/device/sqrt.hpp>
+#include <migraphx/gpu/device/nary.hpp>
+#include <migraphx/gpu/device/types.hpp>
+
+namespace migraphx {
+inline namespace MIGRAPHX_INLINE_NS {
+namespace gpu {
+namespace device {
+
+void sqrt(hipStream_t stream, const argument& result, const argument& arg)
+{
+    nary(stream, result, arg)([](auto x) { return ::sqrt(to_hip_type(x)); });
+}
+
+} // namespace device
+} // namespace gpu
+} // namespace MIGRAPHX_INLINE_NS
+} // namespace migraphx

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

+#ifndef MIGRAPHX_GUARD_RTGLIB_DEVICE_SQRT_HPP
+#define MIGRAPHX_GUARD_RTGLIB_DEVICE_SQRT_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 sqrt(hipStream_t stream, const argument& result, const argument& arg);
+
+} // namespace device
+} // namespace gpu
+} // namespace MIGRAPHX_INLINE_NS
+} // namespace migraphx
+
+#endif

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

+#ifndef MIGRAPHX_GUARD_RTGLIB_SQRT_HPP
+#define MIGRAPHX_GUARD_RTGLIB_SQRT_HPP
+
+#include <migraphx/gpu/oper.hpp>
+#include <migraphx/gpu/device/sqrt.hpp>
+
+namespace migraphx {
+inline namespace MIGRAPHX_INLINE_NS {
+namespace gpu {
+
+struct hip_sqrt : unary_device<hip_sqrt, device::sqrt>
+{
+};
+
+} // namespace gpu
+} // namespace MIGRAPHX_INLINE_NS
+} // namespace migraphx
+
+#endif

src/targets/gpu/lowering.cpp

@@ -52,6 +52,7 @@
 #include <migraphx/gpu/convert.hpp>
 #include <migraphx/gpu/clip.hpp>
 #include <migraphx/gpu/reduce_sum.hpp>
+#include <migraphx/gpu/sqrt.hpp>
 #include <migraphx/gpu/reduce_mean.hpp>
 #include <migraphx/gpu/pow.hpp>
 #include <migraphx/gpu/sqdiff.hpp>
@@ -101,6 +102,7 @@ struct miopen_apply
         add_generic_op<hip_asin>("asin");
         add_generic_op<hip_acos>("acos");
         add_generic_op<hip_atan>("atan");
+        add_generic_op<hip_sqrt>("sqrt");
         add_generic_op<hip_mul>("mul");
         add_generic_op<hip_div>("div");
         add_generic_op<hip_max>("max");

test/cpu_ops_test.cpp

@@ -542,6 +542,21 @@ TEST_CASE(erf_test)
     EXPECT(migraphx::verify_range(results_vector, gold));
 }
 
+TEST_CASE(sqrt_test)
+{
+    migraphx::program p;
+    migraphx::shape s{migraphx::shape::float_type, {5}};
+    auto l = p.add_literal(
+        migraphx::literal{s, {1.02481645, 0.85643062, 0.03404123, 0.92791926, 0.10569184}});
+    p.add_instruction(migraphx::op::sqrt{}, l);
+    p.compile(migraphx::cpu::target{});
+    auto result = p.eval({});
+    std::vector<float> results_vector;
+    result.visit([&](auto output) { results_vector.assign(output.begin(), output.end()); });
+    std::vector<float> gold = {1.01233218, 0.92543537, 0.18450265, 0.96328566, 0.32510282};
+    EXPECT(migraphx::verify_range(results_vector, gold));
+}
+
 TEST_CASE(log_test)
 {
     migraphx::program p;

test/gpu/miopen.cpp

@@ -255,6 +255,19 @@ struct test_erf : verify_program<test_erf>
     }
 };
 
+struct test_sqrt : verify_program<test_sqrt>
+{
+    migraphx::program create_program() const
+    {
+        migraphx::program p;
+        migraphx::shape s{migraphx::shape::float_type, {2, 3, 4, 6}};
+        auto param     = p.add_parameter("x", s);
+        auto param_abs = p.add_instruction(migraphx::op::abs{}, param);
+        p.add_instruction(migraphx::op::sqrt{}, param_abs);
+        return p;
+    }
+};
+
 struct test_log : verify_program<test_log>
 {
     migraphx::program create_program() const

test/onnx/onnx_test.cpp

@@ -202,6 +202,16 @@ TEST_CASE(erf_test)
     EXPECT(p == prog);
 }
 
+TEST_CASE(sqrt_test)
+{
+    migraphx::program p;
+    auto input = p.add_parameter("x", migraphx::shape{migraphx::shape::float_type, {10, 15}});
+    p.add_instruction(migraphx::op::sqrt{}, input);
+
+    auto prog = migraphx::parse_onnx("sqrt_test.onnx");
+    EXPECT(p == prog);
+}
+
 TEST_CASE(log_test)
 {
     migraphx::program p;

test/onnx/sqrt_test.onnx

+sqrt-example:C
+
+
x
y"Sqrt	test_sqrtZ
+
x
+

+
+
+b
+
y
+

+
+
+B