Merge branch 'develop' into hcc26

src/include/migraphx/op/pow.hpp

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

src/include/migraphx/op/rsqrt.hpp

+#ifndef MIGRAPHX_GUARD_OPERATORS_RSQRT_HPP
+#define MIGRAPHX_GUARD_OPERATORS_RSQRT_HPP
+
+#include <migraphx/op/unary.hpp>
+#include <cmath>
+
+namespace migraphx {
+inline namespace MIGRAPHX_INLINE_NS {
+namespace op {
+
+struct rsqrt : unary<rsqrt>
+{
+    auto apply() const
+    {
+        return [](auto x) { return 1 / std::sqrt(x); };
+    }
+};
+
+} // namespace op
+} // namespace MIGRAPHX_INLINE_NS
+} // namespace migraphx
+
+#endif

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/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

@@ -45,6 +45,7 @@
 #include <migraphx/op/outline.hpp>
 #include <migraphx/op/pad.hpp>
 #include <migraphx/op/pooling.hpp>
+#include <migraphx/op/pow.hpp>
 #include <migraphx/op/reduce_sum.hpp>
 #include <migraphx/op/reduce_mean.hpp>
 #include <migraphx/op/relu.hpp>
@@ -52,12 +53,15 @@
 #include <migraphx/op/rnn.hpp>
 #include <migraphx/op/rnn_last_cell_output.hpp>
 #include <migraphx/op/rnn_last_output.hpp>
+#include <migraphx/op/rsqrt.hpp>
 #include <migraphx/op/scalar.hpp>
 #include <migraphx/op/sigmoid.hpp>
 #include <migraphx/op/sinh.hpp>
 #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>
 #include <migraphx/op/tanh.hpp>

src/onnx/onnx.cpp

@@ -54,11 +54,13 @@ 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{});
         add_binary_op("Mul", op::mul{});
         add_binary_op("Sub", op::sub{});
+        add_binary_op("Pow", op::pow{});
 
         add_variadic_op("Sum", op::add{});
         add_variadic_op("Max", op::max{});

src/targets/gpu/CMakeLists.txt

@@ -37,9 +37,14 @@ add_library(migraphx_device
     device/pad.cpp
     device/gather.cpp
     device/sub.cpp
+    device/div.cpp
     device/clip.cpp
     device/reduce_sum.cpp
+    device/rsqrt.cpp
+    device/sqrt.cpp
     device/reduce_mean.cpp
+    device/pow.cpp
+    device/sqdiff.cpp
 )
 set_target_properties(migraphx_device PROPERTIES EXPORT_NAME device)
 rocm_clang_tidy_check(migraphx_device)

src/targets/gpu/device/div.cpp

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

src/targets/gpu/device/pow.cpp

+#include <migraphx/gpu/device/pow.hpp>
+#include <migraphx/gpu/device/nary.hpp>
+
+namespace migraphx {
+inline namespace MIGRAPHX_INLINE_NS {
+namespace gpu {
+namespace device {
+
+void pow(hipStream_t stream, const argument& result, const argument& arg1, const argument& arg2)
+{
+    nary(stream, result, arg1, arg2)(
+        [](auto b, auto e) { return ::pow(to_hip_type(b), to_hip_type(e)); });
+}
+
+} // namespace device
+} // namespace gpu
+} // namespace MIGRAPHX_INLINE_NS
+} // namespace migraphx

src/targets/gpu/device/rsqrt.cpp

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

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, arg1, arg2)([](auto x, auto y) { return (x - y) * (x - y); });
+}
+
+} // namespace device
+} // namespace gpu
+} // namespace MIGRAPHX_INLINE_NS
+} // namespace migraphx

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/device/sub.cpp

@@ -8,7 +8,7 @@ namespace device {
 
 void sub(hipStream_t stream, const argument& result, const argument& arg1, const argument& arg2)
 {
-    nary(stream, result, arg1, arg2)([](auto x, auto y) { return y - x; });
+    nary(stream, result, arg1, arg2)([](auto x, auto y) { return x - y; });
 }
 
 } // namespace device

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

+#ifndef MIGRAPHX_GUARD_RTGLIB_DEVICE_DIV_HPP
+#define MIGRAPHX_GUARD_RTGLIB_DEVICE_DIV_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 div(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/device/pow.hpp

+
+#ifndef MIGRAPHX_GUARD_RTGLIB_DEVICE_POW_HPP
+#define MIGRAPHX_GUARD_RTGLIB_DEVICE_POW_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 pow(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/device/rsqrt.hpp

+#ifndef MIGRAPHX_GUARD_RTGLIB_DEVICE_RSQRT_HPP
+#define MIGRAPHX_GUARD_RTGLIB_DEVICE_RSQRT_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 rsqrt(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/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/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/div.hpp

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

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

@@ -88,7 +88,7 @@ struct binary_device : oper<Derived>
 
     argument compute(context& ctx, const shape&, const std::vector<argument>& args) const
     {
-        F(ctx.get_stream().get(), args[2], args[1], args[0]);
+        F(ctx.get_stream().get(), args[2], args[0], args[1]);
         return args[2];
     }