Merge branch 'develop' into driver

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

 #ifndef MIGRAPHX_GUARD_RTGLIB_UNARY_HPP
 #define MIGRAPHX_GUARD_RTGLIB_UNARY_HPP
 
-#include <migraphx/gpu/lowering.hpp>
-#include <migraphx/manage_ptr.hpp>
-#include <migraphx/instruction.hpp>
-#include <migraphx/operators.hpp>
-#include <migraphx/generate.hpp>
-#include <migraphx/shape_for_each.hpp>
-#include <migraphx/gpu/miopen.hpp>
 #include <migraphx/gpu/hip.hpp>
-#include <migraphx/dfor.hpp>
-#include <migraphx/type_name.hpp>
-#include <migraphx/gpu/device/contiguous.hpp>
-#include <migraphx/iterator_for.hpp>
-#include <migraphx/gpu/rocblas.hpp>
 #include <migraphx/gpu/context.hpp>
+#include <migraphx/shape.hpp>
+#include <migraphx/argument.hpp>
 #include <migraphx/config.hpp>
+#include <migraphx/type_name.hpp>
 #include <utility>
 #include <iostream>
 
@@ -54,7 +45,15 @@ struct unary_device : oper<Derived>
     shape compute_shape(const std::vector<shape>& inputs) const
     {
         check_shapes{inputs, *this}.has(2);
-        return inputs.at(0);
+        auto s = inputs.at(0);
+        if(s.packed())
+        {
+            return s;
+        }
+        else
+        {
+            return {s.type(), s.lens()};
+        }
     }
 
     argument compute(context& ctx, const shape&, const std::vector<argument>& args) const
@@ -63,7 +62,10 @@ struct unary_device : oper<Derived>
         return args[1];
     }
 
-    int output_alias(const std::vector<shape>& shapes) const { return shapes.size() - 1; }
+    std::ptrdiff_t output_alias(const std::vector<shape>& shapes) const
+    {
+        return shapes.size() - 1;
+    }
 };
 
 template <class Derived, void (*F)(hipStream_t, const argument&, const argument&, const argument&)>
@@ -72,7 +74,16 @@ struct binary_device : oper<Derived>
     shape compute_shape(const std::vector<shape>& inputs) const
     {
         check_shapes{inputs, *this}.has(3);
-        return inputs.at(0);
+        auto s0 = inputs.at(0);
+        auto s1 = inputs.at(1);
+        if(s0 == s1 and s0.packed())
+        {
+            return s0;
+        }
+        else
+        {
+            return {s0.type(), s0.lens()};
+        }
     }
 
     argument compute(context& ctx, const shape&, const std::vector<argument>& args) const
@@ -81,7 +92,10 @@ struct binary_device : oper<Derived>
         return args[2];
     }
 
-    int output_alias(const std::vector<shape>& shapes) const { return shapes.size() - 1; }
+    std::ptrdiff_t output_alias(const std::vector<shape>& shapes) const
+    {
+        return shapes.size() - 1;
+    }
 };
 
 } // namespace gpu

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

 #ifndef MIGRAPHX_GUARD_RTGLIB_PAD_HPP
 #define MIGRAPHX_GUARD_RTGLIB_PAD_HPP
 
-#include <migraphx/gpu/lowering.hpp>
-#include <migraphx/manage_ptr.hpp>
-#include <migraphx/instruction.hpp>
-#include <migraphx/operators.hpp>
-#include <migraphx/generate.hpp>
-#include <migraphx/shape_for_each.hpp>
-#include <migraphx/config.hpp>
-#include <migraphx/gpu/miopen.hpp>
-#include <migraphx/gpu/hip.hpp>
-#include <migraphx/dfor.hpp>
-#include <migraphx/gpu/device/pad.hpp>
-#include <migraphx/gpu/device/add.hpp>
-#include <migraphx/iterator_for.hpp>
-#include <migraphx/gpu/rocblas.hpp>
-#include <migraphx/gpu/context.hpp>
-#include <utility>
+#include <migraphx/shape.hpp>
+#include <migraphx/op/pad.hpp>
 
 namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {
 namespace gpu {
 
+struct context;
+
 struct hip_pad
 {
     op::pad op;
 
+    template <class Self, class F>
+    static auto reflect(Self& self, F f)
+    {
+        return migraphx::reflect(self.op, f);
+    }
+
     std::string name() const { return "gpu::pad"; }
     shape compute_shape(std::vector<shape> inputs) const;
     argument
     compute(context& ctx, const shape& output_shape, const std::vector<argument>& args) const;
-    int output_alias(const std::vector<shape>& shapes) const { return shapes.size() - 1; }
+    std::ptrdiff_t output_alias(const std::vector<shape>& shapes) const
+    {
+        return shapes.size() - 1;
+    }
 };
 
 } // namespace gpu

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

 #ifndef MIGRAPHX_GUARD_RTGLIB_POOLING_HPP
 #define MIGRAPHX_GUARD_RTGLIB_POOLING_HPP
 
-#include <migraphx/gpu/lowering.hpp>
-#include <migraphx/manage_ptr.hpp>
-#include <migraphx/instruction.hpp>
-#include <migraphx/operators.hpp>
-#include <migraphx/generate.hpp>
-#include <migraphx/shape_for_each.hpp>
-#include <migraphx/config.hpp>
+#include <migraphx/shape.hpp>
+#include <migraphx/op/pooling.hpp>
 #include <migraphx/gpu/miopen.hpp>
-#include <migraphx/gpu/hip.hpp>
-#include <migraphx/dfor.hpp>
-#include <migraphx/gpu/device/contiguous.hpp>
-#include <migraphx/gpu/device/add.hpp>
-#include <migraphx/iterator_for.hpp>
-#include <migraphx/gpu/rocblas.hpp>
-#include <migraphx/gpu/context.hpp>
-#include <utility>
 
 namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {
 namespace gpu {
 
+struct context;
+
 struct miopen_pooling
 {
     op::pooling op;
     shared<pooling_descriptor> pd;
 
+    template <class Self, class F>
+    static auto reflect(Self& self, F f)
+    {
+        return migraphx::reflect(self.op, f);
+    }
+
     std::string name() const { return "gpu::pooling"; }
     shape compute_shape(const std::vector<shape>& inputs) const;
     argument
     compute(context& ctx, const shape& output_shape, const std::vector<argument>& args) const;
-    int output_alias(const std::vector<shape>& shapes) const { return shapes.size() - 1; }
+    std::ptrdiff_t output_alias(const std::vector<shape>& shapes) const
+    {
+        return shapes.size() - 1;
+    }
 };
 
 } // namespace gpu

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

 #ifndef MIGRAPHX_GUARD_RTGLIB_RELU_HPP
 #define MIGRAPHX_GUARD_RTGLIB_RELU_HPP
 
-#include <migraphx/gpu/lowering.hpp>
-#include <migraphx/manage_ptr.hpp>
-#include <migraphx/instruction.hpp>
-#include <migraphx/operators.hpp>
-#include <migraphx/generate.hpp>
-#include <migraphx/shape_for_each.hpp>
-#include <migraphx/config.hpp>
+#include <migraphx/shape.hpp>
 #include <migraphx/gpu/miopen.hpp>
-#include <migraphx/gpu/hip.hpp>
-#include <migraphx/dfor.hpp>
-#include <migraphx/gpu/device/contiguous.hpp>
-#include <migraphx/gpu/device/add.hpp>
-#include <migraphx/iterator_for.hpp>
-#include <migraphx/gpu/rocblas.hpp>
-#include <migraphx/gpu/context.hpp>
-#include <utility>
 
 namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {
 namespace gpu {
 
+struct context;
+
 struct miopen_relu
 {
     shared<activation_descriptor> ad;
+
+    template <class Self, class F>
+    static auto reflect(Self& self, F f)
+    {
+        return gpu::reflect(self.ad.get(), f);
+    }
+
     std::string name() const { return "gpu::relu"; }
     shape compute_shape(const std::vector<shape>& inputs) const;
     argument
     compute(context& ctx, const shape& output_shape, const std::vector<argument>& args) const;
-    int output_alias(const std::vector<shape>& shapes) const { return shapes.size() - 1; }
+    std::ptrdiff_t output_alias(const std::vector<shape>& shapes) const
+    {
+        return shapes.size() - 1;
+    }
 };
 
 } // namespace gpu

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

@@ -2,7 +2,6 @@
 #define MIGRAPHX_GUARD_MIGRAPHLIB_ROCBLAS_HPP
 
 #include <migraphx/manage_ptr.hpp>
-#include <migraphx/operators.hpp>
 #include <migraphx/config.hpp>
 #include <rocblas.h>
 

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

+#ifndef MIGRAPHX_GUARD_RTGLIB_GPU_SCHEDULE_MODEL_HPP
+#define MIGRAPHX_GUARD_RTGLIB_GPU_SCHEDULE_MODEL_HPP
+
+#include <migraphx/config.hpp>
+#include <migraphx/instruction_ref.hpp>
+#include <vector>
+
+namespace migraphx {
+inline namespace MIGRAPHX_INLINE_NS {
+
+struct program;
+struct operation;
+
+namespace gpu {
+
+struct schedule_model
+{
+    std::size_t streams = 0;
+    std::size_t concurrency() const;
+    void sched(program& p, instruction_ref ins, std::size_t n) const;
+    void wait(program& p, instruction_ref ins, std::size_t wait_id) const;
+    void record(program& p, instruction_ref ins, std::size_t wait_id) const;
+    std::size_t weight(const operation& op) const;
+};
+
+} // namespace gpu
+} // namespace MIGRAPHX_INLINE_NS
+} // namespace migraphx
+
+#endif

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

 #ifndef MIGRAPHX_GUARD_RTGLIB_SIGMOID_HPP
 #define MIGRAPHX_GUARD_RTGLIB_SIGMOID_HPP
 
-#include <migraphx/gpu/lowering.hpp>
-#include <migraphx/manage_ptr.hpp>
-#include <migraphx/instruction.hpp>
-#include <migraphx/operators.hpp>
-#include <migraphx/generate.hpp>
-#include <migraphx/shape_for_each.hpp>
-#include <migraphx/config.hpp>
+#include <migraphx/shape.hpp>
 #include <migraphx/gpu/miopen.hpp>
-#include <migraphx/gpu/hip.hpp>
-#include <migraphx/dfor.hpp>
-#include <migraphx/gpu/device/contiguous.hpp>
-#include <migraphx/gpu/device/add.hpp>
-#include <migraphx/iterator_for.hpp>
-#include <migraphx/gpu/rocblas.hpp>
-#include <migraphx/gpu/context.hpp>
-#include <utility>
 
 namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {
 namespace gpu {
 
+struct context;
+
 struct miopen_sigmoid
 {
     shared<activation_descriptor> ad;
+
+    template <class Self, class F>
+    static auto reflect(Self& self, F f)
+    {
+        return gpu::reflect(self.ad.get(), f);
+    }
+
     std::string name() const { return "gpu::sigmoid"; }
     shape compute_shape(const std::vector<shape>& inputs) const;
     argument
     compute(context& ctx, const shape& output_shape, const std::vector<argument>& args) const;
-    int output_alias(const std::vector<shape>& shapes) const { return shapes.size() - 1; }
+    std::ptrdiff_t output_alias(const std::vector<shape>& shapes) const
+    {
+        return shapes.size() - 1;
+    }
 };
 
 } // namespace gpu

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

 #ifndef MIGRAPHX_GUARD_RTGLIB_SIN_HPP
 #define MIGRAPHX_GUARD_RTGLIB_SIN_HPP
 
-#include <migraphx/gpu/lowering.hpp>
 #include <migraphx/gpu/oper.hpp>
-#include <migraphx/manage_ptr.hpp>
-#include <migraphx/instruction.hpp>
-#include <migraphx/operators.hpp>
-#include <migraphx/generate.hpp>
-#include <migraphx/shape_for_each.hpp>
-#include <migraphx/gpu/miopen.hpp>
-#include <migraphx/gpu/hip.hpp>
-#include <migraphx/dfor.hpp>
-#include <migraphx/gpu/device/contiguous.hpp>
 #include <migraphx/gpu/device/sin.hpp>
-#include <migraphx/iterator_for.hpp>
-#include <migraphx/gpu/rocblas.hpp>
-#include <migraphx/gpu/context.hpp>
-#include <migraphx/config.hpp>
-#include <utility>
 
 namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {

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

 #ifndef MIGRAPHX_GUARD_RTGLIB_SINH_HPP
 #define MIGRAPHX_GUARD_RTGLIB_SINH_HPP
 
-#include <migraphx/gpu/lowering.hpp>
 #include <migraphx/gpu/oper.hpp>
-#include <migraphx/manage_ptr.hpp>
-#include <migraphx/instruction.hpp>
-#include <migraphx/operators.hpp>
-#include <migraphx/generate.hpp>
-#include <migraphx/shape_for_each.hpp>
-#include <migraphx/gpu/miopen.hpp>
-#include <migraphx/gpu/hip.hpp>
-#include <migraphx/dfor.hpp>
-#include <migraphx/gpu/device/contiguous.hpp>
 #include <migraphx/gpu/device/sinh.hpp>
-#include <migraphx/iterator_for.hpp>
-#include <migraphx/gpu/rocblas.hpp>
-#include <migraphx/gpu/context.hpp>
-#include <migraphx/config.hpp>
-#include <utility>
 
 namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {

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

 #ifndef MIGRAPHX_GUARD_RTGLIB_SOFTMAX_HPP
 #define MIGRAPHX_GUARD_RTGLIB_SOFTMAX_HPP
 
-#include <migraphx/gpu/lowering.hpp>
-#include <migraphx/manage_ptr.hpp>
-#include <migraphx/instruction.hpp>
-#include <migraphx/operators.hpp>
-#include <migraphx/generate.hpp>
-#include <migraphx/shape_for_each.hpp>
-#include <migraphx/config.hpp>
-#include <migraphx/gpu/miopen.hpp>
-#include <migraphx/gpu/hip.hpp>
-#include <migraphx/dfor.hpp>
-#include <migraphx/gpu/device/contiguous.hpp>
-#include <migraphx/gpu/device/add.hpp>
-#include <migraphx/iterator_for.hpp>
-#include <migraphx/gpu/rocblas.hpp>
-#include <migraphx/gpu/context.hpp>
-#include <utility>
+#include <migraphx/shape.hpp>
+#include <migraphx/op/softmax.hpp>
 
 namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {
 namespace gpu {
 
+struct context;
+
 struct miopen_softmax
 {
     op::softmax op;
+
+    template <class Self, class F>
+    static auto reflect(Self& self, F f)
+    {
+        return migraphx::reflect(self.op, f);
+    }
+
     std::string name() const { return "gpu::softmax"; }
     shape compute_shape(const std::vector<shape>& inputs) const;
     argument
     compute(context& ctx, const shape& output_shape, const std::vector<argument>& args) const;
-    int output_alias(const std::vector<shape>& shapes) const { return shapes.size() - 1; }
+    std::ptrdiff_t output_alias(const std::vector<shape>& shapes) const
+    {
+        return shapes.size() - 1;
+    }
 };
 
 } // namespace gpu

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

 #ifndef MIGRAPHX_GUARD_RTGLIB_SUB_HPP
 #define MIGRAPHX_GUARD_RTGLIB_SUB_HPP
 
-#include <migraphx/gpu/lowering.hpp>
 #include <migraphx/gpu/oper.hpp>
-#include <migraphx/manage_ptr.hpp>
-#include <migraphx/instruction.hpp>
-#include <migraphx/operators.hpp>
-#include <migraphx/generate.hpp>
-#include <migraphx/shape_for_each.hpp>
-#include <migraphx/gpu/miopen.hpp>
-#include <migraphx/gpu/hip.hpp>
-#include <migraphx/dfor.hpp>
-#include <migraphx/gpu/device/contiguous.hpp>
 #include <migraphx/gpu/device/sub.hpp>
-#include <migraphx/iterator_for.hpp>
-#include <migraphx/gpu/rocblas.hpp>
-#include <migraphx/gpu/context.hpp>
-#include <migraphx/config.hpp>
-#include <utility>
 
 namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {

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

 #ifndef MIGRAPHX_GUARD_RTGLIB_TAN_HPP
 #define MIGRAPHX_GUARD_RTGLIB_TAN_HPP
 
-#include <migraphx/gpu/lowering.hpp>
 #include <migraphx/gpu/oper.hpp>
-#include <migraphx/manage_ptr.hpp>
-#include <migraphx/instruction.hpp>
-#include <migraphx/operators.hpp>
-#include <migraphx/generate.hpp>
-#include <migraphx/shape_for_each.hpp>
-#include <migraphx/gpu/miopen.hpp>
-#include <migraphx/gpu/hip.hpp>
-#include <migraphx/dfor.hpp>
-#include <migraphx/gpu/device/contiguous.hpp>
 #include <migraphx/gpu/device/tan.hpp>
-#include <migraphx/iterator_for.hpp>
-#include <migraphx/gpu/rocblas.hpp>
-#include <migraphx/gpu/context.hpp>
-#include <migraphx/config.hpp>
-#include <utility>
 
 namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {

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

 #ifndef MIGRAPHX_GUARD_RTGLIB_TANH_HPP
 #define MIGRAPHX_GUARD_RTGLIB_TANH_HPP
 
-#include <migraphx/gpu/lowering.hpp>
-#include <migraphx/manage_ptr.hpp>
-#include <migraphx/instruction.hpp>
-#include <migraphx/operators.hpp>
-#include <migraphx/generate.hpp>
-#include <migraphx/shape_for_each.hpp>
-#include <migraphx/config.hpp>
+#include <migraphx/shape.hpp>
 #include <migraphx/gpu/miopen.hpp>
-#include <migraphx/gpu/hip.hpp>
-#include <migraphx/dfor.hpp>
-#include <migraphx/gpu/device/contiguous.hpp>
-#include <migraphx/gpu/device/add.hpp>
-#include <migraphx/iterator_for.hpp>
-#include <migraphx/gpu/rocblas.hpp>
-#include <migraphx/gpu/context.hpp>
-#include <utility>
 
 namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {
 namespace gpu {
 
+struct context;
+
 struct miopen_tanh
 {
     shared<activation_descriptor> ad;
+
+    template <class Self, class F>
+    static auto reflect(Self& self, F f)
+    {
+        return gpu::reflect(self.ad.get(), f);
+    }
+
     std::string name() const { return "gpu::tanh"; }
     shape compute_shape(const std::vector<shape>& inputs) const;
     argument
     compute(context& ctx, const shape& output_shape, const std::vector<argument>& args) const;
-    int output_alias(const std::vector<shape>& shapes) const { return shapes.size() - 1; }
+    std::ptrdiff_t output_alias(const std::vector<shape>& shapes) const
+    {
+        return shapes.size() - 1;
+    }
 };
 
 } // namespace gpu

src/targets/gpu/leaky_relu.cpp

 #include <migraphx/gpu/leaky_relu.hpp>
-#include <migraphx/operators.hpp>
-#include <migraphx/manage_ptr.hpp>
-#include <migraphx/gpu/miopen.hpp>
-#include <utility>
+#include <migraphx/gpu/context.hpp>
 
 namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {

src/targets/gpu/logsoftmax.cpp

+#include <migraphx/gpu/logsoftmax.hpp>
+#include <migraphx/gpu/device/logsoftmax.hpp>
+#include <migraphx/op/logsoftmax.hpp>
+#include <migraphx/manage_ptr.hpp>
+#include <migraphx/gpu/miopen.hpp>
+#include <utility>
+
+namespace migraphx {
+inline namespace MIGRAPHX_INLINE_NS {
+namespace gpu {
+
+shape hip_logsoftmax::compute_shape(const std::vector<shape>& inputs) const
+{
+    check_shapes{inputs, *this}.has(2).standard();
+    return op.compute_shape({inputs.at(0)});
+}
+
+argument hip_logsoftmax::compute(context& ctx,
+                                 const shape& output_shape,
+                                 const std::vector<argument>& args) const
+{
+    return device::logsoftmax(ctx.get_stream().get(), output_shape, args, op.axis);
+}
+
+} // namespace gpu
+} // namespace MIGRAPHX_INLINE_NS
+} // namespace migraphx

src/targets/gpu/lowering.cpp

@@ -21,6 +21,7 @@
 #include <migraphx/gpu/leaky_relu.hpp>
 #include <migraphx/gpu/elu.hpp>
 #include <migraphx/gpu/softmax.hpp>
+#include <migraphx/gpu/logsoftmax.hpp>
 #include <migraphx/gpu/add.hpp>
 #include <migraphx/gpu/sub.hpp>
 #include <migraphx/gpu/exp.hpp>
@@ -44,6 +45,8 @@
 #include <migraphx/gpu/pad.hpp>
 #include <migraphx/gpu/gather.hpp>
 #include <migraphx/gpu/lrn.hpp>
+#include <migraphx/gpu/convert.hpp>
+#include <migraphx/gpu/clip.hpp>
 #include <utility>
 #include <functional>
 #include <algorithm>
@@ -97,8 +100,11 @@ struct miopen_apply
         add_extend_op<miopen_contiguous, op::contiguous>("contiguous");
         add_extend_op<hip_concat, op::concat>("concat");
         add_extend_op<miopen_softmax, op::softmax>("softmax");
+        add_extend_op<hip_logsoftmax, op::logsoftmax>("logsoftmax");
         add_extend_op<hip_gather, op::gather>("gather");
         add_extend_op<hip_pad, op::pad>("pad");
+        add_extend_op<hip_convert, op::convert>("convert");
+        add_extend_op<hip_clip, op::clip>("clip");
 
         add_lrn_op();
         add_convolution_op();
@@ -127,8 +133,7 @@ struct miopen_apply
         }
         else
         {
-            auto is     = prog->add_outline(s);
-            auto result = prog->insert_instruction(ins, hip_allocate{std::move(tag)}, is);
+            auto result = prog->insert_instruction(ins, hip_allocate{s, std::move(tag)});
             return result;
         }
     }

src/targets/gpu/lrn.cpp

 #include <migraphx/gpu/lrn.hpp>
-#include <migraphx/operators.hpp>
-#include <migraphx/manage_ptr.hpp>
-#include <migraphx/gpu/miopen.hpp>
-#include <utility>
+#include <migraphx/gpu/context.hpp>
 
 namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {

src/targets/gpu/pad.cpp

 #include <migraphx/gpu/pad.hpp>
-#include <migraphx/operators.hpp>
-#include <migraphx/manage_ptr.hpp>
-#include <migraphx/gpu/miopen.hpp>
+#include <migraphx/gpu/context.hpp>
 #include <migraphx/gpu/device/pad.hpp>
-#include <utility>
 
 namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {

src/targets/gpu/pooling.cpp

 #include <migraphx/gpu/pooling.hpp>
-#include <migraphx/operators.hpp>
-#include <migraphx/manage_ptr.hpp>
-#include <migraphx/gpu/miopen.hpp>
-#include <utility>
+#include <migraphx/gpu/context.hpp>
 
 namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {

src/targets/gpu/relu.cpp

 #include <migraphx/gpu/relu.hpp>
-#include <migraphx/operators.hpp>
-#include <migraphx/manage_ptr.hpp>
-#include <migraphx/gpu/miopen.hpp>
-#include <utility>
+#include <migraphx/gpu/context.hpp>
 
 namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {