separte this pull request into ajust_gpu_allocation and eliminate_contiguous processing.

src/include/migraphx/op/unary.hpp

@@ -21,7 +21,7 @@ struct unary
     shape compute_shape(std::vector<shape> inputs) const
     {
         check_shapes{inputs}.has(1);
-        return {inputs.at(0).type(), inputs.at(0).lens()};
+        return inputs.at(0);
     }
 };
 

src/targets/cpu/lowering.cpp

@@ -603,16 +603,7 @@ struct cpu_unary
         argument result{output_shape};
         result.visit([&](auto output) {
             args[0].visit([&](auto input) {
-                if(input.get_shape().packed())
-                {
                 std::transform(input.begin(), input.end(), output.begin(), op.fcn());
-                }
-                else
-                {
-                    shape_for_each(output.get_shape(), [&](const auto& idx) {
-                        output(idx.begin(), idx.end()) = op.fcn()(input(idx.begin(), idx.end()));
-                    });
-                }
             });
         });
 
@@ -788,9 +779,9 @@ struct cpu_binary
     std::string name() const { return op.name(); }
     shape compute_shape(const std::vector<shape>& inputs) const 
     { 
-        // operator will generate standard output shape
-        return {inputs.front().type(), inputs.front().lens()};
+        return { inputs.front().type(), inputs.front().lens()}; 
     }
+
     argument compute(context&, const shape& output_shape, std::vector<argument> args) const
     {
         argument result{output_shape};

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

@@ -45,7 +45,7 @@ struct unary_device : oper<Derived>
     shape compute_shape(const std::vector<shape>& inputs) const
     {
         check_shapes{inputs, *this}.has(2);
-        return {inputs.at(0).type(), inputs.at(0).lens()};
+        return inputs.at(1);
     }
 
     argument compute(context& ctx, const shape&, const std::vector<argument>& args) const
@@ -63,7 +63,7 @@ struct binary_device : oper<Derived>
     shape compute_shape(const std::vector<shape>& inputs) const
     {
         check_shapes{inputs, *this}.has(3);
-        return {inputs.at(0).type(), inputs.at(0).lens()};
+        return inputs.at(2);
     }
 
     argument compute(context& ctx, const shape&, const std::vector<argument>& args) const

src/targets/gpu/lrn.cpp

@@ -7,7 +7,7 @@ namespace gpu {
 
 shape miopen_lrn::compute_shape(const std::vector<shape>& inputs) const
 {
-    check_shapes{inputs, *this}.has(2).standard();
+    check_shapes{inputs, *this}.has(2).not_broadcasted();
     return inputs.at(1);
 }
 

src/targets/gpu/relu.cpp

@@ -7,7 +7,7 @@ namespace gpu {
 
 shape miopen_relu::compute_shape(const std::vector<shape>& inputs) const
 {
-    check_shapes{inputs, *this}.has(2).not_broadcasted().not_transposed();
+    check_shapes{inputs, *this}.has(2).not_broadcasted();
     return inputs.at(1);
 }
 

src/targets/gpu/tanh.cpp

@@ -7,7 +7,7 @@ namespace gpu {
 
 shape miopen_tanh::compute_shape(const std::vector<shape>& inputs) const
 {
-    check_shapes{inputs, *this}.has(2).not_broadcasted().not_transposed();
+    check_shapes{inputs, *this}.has(2).not_broadcasted();
     return inputs.at(1);
 }