merge changes from gpu_div branch

src/propagate_constant.cpp

@@ -10,8 +10,8 @@ inline namespace MIGRAPHX_INLINE_NS {
 
 bool skip_propogate(instruction_ref ins)
 {
-    if(ins->name() == "@literal")
-        return true;
+    if(ins->name() == "contiguous")
+        return skip_propogate(ins->inputs().front());
     auto&& s = ins->get_shape();
     if(s.broadcasted() and not s.scalar())
         return true;
@@ -33,7 +33,7 @@ void propagate_constant::apply(program& p) const
                                                          ins->outputs().end());
             for(auto child : children)
             {
-                if(skip_propogate(child))
+                if(child->name() == "@literal" or skip_propogate(child))
                 {
                     self(child);
                     continue;

src/targets/gpu/device/div.cpp

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

src/targets/gpu/device/pow.cpp

@@ -9,7 +9,7 @@ namespace device {
 void pow(hipStream_t stream, const argument& result, const argument& arg1, const argument& arg2)
 {
     nary(stream, result, arg1, arg2)(
-        [](auto e, auto b) { return ::pow(to_hip_type(b), to_hip_type(e)); });
+        [](auto b, auto e) { return ::pow(to_hip_type(b), to_hip_type(e)); });
 }
 
 } // namespace device

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/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];
     }