Merge from develop

test/eliminate_contiguous_test.cpp

 #include <migraphx/eliminate_contiguous.hpp>
 #include <migraphx/dead_code_elimination.hpp>
+#include <migraphx/op/identity.hpp>
+#include <migraphx/op/dot.hpp>
+#include <migraphx/op/sin.hpp>
+#include <migraphx/op/slice.hpp>
 #include <migraphx/op/transpose.hpp>
 #include <migraphx/op/contiguous.hpp>
 #include <basic_ops.hpp>
@@ -36,7 +40,46 @@ TEST_CASE(non_standard_op)
     p.add_instruction(pass_op{}, c);
     auto count = std::distance(p.begin(), p.end());
     p.compile(eliminate_contiguous_target{});
+    EXPECT(std::distance(p.begin(), p.end()) == count);
+}
+
+TEST_CASE(transpose_gemm)
+{
+    migraphx::program p;
+    auto l  = p.add_literal(get_2x2());
+    auto t  = p.add_instruction(migraphx::op::transpose{{1, 0}}, l);
+    auto c  = p.add_instruction(migraphx::op::contiguous{}, t);
+    auto ic = p.add_instruction(migraphx::op::identity{}, c);
+    p.add_instruction(migraphx::op::dot{}, ic, l);
+    auto count = std::distance(p.begin(), p.end());
+    p.compile(eliminate_contiguous_target{});
     EXPECT(std::distance(p.begin(), p.end()) == (count - 1));
 }
 
+TEST_CASE(transpose_standard_op)
+{
+    migraphx::program p;
+    auto l  = p.add_literal(get_2x2());
+    auto t  = p.add_instruction(migraphx::op::transpose{{1, 0}}, l);
+    auto c  = p.add_instruction(migraphx::op::contiguous{}, t);
+    auto sn = p.add_instruction(migraphx::op::sin{}, c);
+    p.add_instruction(pass_standard_op{}, sn);
+    auto count = std::distance(p.begin(), p.end());
+    p.compile(eliminate_contiguous_target{});
+    EXPECT(std::distance(p.begin(), p.end()) == count);
+}
+
+TEST_CASE(no_packed_unary_op)
+{
+    migraphx::program p;
+    auto l  = p.add_literal(get_2x2());
+    auto t  = p.add_instruction(migraphx::op::slice{{1}, {1}, {2}}, l);
+    auto c  = p.add_instruction(migraphx::op::contiguous{}, t);
+    auto sn = p.add_instruction(migraphx::op::sin{}, c);
+    p.add_instruction(pass_standard_op{}, sn);
+    auto count = std::distance(p.begin(), p.end());
+    p.compile(eliminate_contiguous_target{});
+    EXPECT(std::distance(p.begin(), p.end()) == count - 1);
+}
+
 int main(int argc, const char* argv[]) { test::run(argc, argv); }

test/gpu/adjust_allocation.cpp

@@ -8,6 +8,7 @@
 #include <migraphx/iterator_for.hpp>
 #include <migraphx/op/add.hpp>
 #include <migraphx/op/transpose.hpp>
+#include <migraphx/op/contiguous.hpp>
 #include <migraphx/instruction.hpp>
 #include <migraphx/pass_manager.hpp>
 #include <migraphx/op/tanh.hpp>
@@ -37,7 +38,8 @@ TEST_CASE(tanh_shape)
         auto x   = p.add_parameter("x", s);
         auto tx  = p.add_instruction(migraphx::op::transpose{{1, 0}}, x);
         auto txh = p.add_instruction(migraphx::op::tanh{}, tx);
-        p.add_instruction(migraphx::op::add{}, txh, txh);
+        auto sum = p.add_instruction(migraphx::op::add{}, txh, txh);
+        p.add_instruction(migraphx::op::contiguous{}, sum);
 
         return p;
     };
@@ -55,8 +57,8 @@ TEST_CASE(tanh_shape)
     {
         if(ins->name() == "hip::allocate")
         {
-            migraphx::shape wrong_s{migraphx::shape::float_type, {3, 2}, {1, 3}};
-            ins->replace(wrong_s);
+            migraphx::shape new_s{migraphx::shape::float_type, {3, 2}, {1, 3}};
+            ins->replace(migraphx::gpu::hip_allocate{new_s});
         }
     }
     EXPECT(p1 != p2);

test/gpu/miopen.cpp

@@ -333,7 +333,22 @@ struct test_trans_tanh : verify_program<test_trans_tanh>
         auto x  = p.add_parameter("x", migraphx::shape{migraphx::shape::float_type, {4, 3, 3, 3}});
         auto tx = p.add_instruction(migraphx::op::transpose{{0, 1, 3, 2}}, x);
         auto tanhx = p.add_instruction(migraphx::op::tanh{}, tx);
-        p.add_instruction(migraphx::op::add{}, tanhx, tanhx);
+        auto r     = p.add_instruction(migraphx::op::add{}, tanhx, tanhx);
+        p.add_instruction(migraphx::op::contiguous{}, r);
+
+        return p;
+    }
+};
+
+struct test_slice_sin : verify_program<test_slice_sin>
+{
+    migraphx::program create_program() const
+    {
+        migraphx::program p;
+        auto l = p.add_parameter("x", migraphx::shape{migraphx::shape::float_type, {2, 2}});
+        auto t = p.add_instruction(migraphx::op::slice{{1}, {1}, {2}}, l);
+        p.add_instruction(migraphx::op::sin{}, t);
+
         return p;
     }
 };
@@ -692,8 +707,10 @@ struct test_trans_abs : verify_program<test_trans_abs>
         migraphx::program p;
         auto x  = p.add_parameter("x", migraphx::shape{migraphx::shape::float_type, {4, 3, 3, 3}});
         auto tx = p.add_instruction(migraphx::op::transpose{{0, 1, 3, 2}}, x);
-        auto tanhx = p.add_instruction(migraphx::op::abs{}, tx);
-        p.add_instruction(migraphx::op::add{}, tanhx, tanhx);
+        auto absx = p.add_instruction(migraphx::op::abs{}, tx);
+        auto r    = p.add_instruction(migraphx::op::add{}, absx, absx);
+        p.add_instruction(migraphx::op::contiguous{}, r);
+
         return p;
     }
 };
@@ -1293,6 +1310,17 @@ struct test_batchnorm_inference : verify_program<test_batchnorm_inference>
     }
 };
 
+struct test_clip : verify_program<test_clip>
+{
+    migraphx::program create_program() const
+    {
+        migraphx::program p;
+        auto x = p.add_parameter("x", migraphx::shape{migraphx::shape::float_type, {3}});
+        p.add_instruction(migraphx::op::clip{6.0, 0.0}, x);
+        return p;
+    }
+};
+
 struct test_conv_bn : verify_program<test_conv_bn>
 {
     migraphx::program create_program() const

test/memory_coloring_test.cpp

@@ -18,6 +18,13 @@ struct memory_coloring_target
 struct allocate
 {
     migraphx::shape s{};
+
+    template <class Self, class F>
+    static auto reflect(Self& self, F f)
+    {
+        return migraphx::pack(f(self.s, "shape"));
+    }
+
     std::string name() const { return "allocate"; }
     migraphx::shape compute_shape(const std::vector<migraphx::shape>& inputs) const
     {

test/onnx/onnx_test.cpp

@@ -699,8 +699,7 @@ TEST_CASE(add_scalar_test)
 {
     migraphx::program p;
     auto l0 = p.add_parameter("0", migraphx::shape{migraphx::shape::float_type, {2, 3, 4, 5}});
-    auto l1 =
-        p.add_literal(migraphx::literal{migraphx::shape{migraphx::shape::float_type, {1}}, {1}});
+    auto l1 = p.add_literal(migraphx::literal{migraphx::shape{migraphx::shape::float_type}, {1}});
     auto m0 = p.add_instruction(migraphx::op::multibroadcast{{2, 3, 4, 5}}, l0);
     auto m1 = p.add_instruction(migraphx::op::multibroadcast{{2, 3, 4, 5}}, l1);
     p.add_instruction(migraphx::op::add{}, m0, m1);
@@ -795,4 +794,14 @@ TEST_CASE(no_pad_test)
     EXPECT(p == prog);
 }
 
+TEST_CASE(clip_test)
+{
+    migraphx::program p;
+    auto l0 = p.add_parameter("0", migraphx::shape{migraphx::shape::float_type, {3}});
+    p.add_instruction(migraphx::op::clip{6.0, 0.0}, l0);
+    auto prog = migraphx::parse_onnx("clip_test.onnx");
+
+    EXPECT(p == prog);
+}
+
 int main(int argc, const char* argv[]) { test::run(argc, argv); }

test/print_graph_test.cpp

@@ -29,12 +29,12 @@ TEST_CASE(basic_graph_test)
     EXPECT(migraphx::contains(test, "\"@0\"[label=\"@literal\"]"));
     EXPECT(migraphx::contains(test, "\"y\"[label=\"@param:y\"]"));
     EXPECT(migraphx::contains(test, "\"x\"[label=\"@param:x\"]"));
-    EXPECT(migraphx::contains(test, "\"@3\"[label=\"sum\"]"));
-    EXPECT(migraphx::contains(test, "\"@4\"[label=\"sum\"]"));
-    EXPECT(migraphx::contains(test, "\"x\" -> \"@3\""));
-    EXPECT(migraphx::contains(test, "\"y\" -> \"@3\""));
-    EXPECT(migraphx::contains(test, "\"@3\" -> \"@4\""));
-    EXPECT(migraphx::contains(test, "\"@0\" -> \"@4\""));
+    EXPECT(migraphx::contains(test, "\"@1\"[label=\"sum\"]"));
+    EXPECT(migraphx::contains(test, "\"@2\"[label=\"sum\"]"));
+    EXPECT(migraphx::contains(test, "\"x\" -> \"@1\""));
+    EXPECT(migraphx::contains(test, "\"y\" -> \"@1\""));
+    EXPECT(migraphx::contains(test, "\"@1\" -> \"@2\""));
+    EXPECT(migraphx::contains(test, "\"@0\" -> \"@2\""));
     EXPECT(migraphx::contains(test, "[label=\"int64_type, {1}, {0}\"]"));
 }
 

test/program_test.cpp

@@ -2,6 +2,10 @@
 #include <migraphx/program.hpp>
 #include <migraphx/iterator_for.hpp>
 #include <migraphx/instruction.hpp>
+#include <migraphx/op/add.hpp>
+#include <migraphx/op/dot.hpp>
+#include <migraphx/op/mul.hpp>
+#include <migraphx/cpu/target.hpp>
 #include <sstream>
 #include "test.hpp"
 #include <basic_ops.hpp>
@@ -27,4 +31,78 @@ TEST_CASE(program_equality)
     EXPECT(x == y);
 }
 
+TEST_CASE(program_copy)
+{
+    auto create_program_1 = [] {
+        migraphx::program p;
+        migraphx::shape s{migraphx::shape::float_type, {3, 4, 5}};
+        std::vector<float> data(3 * 4 * 5);
+        std::iota(data.begin(), data.end(), 1.0f);
+        auto l2  = p.add_literal(migraphx::literal(s, data));
+        auto p1  = p.add_parameter("x", s);
+        auto po  = p.add_outline(s);
+        auto sum = p.add_instruction(migraphx::op::add{}, l2, po);
+        p.add_instruction(migraphx::op::mul{}, sum, p1);
+
+        return p;
+    };
+
+    {
+        auto p1 = create_program_1();
+        migraphx::program p2{};
+        p2 = p1;
+
+        p2.compile(migraphx::cpu::target{});
+        EXPECT(p1 != p2);
+
+        p1.compile(migraphx::cpu::target{});
+        EXPECT(p1 == p2);
+    }
+
+    {
+        auto p1 = create_program_1();
+        auto p2(p1);
+        EXPECT(p1 == p2);
+
+        p1.compile(migraphx::cpu::target{});
+        EXPECT(p1 != p2);
+
+        p2 = p1;
+        EXPECT(p1 == p2);
+    }
+
+    {
+        auto p1 = create_program_1();
+        auto p2 = create_program();
+        EXPECT(p1 != p2);
+
+        p2 = p1;
+        EXPECT(p1 == p2);
+
+        p1.compile(migraphx::cpu::target{});
+        p2.compile(migraphx::cpu::target{});
+
+        EXPECT(p1 == p2);
+    }
+
+    {
+        migraphx::program p1;
+        migraphx::shape s1{migraphx::shape::float_type, {2, 3}};
+        migraphx::shape s2{migraphx::shape::float_type, {3, 6}};
+        migraphx::shape s3{migraphx::shape::float_type, {2, 6}};
+        auto para1 = p1.add_parameter("m1", s1);
+        auto para2 = p1.add_parameter("m2", s2);
+        auto para3 = p1.add_parameter("m3", s3);
+        p1.add_instruction(migraphx::op::dot{0.31f, 0.28f}, para1, para2, para3);
+
+        migraphx::program p2{};
+        p2 = p1;
+        EXPECT(p2 == p1);
+
+        p1.compile(migraphx::cpu::target{});
+        p2.compile(migraphx::cpu::target{});
+        EXPECT(p2 == p1);
+    }
+}
+
 int main(int argc, const char* argv[]) { test::run(argc, argv); }

test/tf/relu6_test.pb

+
+:
+
0Placeholder*
+dtype0
*
+shape:

+
+relu6Relu6
0*
+
T0
"
\ No newline at end of file

test/tf/tf_test.cpp

@@ -80,7 +80,7 @@ TEST_CASE(concat_test)
     int axis = 1;
     // tf uses axis as the third input, and it is in int32 format
     // add the literal using a vector in order to set stride to 1 (like in tf parser)
-    p.add_literal(migraphx::shape{migraphx::shape::int32_type, {1}}, std::vector<int>{axis});
+    p.add_literal(migraphx::shape{migraphx::shape::int32_type}, std::vector<int>{axis});
 
     p.add_instruction(migraphx::op::concat{static_cast<std::size_t>(axis)}, l0, l1);
     auto prog = migraphx::parse_tf("concat_test.pb", false);
@@ -91,7 +91,7 @@ TEST_CASE(concat_test)
 TEST_CASE(const_test)
 {
     migraphx::program p;
-    p.add_literal(migraphx::shape{migraphx::shape::float_type, {1}}, std::vector<float>{1.0f});
+    p.add_literal(migraphx::shape{migraphx::shape::float_type}, std::vector<float>{1.0f});
     auto prog = migraphx::parse_tf("constant_test.pb", false);
 
     EXPECT(p == prog);
@@ -119,6 +119,30 @@ TEST_CASE(conv_test)
     EXPECT(p == prog);
 }
 
+TEST_CASE(depthwiseconv_test)
+{
+    migraphx::program p;
+
+    auto l0 = p.add_parameter("0", migraphx::shape{migraphx::shape::float_type, {1, 3, 16, 16}});
+    std::vector<float> weight_data(3 * 3 * 3 * 1);
+    std::fill(weight_data.begin(), weight_data.end(), 1.0f);
+    auto l1 =
+        p.add_literal(migraphx::shape{migraphx::shape::float_type, {3, 3, 3, 1}}, weight_data);
+
+    migraphx::op::convolution op;
+    op.padding_mode = migraphx::op::padding_mode_t::same;
+    op.stride       = {1, 1};
+    op.dilation     = {1, 1};
+    op.group        = 3;
+    auto l2         = p.add_instruction(migraphx::op::transpose{{0, 3, 1, 2}}, l1);
+    auto l3         = p.add_instruction(migraphx::op::transpose{{1, 3, 0, 2}}, l2);
+    auto l4         = p.add_instruction(migraphx::op::reshape{{3, 1, 3, 3}}, l3);
+    p.add_instruction(op, l0, l4);
+    auto prog = migraphx::parse_tf("depthwise_conv_test.pb", true);
+
+    EXPECT(p == prog);
+}
+
 TEST_CASE(identity_test)
 {
     migraphx::program p;
@@ -229,6 +253,16 @@ TEST_CASE(relu_test)
     EXPECT(p == prog);
 }
 
+TEST_CASE(relu6_test)
+{
+    migraphx::program p;
+    auto l0 = p.add_parameter("0", migraphx::shape{migraphx::shape::float_type, {1, 3, 16, 16}});
+    p.add_instruction(migraphx::op::clip{6.0, 0.0}, l0);
+    auto prog = migraphx::parse_tf("relu6_test.pb", false);
+
+    EXPECT(p == prog);
+}
+
 TEST_CASE(reshape_test)
 {
     migraphx::program p;

tools/include/operation.hpp

@@ -49,7 +49,7 @@ struct operation
     argument compute(context& ctx, const shape& output, const std::vector<argument>& input) const;
     /// An optional method to return which argument the output will alias. If
     /// there is no aliased output then -1 can be returned.
-    int output_alias(const std::vector<shape>& input) const;
+    std::ptrdiff_t output_alias(const std::vector<shape>& input) const;
     /// An optional stream operator to print the operation. When this is not
     /// implemented, it will just print the operation's name.
     friend std::ostream& operator<<(std::ostream& os, const operation& op);
@@ -69,7 +69,7 @@ auto operator<<(std::ostream& os, const T& x) -> decltype(os << x.name())
 {
     os << x.name();
     char delim = '[';
-    reflect_each(x, [&](auto& y, auto name) {
+    reflect_each(x, [&](auto&& y, auto name) {
         os << delim;
         os << name << "=";
         stream_write_value(os, y);
@@ -87,6 +87,8 @@ namespace operation_equal {
 template <class T, class U>
 auto operator==(const T& x, const U& y) -> decltype(x.name() == y.name())
 {
+    static_assert(is_reflectable<T>{} or sizeof(T) <= 1,
+                  "Missing equality operator or reflect method.");
     if(x.name() != y.name())
         return false;
     const auto& yy = any_cast<T>(y);
@@ -175,7 +177,7 @@ auto is_context_free_op(const T& x) -> decltype(is_context_free_op(
 }
 
 template <class T>
-int output_alias_op(rank<0>, const T&, const std::vector<shape>&)
+std::ptrdiff_t output_alias_op(rank<0>, const T&, const std::vector<shape>&)
 {
     return -1;
 }
@@ -188,7 +190,7 @@ auto output_alias_op(rank<1>, const T& x, const std::vector<shape>& shapes)
 }
 
 template <class T>
-int output_alias_op(const T& x, const std::vector<shape>& shapes)
+std::ptrdiff_t output_alias_op(const T& x, const std::vector<shape>& shapes)
 {
     return output_alias_op(rank<1>{}, x, shapes);
 }
@@ -238,7 +240,7 @@ auto has_finalize_op(const T&) -> decltype(has_finalize_op(rank<1>{},
      virtual('is_context_free', returns = 'bool', const = True, default = 'is_context_free_op'),
      virtual('has_finalize', returns = 'bool', const = True, default = 'has_finalize_op'),
      virtual('output_alias',
-             returns = 'int',
+             returns = 'std::ptrdiff_t',
              input   = 'const std::vector<shape>&',
              const   = True,
              default = 'output_alias_op'),