Merge branch 'develop' into jenkins_reorder

test/gpu/manage_host_buffer.cpp

@@ -53,7 +53,6 @@ TEST_CASE(host_same_buffer_copy)
     migraphx::parameter_map pp;
     std::vector<float> a_vec(ss.elements(), -1);
     std::vector<float> b_vec(ss.elements(), 2);
-    std::vector<float> c_vec(ss.elements(), 0);
     pp["a"] = migraphx::argument(ss, a_vec.data());
     pp["b"] = migraphx::argument(ss, b_vec.data());
     std::vector<float> gpu_result;
@@ -64,7 +63,8 @@ TEST_CASE(host_same_buffer_copy)
     auto result = p.eval(pp).back();
     std::vector<float> results_vector(ss.elements(), -1);
     result.visit([&](auto output) { results_vector.assign(output.begin(), output.end()); });
-    EXPECT(migraphx::verify::verify_range(c_vec, results_vector));
+    std::vector<float> gold_vec(ss.elements(), 0);
+    EXPECT(migraphx::verify::verify_rms_range(results_vector, gold_vec));
 }
 
 TEST_CASE(arguments_lifetime)

test/gpu/mlir.cpp

@@ -133,7 +133,8 @@ bool verify_mlir(const migraphx::module& mmlir)
     auto inputs = generate_params(ref);
 
     auto mlir = create_program_from_mlir(mmlir);
-    return migraphx::verify_args("mlir", run_ref(ref, inputs), run_gpu(mlir, inputs));
+    return migraphx::verify_args_with_tolerance(
+        "mlir", run_gpu(mlir, inputs), migraphx::verify::expected{run_ref(ref, inputs)});
 }
 
 TEST_CASE(conv)

test/gpu/quantization.cpp

@@ -40,7 +40,6 @@
 TEST_CASE(gpu_target_copy)
 {
     migraphx::target gpu_t = migraphx::make_target("gpu");
-    migraphx::target ref_t = migraphx::make_target("ref");
     migraphx::shape s{migraphx::shape::int8_type, {2, 3, 4, 5}};
 
     auto ref_arg_orig  = migraphx::generate_argument(s, 0x123456L);
@@ -52,7 +51,7 @@ TEST_CASE(gpu_target_copy)
     std::vector<int8_t> val_final;
     ref_arg_final.visit([&](auto v) { val_final.assign(v.begin(), v.end()); });
 
-    EXPECT(migraphx::verify::verify_range(val_orig, val_final));
+    EXPECT(migraphx::verify::verify_rms_range(val_orig, val_final));
 }
 
 TEST_CASE(int8_quantization)
@@ -118,9 +117,12 @@ TEST_CASE(int8_quantization)
         // the regular pipeline uses the rewrite_quantization in the much
         // earlier stage.
         if(migraphx::gpu::mlir_enabled())
-            EXPECT(migraphx::verify::verify_range(ref_result, gpu_result, 1e5));
+            EXPECT(migraphx::verify::verify_range_with_tolerance(
+                gpu_result,
+                migraphx::verify::expected{ref_result},
+                migraphx::verify::tolerance{0.01}));
         else
-            EXPECT(migraphx::verify::verify_range(ref_result, gpu_result));
+            EXPECT(migraphx::verify::verify_rms_range(gpu_result, ref_result));
     }
 }
 

test/include/pointwise.hpp

@@ -24,16 +24,16 @@
 #ifndef MIGRAPHX_GUARD_TEST_INCLUDE_POINTWISE_HPP
 #define MIGRAPHX_GUARD_TEST_INCLUDE_POINTWISE_HPP
 
+#include <migraphx/instruction_ref.hpp>
 #include <migraphx/program.hpp>
 #include <migraphx/module.hpp>
 #include <migraphx/make_op.hpp>
 
 template <class F>
-migraphx::instruction_ref add_pointwise(migraphx::program& p,
-                                        migraphx::module_ref mm,
-                                        const std::string& name,
-                                        std::vector<migraphx::instruction_ref> inputs,
-                                        F f)
+migraphx::module_ref create_pointwise_module(migraphx::program& p,
+                                             const std::string& name,
+                                             std::vector<migraphx::instruction_ref> inputs,
+                                             F f)
 {
     auto* pm = p.create_module(name);
     pm->set_bypass();
@@ -44,6 +44,17 @@ migraphx::instruction_ref add_pointwise(migraphx::program& p,
     });
     auto r = f(pm, params);
     pm->add_return({r});
+    return pm;
+}
+
+template <class F>
+migraphx::instruction_ref add_pointwise(migraphx::program& p,
+                                        migraphx::module_ref mm,
+                                        const std::string& name,
+                                        std::vector<migraphx::instruction_ref> inputs,
+                                        F f)
+{
+    auto* pm = create_pointwise_module(p, name, inputs, f);
     return mm->add_instruction(migraphx::make_op("pointwise"), inputs, {pm});
 }
 

test/onnx/const_of_shape_default_test.onnx

+const_of_shape_default_test:

+6shape"Constant*#
+value*:Bshape_tensor

+
+shape
y"ConstantOfShapeconst_of_shape_default_testb
+
y
+
+
+
+B
\ No newline at end of file

test/onnx/const_of_shape_dyn_int64_test.onnx

+const_of_shape_dyn_int64_test:

+=
+output_dims
y"ConstantOfShape*
+value*
:

+Bvalue
const_of_shape_dyn_int64_testZ
+output_dims
+
+
+b
+
y
+
+
+
+B
\ No newline at end of file

test/onnx/const_of_shape_int64_test.onnx

-constant-of-shape:

+const_of_shape_int64_test:

 6shape"Constant*#
-value**Bshape_tensor 

+value*:Bshape_tensor

 7
 shape
y"ConstantOfShape*
 value*
:

-Bvalue 
constant_of_shapeb
+Bvalue
const_of_shape_int64_testb
 
y
-

+
 
 
-B
+B
\ No newline at end of file

test/onnx/const_of_shape_no_value_attr_test.onnx

-constant-of-shape:

+!const_of_shape_no_value_attr_test:

 6shape"Constant*#
-value**Bshape_tensor 

+value*:Bshape_tensor

 
-shape
y"ConstantOfShapeconstant_of_shapeb
+shape
y"ConstantOfShape!const_of_shape_no_value_attr_testb
 
y
 

 
 
-B
+B
\ No newline at end of file

test/onnx/gen_onnx.py

@@ -1007,9 +1007,9 @@ def const_of_shape_empty_input_test():
                                          [10])
     empty_val = np.array([]).astype(np.int64)
     empty_ts = helper.make_tensor(name='empty_tensor',
-                                  data_type=TensorProto.INT32,
+                                  data_type=TensorProto.INT64,
                                   dims=empty_val.shape,
-                                  vals=empty_val.flatten().astype(int))
+                                  vals=empty_val.flatten().astype(np.int64))
     shape_const = helper.make_node(
         'Constant',
         inputs=[],
@@ -1035,9 +1035,9 @@ def const_of_shape_float_test():
 
     shape_val = np.array([2, 3, 4]).astype(np.int64)
     shape_ts = helper.make_tensor(name='shape_tensor',
-                                  data_type=TensorProto.INT32,
+                                  data_type=TensorProto.INT64,
                                   dims=shape_val.shape,
-                                  vals=shape_val.flatten().astype(int))
+                                  vals=shape_val.flatten().astype(np.int64))
 
     shape_const = helper.make_node(
         'Constant',
@@ -1055,22 +1055,44 @@ def const_of_shape_float_test():
     return ([shape_const, node], [], [y])
 
 
+@onnx_test()
+def const_of_shape_default_test():
+    shape_val = np.array([2, 3, 4]).astype(np.int64)
+    shape_ts = helper.make_tensor(name='shape_tensor',
+                                  data_type=TensorProto.INT64,
+                                  dims=shape_val.shape,
+                                  vals=shape_val.flatten().astype(np.int64))
+    shape_const = helper.make_node(
+        'Constant',
+        inputs=[],
+        outputs=['shape'],
+        value=shape_ts,
+    )
+    y = helper.make_tensor_value_info('y', TensorProto.INT64, [2, 3, 4])
+
+    node = onnx.helper.make_node('ConstantOfShape',
+                                 inputs=['shape'],
+                                 outputs=['y'])
+
+    return ([shape_const, node], [], [y])
+
+
 @onnx_test()
 def const_of_shape_int64_test():
     tensor_val = onnx.helper.make_tensor('value', onnx.TensorProto.INT64, [1],
                                          [10])
     shape_val = np.array([2, 3, 4]).astype(np.int64)
     shape_ts = helper.make_tensor(name='shape_tensor',
-                                  data_type=TensorProto.INT32,
+                                  data_type=TensorProto.INT64,
                                   dims=shape_val.shape,
-                                  vals=shape_val.flatten().astype(int))
+                                  vals=shape_val.flatten().astype(np.int64))
     shape_const = helper.make_node(
         'Constant',
         inputs=[],
         outputs=['shape'],
         value=shape_ts,
     )
-    y = helper.make_tensor_value_info('y', TensorProto.FLOAT, [2, 3, 4])
+    y = helper.make_tensor_value_info('y', TensorProto.INT64, [2, 3, 4])
 
     node = onnx.helper.make_node('ConstantOfShape',
                                  inputs=['shape'],
@@ -1084,9 +1106,9 @@ def const_of_shape_int64_test():
 def const_of_shape_no_value_attr_test():
     shape_val = np.array([2, 3, 4]).astype(np.int64)
     shape_ts = helper.make_tensor(name='shape_tensor',
-                                  data_type=TensorProto.INT32,
+                                  data_type=TensorProto.INT64,
                                   dims=shape_val.shape,
-                                  vals=shape_val.flatten().astype(int))
+                                  vals=shape_val.flatten().astype(np.int64))
     shape_const = helper.make_node(
         'Constant',
         inputs=[],
@@ -1104,6 +1126,40 @@ def const_of_shape_no_value_attr_test():
     return ([shape_const, node], [], [y])
 
 
+@onnx_test()
+def const_of_shape_dyn_float_test():
+    tensor_val = onnx.helper.make_tensor('value', onnx.TensorProto.FLOAT, [1],
+                                         [10])
+
+    output_dims = helper.make_tensor_value_info('output_dims',
+                                                TensorProto.INT64, [3])
+    y = helper.make_tensor_value_info('y', TensorProto.FLOAT, [2, 3, 4])
+
+    node = onnx.helper.make_node('ConstantOfShape',
+                                 inputs=['output_dims'],
+                                 outputs=['y'],
+                                 value=tensor_val)
+
+    return ([node], [output_dims], [y])
+
+
+@onnx_test()
+def const_of_shape_dyn_int64_test():
+    tensor_val = onnx.helper.make_tensor('value', onnx.TensorProto.INT64, [1],
+                                         [10])
+
+    output_dims = helper.make_tensor_value_info('output_dims',
+                                                TensorProto.INT64, [3])
+    y = helper.make_tensor_value_info('y', TensorProto.INT64, [2, 3, 4])
+
+    node = onnx.helper.make_node('ConstantOfShape',
+                                 inputs=['output_dims'],
+                                 outputs=['y'],
+                                 value=tensor_val)
+
+    return ([node], [output_dims], [y])
+
+
 @onnx_test()
 def conv_1d_test():
     x = helper.make_tensor_value_info('0', TensorProto.FLOAT, [1, 3, 5])

test/onnx/onnx_test.cpp

@@ -1040,11 +1040,25 @@ TEST_CASE(constant_one_val_int64_test)
     EXPECT(p == prog);
 }
 
+TEST_CASE(const_of_shape_default_test)
+{
+    migraphx::program p;
+    auto* mm = p.get_main_module();
+    migraphx::shape output_dims_shape(migraphx::shape::int64_type, {3});
+    mm->add_literal(migraphx::literal(output_dims_shape, {2, 3, 4}));
+    migraphx::shape output_shape{migraphx::shape::float_type, {2, 3, 4}};
+    std::vector<float> vec(output_shape.elements(), 0.0);
+    mm->add_literal(migraphx::literal(output_shape, vec));
+
+    auto prog = optimize_onnx("const_of_shape_default_test.onnx");
+    EXPECT(p == prog);
+}
+
 TEST_CASE(const_of_shape_empty_input_test)
 {
     migraphx::program p;
     auto* mm = p.get_main_module();
-    mm->add_literal(migraphx::literal(migraphx::shape::int32_type));
+    mm->add_literal(migraphx::literal(migraphx::shape::int64_type));
     migraphx::shape s(migraphx::shape::int64_type, {1}, {0});
     std::vector<int64_t> vec(s.elements(), 10);
     mm->add_literal(migraphx::literal(s, vec));
@@ -1057,7 +1071,7 @@ TEST_CASE(const_of_shape_float_test)
 {
     migraphx::program p;
     auto* mm = p.get_main_module();
-    migraphx::shape ss(migraphx::shape::int32_type, {3});
+    migraphx::shape ss(migraphx::shape::int64_type, {3});
     mm->add_literal(migraphx::literal(ss, {2, 3, 4}));
     migraphx::shape s(migraphx::shape::float_type, {2, 3, 4});
     std::vector<float> vec(s.elements(), 10.0f);
@@ -1071,8 +1085,10 @@ TEST_CASE(const_of_shape_int64_test)
 {
     migraphx::program p;
     auto* mm = p.get_main_module();
-    migraphx::shape ss(migraphx::shape::int32_type, {3});
+    // output_dims
+    migraphx::shape ss(migraphx::shape::int64_type, {3});
     mm->add_literal(migraphx::literal(ss, {2, 3, 4}));
+    // constant shape literal
     migraphx::shape s(migraphx::shape::int64_type, {2, 3, 4});
     std::vector<int64_t> vec(s.elements(), 10);
     mm->add_literal(migraphx::literal(s, vec));
@@ -1085,7 +1101,7 @@ TEST_CASE(const_of_shape_no_value_attr_test)
 {
     migraphx::program p;
     auto* mm = p.get_main_module();
-    migraphx::shape ss(migraphx::shape::int32_type, {3});
+    migraphx::shape ss(migraphx::shape::int64_type, {3});
     mm->add_literal(migraphx::literal(ss, {2, 3, 4}));
     migraphx::shape s(migraphx::shape::float_type, {2, 3, 4});
     std::vector<float> vec(s.elements(), 0.0f);
@@ -1095,6 +1111,42 @@ TEST_CASE(const_of_shape_no_value_attr_test)
     EXPECT(p == prog);
 }
 
+TEST_CASE(const_of_shape_dyn_float_test)
+{
+    migraphx::program p;
+    auto* mm = p.get_main_module();
+    auto od_param =
+        mm->add_parameter("output_dims", migraphx::shape{migraphx::shape::int64_type, {3}});
+    auto alloc_ins = mm->add_instruction(
+        migraphx::make_op("allocate", {{"buf_type", migraphx::shape::float_type}}), od_param);
+    migraphx::shape dv_shape(migraphx::shape::float_type, {1}, {0});
+    auto dv_lit   = mm->add_literal(migraphx::literal(dv_shape, {10}));
+    auto fill_ins = mm->add_instruction(migraphx::make_op("fill"), dv_lit, alloc_ins);
+    mm->add_return({fill_ins});
+
+    migraphx::onnx_options options;
+    auto prog = parse_onnx("const_of_shape_dyn_float_test.onnx", options);
+    EXPECT(p == prog);
+}
+
+TEST_CASE(const_of_shape_dyn_int64_test)
+{
+    migraphx::program p;
+    auto* mm = p.get_main_module();
+    auto od_param =
+        mm->add_parameter("output_dims", migraphx::shape{migraphx::shape::int64_type, {3}});
+    auto alloc_ins = mm->add_instruction(
+        migraphx::make_op("allocate", {{"buf_type", migraphx::shape::int64_type}}), od_param);
+    migraphx::shape dv_shape(migraphx::shape::int64_type, {1}, {0});
+    auto dv_lit   = mm->add_literal(migraphx::literal(dv_shape, {10}));
+    auto fill_ins = mm->add_instruction(migraphx::make_op("fill"), dv_lit, alloc_ins);
+    mm->add_return({fill_ins});
+
+    migraphx::onnx_options options;
+    auto prog = parse_onnx("const_of_shape_dyn_int64_test.onnx", options);
+    EXPECT(p == prog);
+}
+
 TEST_CASE(conv_autopad_fail_test)
 {
     EXPECT(test::throws([&] { optimize_onnx("conv_autopad_fail_test.onnx"); }));

test/op_shape_test.cpp

@@ -88,6 +88,13 @@ TEST_CASE(allocate_static)
     expect_shape(out_shape, migraphx::make_op("allocate", {{"shape", to_value(out_shape)}}));
 }
 
+TEST_CASE(allocate_static_input_error)
+{
+    migraphx::shape input{migraphx::shape::int64_type, {3}};
+    migraphx::shape out_shape{migraphx::shape::float_type, {2, 3, 4}};
+    expect_shape(out_shape, migraphx::make_op("allocate", {{"shape", to_value(out_shape)}}), input);
+}
+
 TEST_CASE(allocate_dyn)
 {
     migraphx::shape input{migraphx::shape::int64_type, {2}};
@@ -109,6 +116,14 @@ TEST_CASE(allocate_dyn_with_shape_attr)
                  input);
 }
 
+TEST_CASE(allocate_dyn_no_input_error)
+{
+    migraphx::shape shape_attr{migraphx::shape::float_type,
+                               {{1, 4}, {3, 3}, {4, 8, {4, 6}}, {4, 8}, {4, 6}}};
+    expect_shape(shape_attr,
+                 migraphx::make_op("allocate", {{"shape", migraphx::to_value(shape_attr)}}));
+}
+
 TEST_CASE(argmax_axis0)
 {
     migraphx::shape input{migraphx::shape::half_type, {2, 3, 4, 5}};
@@ -2524,13 +2539,21 @@ TEST_CASE(reshape_shape)
         migraphx::shape output{migraphx::shape::float_type, lens};
         expect_shape(output, migraphx::make_op("reshape", {{"dims", new_shape}}), input);
     }
+}
 
+TEST_CASE(reshape_shape_invalid)
+{
+    migraphx::shape input{migraphx::shape::float_type, {24, 1, 1, 1}};
     for(auto&& new_shape :
         std::vector<std::vector<int64_t>>{{8, 3, 2, 2}, {1, 3, -1, -1}, {3, 0}, {3, 2}})
     {
         throws_shape(migraphx::make_op("reshape", {{"dims", new_shape}}), input);
     }
+}
 
+TEST_CASE(reshape_shape_minus1_reshapes)
+{
+    migraphx::shape input{migraphx::shape::float_type, {24, 1, 1, 1}};
     std::vector<std::pair<std::vector<int64_t>, migraphx::shape>> minus1_tests{
         {{2, -1, 3}, {migraphx::shape::float_type, {2, 4, 3}}},
         {{0, -1, 0}, {migraphx::shape::float_type, {24, 1, 1}}},
@@ -2654,11 +2677,11 @@ TEST_CASE(reshape_broadcast_squeeze)
     expect_shape(output, migraphx::make_op("reshape", {{"dims", output.lens()}}), input);
 }
 
-TEST_CASE(reshape_broadcast_squeeze_error)
+TEST_CASE(reshape_broadcast_squeeze_memlayout_change)
 {
     migraphx::shape input{migraphx::shape::float_type, {2, 16, 16, 1280}, {0, 0, 0, 1}};
-    std::vector<int64_t> new_shape = {2, 16, 20480};
-    throws_shape(migraphx::make_op("reshape", {{"dims", new_shape}}), input);
+    migraphx::shape output{migraphx::shape::float_type, {2, 16, 256, 80}, {0, 0, 0, 16}};
+    expect_shape(output, migraphx::make_op("reshape", {{"dims", output.lens()}}), input);
 }
 
 TEST_CASE(reshape_dyn_shape)
@@ -2706,6 +2729,199 @@ TEST_CASE(reshape_non_fixed_not_matching_error)
     throws_shape(migraphx::make_op("reshape", {{"dims", new_shape}}), input);
 }
 
+TEST_CASE(reshape_lazy_shape)
+{
+    migraphx::shape input{migraphx::shape::float_type, {24, 1, 1, 1}};
+    for(auto&& new_shape :
+        std::vector<std::vector<int64_t>>{{8, 3, 1, 1}, {1, 3, 4, 2}, {1, 3, 4, 2}})
+    {
+        std::vector<std::size_t> lens(new_shape.size());
+        std::copy(new_shape.begin(), new_shape.end(), lens.begin());
+        migraphx::shape output{migraphx::shape::float_type, lens};
+        expect_shape(output, migraphx::make_op("reshape_lazy", {{"dims", new_shape}}), input);
+    }
+
+    for(auto&& new_shape :
+        std::vector<std::vector<int64_t>>{{8, 3, 2, 2}, {1, 3, -1, -1}, {3, 0}, {3, 2}})
+    {
+        throws_shape(migraphx::make_op("reshape_lazy", {{"dims", new_shape}}), input);
+    }
+
+    std::vector<std::pair<std::vector<int64_t>, migraphx::shape>> minus1_tests{
+        {{2, -1, 3}, {migraphx::shape::float_type, {2, 4, 3}}},
+        {{0, -1, 0}, {migraphx::shape::float_type, {24, 1, 1}}},
+        {{2, -1, 0}, {migraphx::shape::float_type, {2, 12, 1}}},
+        {{0, 0, -1}, {migraphx::shape::float_type, {24, 1, 1}}},
+        {{2, 0, -1}, {migraphx::shape::float_type, {2, 1, 12}}},
+        {{-1, 2, 3}, {migraphx::shape::float_type, {4, 2, 3}}},
+        {{-1, 0, 3}, {migraphx::shape::float_type, {8, 1, 3}}},
+        {{-1, 0, 0}, {migraphx::shape::float_type, {24, 1, 1}}},
+        {{-1, 3, 0}, {migraphx::shape::float_type, {8, 3, 1}}}};
+
+    for(auto& it : minus1_tests)
+    {
+        expect_shape(it.second, migraphx::make_op("reshape_lazy", {{"dims", it.first}}), input);
+    }
+}
+
+// This uses the permutation to compute the reshape_lazy since its simpler than
+// trying to calculate strides. As we collapse or expand dimensions, we
+// remove the collapsed dimensions or duplicate the expanded dimensions in
+// the permutation. Then we renumber the permutation. So for dimensions of 4,
+// 24, 1, 1, 1 with a permutation of 1, 0, 2, 3, 4 that reshape_lazys to 4, 1, 3,
+// 4, 2, we first remove the collapsed dimensions or duplicate the expanded
+// dimensions which gives 1, 0, 0, 0, 0. Then after renumbering we get a
+// final permutation of 4, 0, 1, 2, 3.
+TEST_CASE(reshape_lazy_nonstandard)
+{
+    auto input = migraphx::shape::from_permutation(migraphx::shape::float_type,
+                                                   {4, 24, 1, 1, 1},
+                                                   migraphx::invert_permutation({1, 0, 2, 3, 4}));
+    std::vector<std::pair<std::vector<std::size_t>, std::vector<int64_t>>> tests{
+        {{4, 24}, {1, 0}},
+        {{4, 24, 1, 1, 1, 1}, {1, 0, 2, 3, 4, 5}},
+        {{4, 8, 3, 1, 1}, {2, 0, 1, 3, 4}},
+        {{4, 1, 3, 4, 2}, {4, 0, 1, 2, 3}},
+        {{4, 1, 4, 3, 2}, {4, 0, 1, 2, 3}},
+        {{4, 2, 4, 3}, {3, 0, 1, 2}},
+        {{4, 2, 12, 1}, {2, 0, 1, 3}},
+        {{4, 2, 1, 12}, {3, 0, 1, 2}},
+        {{4, 4, 2, 3}, {3, 0, 1, 2}},
+        {{4, 8, 1, 3}, {3, 0, 1, 2}},
+        {{4, 8, 3, 1}, {2, 0, 1, 3}}};
+
+    for(const auto& [dims, perm] : tests)
+    {
+        migraphx::shape output = migraphx::shape::from_permutation(
+            migraphx::shape::float_type, dims, migraphx::invert_permutation(perm));
+        expect_shape(output, migraphx::make_op("reshape_lazy", {{"dims", dims}}), input);
+    }
+}
+
+TEST_CASE(reshape_lazy_nonstandard_squeeze)
+{
+    auto input = migraphx::shape::from_permutation(
+        migraphx::shape::float_type, {2, 16, 16, 1280}, migraphx::invert_permutation({0, 2, 3, 1}));
+    std::vector<std::size_t> lens = {2, 256, 1280};
+    migraphx::shape output        = migraphx::shape::from_permutation(
+        migraphx::shape::float_type, lens, migraphx::invert_permutation({0, 2, 1}));
+    expect_shape(output, migraphx::make_op("reshape_lazy", {{"dims", lens}}), input);
+}
+
+TEST_CASE(reshape_lazy_nonstandard_error)
+{
+    auto input = migraphx::shape::from_permutation(migraphx::shape::float_type,
+                                                   {4, 24, 1, 1, 1},
+                                                   migraphx::invert_permutation({1, 0, 2, 3, 4}));
+    for(auto&& new_shape : std::vector<std::vector<int64_t>>{{4, 8, 3, 2, 2},
+                                                             {1},
+                                                             {4, 8, 4},
+                                                             {4, 24, 1, 1, 1, 1, 2},
+                                                             {8, 4, 4},
+                                                             {4, 1, 3, -1, -1},
+                                                             {4, 3, 0},
+                                                             {4, 3, 2},
+                                                             {3, 0},
+                                                             {3, 2}})
+    {
+        throws_shape(migraphx::make_op("reshape_lazy", {{"dims", new_shape}}), input);
+    }
+}
+
+TEST_CASE(reshape_lazy_nonpacked_unsqueeze1)
+{
+    migraphx::shape input{migraphx::shape::float_type, {4, 16}, {32, 2}};
+    migraphx::shape output{migraphx::shape::float_type, {4, 2, 8}, {32, 16, 2}};
+    expect_shape(output, migraphx::make_op("reshape_lazy", {{"dims", output.lens()}}), input);
+}
+
+TEST_CASE(reshape_lazy_nonpacked_unsqueeze2)
+{
+    migraphx::shape input{migraphx::shape::float_type, {4, 16}, {32, 2}};
+    migraphx::shape output{migraphx::shape::float_type, {2, 2, 16}, {64, 32, 2}};
+    expect_shape(output, migraphx::make_op("reshape_lazy", {{"dims", output.lens()}}), input);
+}
+
+TEST_CASE(reshape_lazy_nonpacked_squeeze)
+{
+    migraphx::shape input{migraphx::shape::float_type, {4, 16}, {32, 2}};
+    migraphx::shape output{migraphx::shape::float_type, {64}, {2}};
+    expect_shape(output, migraphx::make_op("reshape_lazy", {{"dims", output.lens()}}), input);
+}
+
+TEST_CASE(reshape_lazy_broadcast_unsqueeze1)
+{
+    migraphx::shape input{migraphx::shape::float_type, {2, 256, 1280}, {0, 0, 1}};
+    migraphx::shape output{migraphx::shape::float_type, {2, 16, 16, 1280}, {0, 0, 0, 1}};
+    expect_shape(output, migraphx::make_op("reshape_lazy", {{"dims", output.lens()}}), input);
+}
+
+TEST_CASE(reshape_lazy_broadcast_unsqueeze2)
+{
+    migraphx::shape input{migraphx::shape::float_type, {2, 256, 1280}, {0, 0, 1}};
+    migraphx::shape output{migraphx::shape::float_type, {2, 256, 16, 80}, {0, 0, 80, 1}};
+    expect_shape(output, migraphx::make_op("reshape_lazy", {{"dims", output.lens()}}), input);
+}
+
+TEST_CASE(reshape_lazy_broadcast_squeeze)
+{
+    migraphx::shape input{migraphx::shape::float_type, {2, 16, 16, 1280}, {0, 0, 0, 1}};
+    migraphx::shape output{migraphx::shape::float_type, {2, 256, 1280}, {0, 0, 1}};
+    expect_shape(output, migraphx::make_op("reshape_lazy", {{"dims", output.lens()}}), input);
+}
+
+TEST_CASE(reshape_lazy_broadcast_squeeze_error)
+{
+    migraphx::shape input{migraphx::shape::float_type, {2, 16, 16, 1280}, {0, 0, 0, 1}};
+    std::vector<int64_t> new_shape = {2, 16, 20480};
+    throws_shape(migraphx::make_op("reshape_lazy", {{"dims", new_shape}}), input);
+}
+
+TEST_CASE(reshape_lazy_dyn_shape)
+{
+    migraphx::shape input{migraphx::shape::float_type, {{1, 4}, {24, 24}, {1, 1}, {1, 1}}};
+    for(auto&& new_shape : std::vector<std::vector<int64_t>>{
+            {-1, 1, 1, 24}, {0, 8, 3, 1}, {-1, 3, 4, 2}, {0, 2, 4, 3}})
+    {
+        std::vector<migraphx::shape::dynamic_dimension> out_dyn_dims{};
+        for(std::size_t i = 0; i < new_shape.size(); ++i)
+        {
+            if(new_shape[i] == 0 or new_shape[i] == -1)
+            {
+                out_dyn_dims.push_back(input.dyn_dims().at(i));
+            }
+            else
+            {
+                std::size_t d = new_shape[i];
+                out_dyn_dims.push_back({d, d});
+            }
+        }
+        migraphx::shape output{migraphx::shape::float_type, out_dyn_dims};
+        expect_shape(output, migraphx::make_op("reshape_lazy", {{"dims", new_shape}}), input);
+    }
+}
+
+TEST_CASE(reshape_lazy_multiple_non_fixed_error)
+{
+    migraphx::shape input{migraphx::shape::float_type, {{1, 4}, {24, 24}, {10, 20}, {1, 1}}};
+    std::vector<int64_t> new_shape = {0, 1, 0, 24};
+    throws_shape(migraphx::make_op("reshape_lazy", {{"dims", new_shape}}), input);
+}
+
+TEST_CASE(reshape_lazy_fixed_ele_not_matching_error)
+{
+    migraphx::shape input{migraphx::shape::float_type, {{1, 4}, {24, 24}, {10, 10}, {1, 1}}};
+    std::vector<int64_t> new_shape = {0, 1, 5, 24};
+    throws_shape(migraphx::make_op("reshape_lazy", {{"dims", new_shape}}), input);
+}
+
+TEST_CASE(reshape_lazy_non_fixed_not_matching_error)
+{
+    migraphx::shape input{migraphx::shape::float_type, {{1, 4}, {24, 24}, {1, 1}, {1, 1}}};
+    std::vector<int64_t> new_shape = {2, 1, 1, 24};
+    throws_shape(migraphx::make_op("reshape_lazy", {{"dims", new_shape}}), input);
+}
+
 TEST_CASE(return_shape_tuple)
 {
     using migraphx::shape;

test/quantization.cpp

@@ -83,7 +83,7 @@ TEST_CASE(param_add)
         auto hs  = mm->add_instruction(migraphx::make_op("add"), hp1, hp2);
         auto fs  = mm->add_instruction(
             migraphx::make_op("convert",
-                              {{"target_type", migraphx::to_value(migraphx::shape::float_type)}}),
+                               {{"target_type", migraphx::to_value(migraphx::shape::float_type)}}),
             hs);
         if(add_return)
         {
@@ -1013,7 +1013,7 @@ TEST_CASE(target_copy)
         std::vector<float> orig_result;
         run_prog(p, ref_t, m, orig_result);
 
-        EXPECT(migraphx::verify::verify_range(ref_result, orig_result));
+        EXPECT(migraphx::verify::verify_rms_range(ref_result, orig_result));
     }
 }
 
@@ -1077,7 +1077,10 @@ TEST_CASE(int8_quantization_dot)
         std::vector<float> no_quant_result;
         run_prog(p, ref_t, m, no_quant_result);
 
-        EXPECT(migraphx::verify::verify_range(quant_result, no_quant_result, 30000));
+        EXPECT(migraphx::verify::verify_range_with_tolerance(
+            quant_result,
+            migraphx::verify::expected{no_quant_result},
+            migraphx::verify::tolerance{0.003}));
     }
 }
 
@@ -1122,7 +1125,7 @@ TEST_CASE(int8_quantization_conv)
         std::vector<float> no_quant_result;
         run_prog(p, ref_t, no_quant_result);
 
-        EXPECT(migraphx::verify::verify_range(quant_result, no_quant_result));
+        EXPECT(migraphx::verify::verify_rms_range(quant_result, no_quant_result));
     }
 }
 
@@ -1274,7 +1277,7 @@ TEST_CASE(test_op_capture)
     cap_res.visit([&](auto output) { cap_vec.assign(output.begin(), output.end()); });
     res.visit([&](auto output) { vec.assign(output.begin(), output.end()); });
 
-    EXPECT(migraphx::verify::verify_range(vec, cap_vec));
+    EXPECT(migraphx::verify::verify_rms_range(vec, cap_vec));
 }
 
 int main(int argc, const char* argv[]) { test::run(argc, argv); }

test/ref/abs.cpp

@@ -42,7 +42,7 @@ TEST_CASE(abs_test)
     std::vector<float> results_vector(4);
     result.visit([&](auto output) { results_vector.assign(output.begin(), output.end()); });
     std::vector<float> gold{1, 2, 3, 4};
-    EXPECT(migraphx::verify::verify_range(results_vector, gold));
+    EXPECT(migraphx::verify::verify_rms_range(results_vector, gold));
 }
 
 TEST_CASE(abs_dyn_test)
@@ -62,5 +62,5 @@ TEST_CASE(abs_dyn_test)
     std::vector<float> results_vector(4);
     result.visit([&](auto output) { results_vector.assign(output.begin(), output.end()); });
     std::vector<float> gold{1, 2, 3, 4};
-    EXPECT(migraphx::verify::verify_range(results_vector, gold));
+    EXPECT(migraphx::verify::verify_rms_range(results_vector, gold));
 }

test/ref/acos.cpp

@@ -45,7 +45,7 @@ TEST_CASE(acos_test)
     std::vector<float> gold = data;
     std::transform(
         gold.begin(), gold.end(), gold.begin(), [](float n) -> float { return acosf(n); });
-    EXPECT(migraphx::verify::verify_range(results_vector, gold));
+    EXPECT(migraphx::verify::verify_rms_range(results_vector, gold));
 }
 
 TEST_CASE(acos_dyn_test)
@@ -68,5 +68,5 @@ TEST_CASE(acos_dyn_test)
     std::vector<float> gold = input_data;
     std::transform(
         gold.begin(), gold.end(), gold.begin(), [](float n) -> float { return acosf(n); });
-    EXPECT(migraphx::verify::verify_range(results_vector, gold));
+    EXPECT(migraphx::verify::verify_rms_range(results_vector, gold));
 }

test/ref/acosh.cpp

@@ -45,7 +45,7 @@ TEST_CASE(acosh_test)
     std::vector<float> gold = data;
     std::transform(
         gold.begin(), gold.end(), gold.begin(), [](float n) -> float { return acoshf(n); });
-    EXPECT(migraphx::verify::verify_range(results_vector, gold));
+    EXPECT(migraphx::verify::verify_rms_range(results_vector, gold));
 }
 
 TEST_CASE(acosh_dyn_test)
@@ -68,5 +68,5 @@ TEST_CASE(acosh_dyn_test)
     std::vector<float> gold = input_data;
     std::transform(
         gold.begin(), gold.end(), gold.begin(), [](float n) -> float { return acoshf(n); });
-    EXPECT(migraphx::verify::verify_range(results_vector, gold));
+    EXPECT(migraphx::verify::verify_rms_range(results_vector, gold));
 }