Merge branch 'develop' of github.com:ROCmSoftwarePlatform/AMDMIGraphX into ref_op_name

test/onnx/randomuniform_generated_seed_test.onnx

+!randomuniform_generated_seed_test:

+2
+inputoutput"
RandomUniform*
+sample_size
+
!randomuniform_generated_seed_testZ
+input
+

+

+
+b
+output
+
+

+
+B
\ No newline at end of file

test/onnx/softplus_nd_test.onnx

+softplus_nd_test:V
+
+
x
y"Softplussoftplus_nd_testZ
+
x
+
+
+
+
+b
+
y
+
+
+
+
+B
\ No newline at end of file

test/onnx/softplus_test.onnx

+
softplus_test:C
+
+
x
y"Softplus
softplus_testZ
+
x
+
+

+b
+
y
+
+

+B
\ No newline at end of file

test/onnx/softsign_nd_test.onnx

+softsign_nd_test:V
+
+
x
y"Softsignsoftsign_nd_testZ
+
x
+
+
+
+
+b
+
y
+
+
+
+
+B
\ No newline at end of file

test/onnx/softsign_test.onnx

+
softsign_test:C
+
+
x
y"Softsign
softsign_testZ
+
x
+
+

+b
+
y
+
+

+B
\ No newline at end of file

test/onnx/verify_onnx.cpp

@@ -126,6 +126,51 @@ TEST_CASE(gather_elements)
     EXPECT(migraphx::verify_range(result_vector, gold));
 }
 
+TEST_CASE(greaterorequal_test)
+{
+    migraphx::program p = migraphx::parse_onnx("greaterorequal_test.onnx");
+    p.compile(migraphx::ref::target{});
+
+    migraphx::shape s{migraphx::shape::float_type, {3}};
+    std::vector<float> data1 = {0.25, 0.75, 0.9375};
+    std::vector<float> data2 = {0.25, 0.74, 0.9411};
+
+    migraphx::parameter_map pp;
+    pp["x1"] = migraphx::argument(s, data1.data());
+    pp["x2"] = migraphx::argument(s, data2.data());
+
+    auto result = p.eval(pp).back();
+    std::vector<float> result_vector;
+    result.visit([&](auto output) { result_vector.assign(output.begin(), output.end()); });
+
+    std::vector<float> gold = {1.0, 1.0, 0.0};
+    EXPECT(migraphx::verify_range(result_vector, gold));
+}
+
+TEST_CASE(hardsigmoid_verify_test)
+{
+    migraphx::program p = migraphx::parse_onnx("hardsigmoid_verify_test.onnx");
+    p.compile(migraphx::ref::target{});
+
+    migraphx::shape s{migraphx::shape::float_type, {2, 5}};
+    std::vector<float> data = {-10.0, -2.5, -1.0, -0.5, 0, 1.0, 2.0, 2.5, 2.6, 100.0};
+
+    float alpha = 0.2;
+    float beta  = 0.5;
+    migraphx::parameter_map pp;
+    pp["x"] = migraphx::argument(s, data.data());
+
+    auto result = p.eval(pp).back();
+    std::vector<float> result_vector;
+    result.visit([&](auto output) { result_vector.assign(output.begin(), output.end()); });
+
+    std::vector<float> gold(10);
+    std::transform(data.begin(), data.end(), gold.begin(), [&](auto x) {
+        return std::max(0.0f, std::min(x * alpha + beta, 1.0f));
+    });
+    EXPECT(migraphx::verify_range(result_vector, gold));
+}
+
 TEST_CASE(if_else_test)
 {
     migraphx::program p = migraphx::parse_onnx("if_else_test.onnx");
@@ -348,6 +393,64 @@ TEST_CASE(lessorequal_test)
     EXPECT(migraphx::verify_range(result_vector, gold));
 }
 
+TEST_CASE(mean_broadcast_test)
+{
+    migraphx::program p = migraphx::parse_onnx("mean_broadcast_test.onnx");
+    p.compile(migraphx::ref::target{});
+
+    migraphx::shape s0{migraphx::shape::float_type, {1, 3, 4}};
+    std::vector<float> data0(12, 1);
+    migraphx::shape s1{migraphx::shape::float_type, {1, 2, 3, 4}};
+    std::vector<float> data1(24, 2);
+    migraphx::shape s2{migraphx::shape::float_type, {4}};
+    std::vector<float> data2(4, 3);
+    migraphx::shape s3{migraphx::shape::float_type, {1}};
+    std::vector<float> data3(1, 4);
+    migraphx::shape s4{migraphx::shape::float_type, {2, 3, 1}};
+    std::vector<float> data4(6, 5);
+
+    migraphx::parameter_map pp;
+    pp["0"] = migraphx::argument(s0, data0.data());
+    pp["1"] = migraphx::argument(s1, data1.data());
+    pp["2"] = migraphx::argument(s2, data2.data());
+    pp["3"] = migraphx::argument(s3, data3.data());
+    pp["4"] = migraphx::argument(s4, data4.data());
+
+    auto result = p.eval(pp).back();
+    std::vector<float> result_vector;
+    result.visit([&](auto output) { result_vector.assign(output.begin(), output.end()); });
+
+    std::vector<float> gold(24, 3);
+    EXPECT(migraphx::verify_range(result_vector, gold));
+}
+
+TEST_CASE(mean_test)
+{
+    migraphx::program p = migraphx::parse_onnx("mean_test.onnx");
+    p.compile(migraphx::ref::target{});
+
+    migraphx::shape s{migraphx::shape::double_type, {2, 2, 2}};
+    const int num_elms = 8;
+    const int num_data = 10;
+    const std::vector<double> scalars{1.0, 2.0, -2.5, 3.3, 10.7, -1.0, 100.0, 7.9, 0.01, -56.8};
+    std::vector<std::vector<double>> data;
+    std::transform(scalars.begin(), scalars.end(), std::back_inserter(data), [&](const auto& i) {
+        return std::vector<double>(num_elms, i);
+    });
+
+    migraphx::parameter_map pp;
+    for(std::size_t i = 0; i < num_data; ++i)
+        pp[std::to_string(i)] = migraphx::argument(s, data[i].data());
+
+    auto result = p.eval(pp).back();
+    std::vector<double> result_vector;
+    result.visit([&](auto output) { result_vector.assign(output.begin(), output.end()); });
+
+    const auto mean = std::accumulate(scalars.begin(), scalars.end(), 0.0) / num_data;
+    std::vector<double> gold(num_elms, mean);
+    EXPECT(migraphx::verify_range(result_vector, gold));
+}
+
 TEST_CASE(nonzero_test)
 {
     migraphx::program p = migraphx::parse_onnx("nonzero_dynamic_test.onnx");
@@ -564,6 +667,48 @@ TEST_CASE(slice_step_test)
     EXPECT(migraphx::verify_range(result_vector, gold));
 }
 
+TEST_CASE(softplus_test)
+{
+    migraphx::program p = migraphx::parse_onnx("softplus_test.onnx");
+    p.compile(migraphx::ref::target{});
+
+    migraphx::shape s{migraphx::shape::float_type, {5}};
+    std::vector<float> data = {0, 1, 2, 3, 4};
+
+    migraphx::parameter_map pp;
+    pp["x"] = migraphx::argument(s, data.data());
+
+    auto result = p.eval(pp).back();
+    std::vector<float> result_vector;
+    result.visit([&](auto output) { result_vector.assign(output.begin(), output.end()); });
+    std::vector<float> gold(5);
+    std::transform(
+        data.begin(), data.end(), gold.begin(), [](auto x) { return std::log1p(std::exp(x)); });
+
+    EXPECT(migraphx::verify_range(result_vector, gold));
+}
+
+TEST_CASE(softsign_test)
+{
+    migraphx::program p = migraphx::parse_onnx("softsign_test.onnx");
+    p.compile(migraphx::ref::target{});
+
+    migraphx::shape s{migraphx::shape::float_type, {5}};
+    std::vector<float> data = {0, 1, 2, 3, 4};
+
+    migraphx::parameter_map pp;
+    pp["x"] = migraphx::argument(s, data.data());
+
+    auto result = p.eval(pp).back();
+    std::vector<float> result_vector;
+    result.visit([&](auto output) { result_vector.assign(output.begin(), output.end()); });
+    std::vector<float> gold(5);
+    std::transform(
+        data.begin(), data.end(), gold.begin(), [](auto x) { return x / (1.0 + std::abs(x)); });
+
+    EXPECT(migraphx::verify_range(result_vector, gold));
+}
+
 TEST_CASE(upsample_test)
 {
     migraphx::program p = migraphx::parse_onnx("upsample_test.onnx");

test/py/onnx_backend_test.py

@@ -119,6 +119,7 @@ def create_backend_test(testname=None, target_device=None):
         backend_test.include(r'.*test_globalmaxpool.*')
         backend_test.include(r'.*test_greater.*')
         backend_test.include(r'.*test_hardsigmoid.*')
+        backend_test.include(r'.*test_hardswish.*')
         backend_test.include(r'.*test_identity.*')
         backend_test.include(r'.*test_if.*')
         backend_test.include(r'.*test_LeakyReLU*')
@@ -266,18 +267,8 @@ def create_backend_test(testname=None, target_device=None):
         backend_test.exclude(r'test_gathernd_example_float32_cpu')
         backend_test.exclude(r'test_gathernd_example_int32_batch_dim1_cpu')
         backend_test.exclude(r'test_gathernd_example_int32_cpu')
-        backend_test.exclude(r'test_greater_equal_bcast_cpu')
-        backend_test.exclude(r'test_greater_equal_bcast_expanded_cpu')
-        backend_test.exclude(r'test_greater_equal_cpu')
-        backend_test.exclude(r'test_greater_equal_expanded_cpu')
-        backend_test.exclude(r'test_hardsigmoid_cpu')
-        backend_test.exclude(r'test_hardsigmoid_default_cpu')
-        backend_test.exclude(r'test_hardsigmoid_example_cpu')
         backend_test.exclude(r'test_identity_sequence_cpu')
         backend_test.exclude(r'test_maxpool_2d_uint8_cpu')
-        backend_test.exclude(r'test_mean_example_cpu')
-        backend_test.exclude(r'test_mean_one_input_cpu')
-        backend_test.exclude(r'test_mean_two_inputs_cpu')
         backend_test.exclude(r'test_negative_log_likelihood_loss_*')
         backend_test.exclude(r'test_scatternd_*')
 
@@ -285,12 +276,7 @@ def create_backend_test(testname=None, target_device=None):
         backend_test.exclude(r'test_size_cpu')
         backend_test.exclude(r'test_size_example_cpu')
         backend_test.exclude(r'test_softmax_cross_entropy_*')
-        backend_test.exclude(r'test_softplus_cpu')
-        backend_test.exclude(r'test_softplus_example_cpu')
-        backend_test.exclude(r'test_softsign_cpu')
-        backend_test.exclude(r'test_softsign_example_cpu')
         backend_test.exclude(r'test_Embedding_cpu')
-        backend_test.exclude(r'test_Softplus_cpu')
 
         # real model tests
         backend_test.exclude(r'test_inception_v1_cpu')

test/ref_ops_nonstd_shape_test.cpp

+#include <iostream>
+#include <vector>
+#include <migraphx/literal.hpp>
+#include <migraphx/instruction.hpp>
+#include <migraphx/ref/target.hpp>
+#include <migraphx/verify.hpp>
+#include <migraphx/make_op.hpp>
+#include <migraphx/auto_contiguous.hpp>
+#include <migraphx/pass_manager.hpp>
+#include "test.hpp"
+
+TEST_CASE(argmax_test_nonstd_shape)
+{
+    migraphx::program p;
+    auto* mm                = p.get_main_module();
+    std::vector<float> data = {1.2255,  1.6834,  -2.0305, -0.3221, 0.4701,  0.2583, 0.7545, 2.5758,
+                               -1.6849, 0.0928,  0.9022,  -0.8765, -0.4090, 0.9301, 2.0724, -1.5706,
+                               0.4867,  -0.1493, 0.6957,  -0.2179, 0.7142,  0.7177, 0.0183, 1.3497};
+    migraphx::shape data_shape{migraphx::shape::float_type, {2, 3, 4}};
+    auto dl = mm->add_literal(migraphx::literal{data_shape, data});
+    auto dl_trans =
+        mm->add_instruction(migraphx::make_op("transpose", {{"permutation", {1, 2, 0}}}), dl);
+    mm->add_instruction(migraphx::make_op("argmax", {{"axis", -3}}), dl_trans);
+    auto p_uncompiled = p;
+    p.compile(migraphx::ref::target{});
+    auto result   = p.eval({}).back();
+    auto res_gold = p_uncompiled.eval({}).back();
+    std::vector<int64_t> result_vec;
+    result.visit([&](auto output) { result_vec.assign(output.begin(), output.end()); });
+    std::vector<int64_t> res_gold_vec;
+    res_gold.visit([&](auto output) { res_gold_vec.assign(output.begin(), output.end()); });
+    EXPECT(migraphx::verify_range(result_vec, res_gold_vec));
+}
+
+TEST_CASE(argmin_test_nonstd_shape)
+{
+    migraphx::program p;
+    auto* mm                = p.get_main_module();
+    std::vector<float> data = {1.2255,  1.6834,  -2.0305, -0.3221, 0.4701,  0.2583, 0.7545, 2.5758,
+                               -1.6849, 0.0928,  0.9022,  -0.8765, -0.4090, 0.9301, 2.0724, -1.5706,
+                               0.4867,  -0.1493, 0.6957,  -0.2179, 0.7142,  0.7177, 0.0183, 1.3497};
+    migraphx::shape data_shape{migraphx::shape::float_type, {2, 3, 4}};
+    auto dl = mm->add_literal(migraphx::literal{data_shape, data});
+    auto dl_trans =
+        mm->add_instruction(migraphx::make_op("transpose", {{"permutation", {1, 2, 0}}}), dl);
+    mm->add_instruction(migraphx::make_op("argmin", {{"axis", -1}}), dl_trans);
+    auto p_uncompiled = p;
+    p.compile(migraphx::ref::target{});
+    auto result   = p.eval({}).back();
+    auto res_gold = p_uncompiled.eval({}).back();
+    std::vector<int64_t> result_vec;
+    result.visit([&](auto output) { result_vec.assign(output.begin(), output.end()); });
+    std::vector<int64_t> res_gold_vec;
+    res_gold.visit([&](auto output) { res_gold_vec.assign(output.begin(), output.end()); });
+    EXPECT(migraphx::verify_range(result_vec, res_gold_vec));
+}
+
+int main(int argc, const char* argv[]) { test::run(argc, argv); }

test/ref_ops_test.cpp

@@ -1473,6 +1473,32 @@ TEST_CASE(fp32_fp16_test)
     test_case({"add"});
 }
 
+TEST_CASE(gather_non_std_test)
+{
+    {
+        migraphx::program p;
+        auto* mm = p.get_main_module();
+
+        std::vector<float> data = {0.5f, 3.5f, 6.5f, 1.5f, 4.5f, 7.5f, 2.5f, 2.5f, 8.5f};
+        migraphx::shape s{migraphx::shape::float_type, {3, 3}};
+        auto d = mm->add_literal(migraphx::literal{s, data});
+        migraphx::shape s_indices{migraphx::shape::int32_type, {2, 2}};
+        std::vector<int> indices{-3, -3, -1, -1};
+        auto ind = mm->add_literal(migraphx::literal{s_indices, indices});
+        auto td = mm->add_instruction(migraphx::make_op("transpose", {{"permutation", {1, 0}}}), d);
+        auto tind =
+            mm->add_instruction(migraphx::make_op("transpose", {{"permutation", {1, 0}}}), ind);
+
+        mm->add_instruction(migraphx::make_op("gather", {{"axis", 0}}), td, tind);
+        auto result               = p.eval({}).back();
+        std::vector<float> golden = {
+            0.5f, 1.5f, 2.5f, 6.5f, 7.5f, 8.5f, 0.5f, 1.5f, 2.5f, 6.5f, 7.5f, 8.5f};
+        std::vector<float> res_data;
+        result.visit([&](auto output) { res_data.assign(output.begin(), output.end()); });
+        EXPECT(migraphx::verify_range(res_data, golden));
+    }
+}
+
 TEST_CASE(gather_test)
 {
     {
@@ -2784,7 +2810,6 @@ TEST_CASE(nms_not_center_test)
     auto output = p.eval({}).back();
     std::vector<int64_t> result;
     output.visit([&](auto out) { result.assign(out.begin(), out.end()); });
-    std::cout << "output = " << output << std::endl;
     std::vector<int64_t> gold = {0, 0, 3, 0, 0, 0, 0, 0, 5, 0, 0, 0, 0, 0, 0, 0, 0, 0};
     EXPECT(migraphx::verify_range(result, gold));
 }
@@ -2818,7 +2843,6 @@ TEST_CASE(nms_test)
     auto output = p.eval({}).back();
     std::vector<int64_t> result;
     output.visit([&](auto out) { result.assign(out.begin(), out.end()); });
-    std::cout << "output = " << output << std::endl;
     std::vector<int64_t> gold = {0, 0, 3, 0, 0, 0, 0, 0, 5, 0, 0, 0, 0, 0, 0, 0, 0, 0};
     EXPECT(migraphx::verify_range(result, gold));
 }

test/verify/run_verify.cpp

@@ -6,6 +6,7 @@
 #include <migraphx/ref/target.hpp>
 #include <migraphx/ranges.hpp>
 #include <migraphx/generate.hpp>
+#include <migraphx/load_save.hpp>
 #include <migraphx/verify_args.hpp>
 #include <set>
 
@@ -15,6 +16,7 @@
 
 MIGRAPHX_DECLARE_ENV_VAR(MIGRAPHX_TRACE_TEST_COMPILE)
 MIGRAPHX_DECLARE_ENV_VAR(MIGRAPHX_TRACE_TEST)
+MIGRAPHX_DECLARE_ENV_VAR(MIGRAPHX_DUMP_TEST)
 
 // An improved async, that doesn't block
 template <class Function>
@@ -125,6 +127,8 @@ void run_verify::verify(const std::string& name, const migraphx::program& p) con
     using result_future =
         std::future<std::pair<migraphx::program, std::vector<migraphx::argument>>>;
     auto_print::set_terminate_handler(name);
+    if(migraphx::enabled(MIGRAPHX_DUMP_TEST{}))
+        migraphx::save(p, name + ".mx");
     std::vector<std::pair<std::string, result_future>> results;
     std::vector<std::string> target_names;
     for(const auto& tname : migraphx::get_targets())

test/verify/test_arg_ops.cpp

@@ -2,34 +2,92 @@
 #include "verify_program.hpp"
 #include <migraphx/program.hpp>
 #include <migraphx/generate.hpp>
+#include <migraphx/make_op.hpp>
 #include <migraphx/op/argmax.hpp>
 #include <migraphx/op/argmin.hpp>
 
-template <class T, int Axis>
-struct test_arg_ops : verify_program<test_arg_ops<T, Axis>>
+template <class T, int Axis, int NonStdShape>
+struct test_arg_ops : verify_program<test_arg_ops<T, Axis, NonStdShape>>
 {
     migraphx::program create_program() const
     {
         migraphx::program p;
         auto* mm = p.get_main_module();
-        migraphx::shape s{migraphx::shape::float_type, {2, 3, 4, 1025}};
+        migraphx::shape s{migraphx::shape::float_type, {2, 1, 4, 1025}};
         auto param = mm->add_parameter("data", s);
+        switch(NonStdShape)
+        {
+        case 0:
+            param = mm->add_instruction(
+                migraphx::make_op("transpose", {{"permutation", {0, 2, 3, 1}}}), param);
+            break;
+        case 1:
+            param = mm->add_instruction(
+                migraphx::make_op("multibroadcast", {{"out_lens", {2, 3, 4, 1025}}}), param);
+            break;
+        case 2:
+            param = mm->add_instruction(
+                migraphx::make_op("slice", {{"axes", {2}}, {"starts", {1}}, {"ends", {3}}}), param);
+            break;
+        default: break;
+        }
         mm->add_instruction(T{Axis}, param);
-
         return p;
     }
 };
-
-template struct test_arg_ops<migraphx::op::argmax, 0>;
-template struct test_arg_ops<migraphx::op::argmax, 1>;
-template struct test_arg_ops<migraphx::op::argmax, 2>;
-template struct test_arg_ops<migraphx::op::argmax, 3>;
-template struct test_arg_ops<migraphx::op::argmax, -1>;
-template struct test_arg_ops<migraphx::op::argmax, -2>;
-
-template struct test_arg_ops<migraphx::op::argmin, 0>;
-template struct test_arg_ops<migraphx::op::argmin, 1>;
-template struct test_arg_ops<migraphx::op::argmin, 2>;
-template struct test_arg_ops<migraphx::op::argmin, 3>;
-template struct test_arg_ops<migraphx::op::argmin, -3>;
-template struct test_arg_ops<migraphx::op::argmin, -4>;
+// transpose argmax tests
+template struct test_arg_ops<migraphx::op::argmax, 0, 0>;
+template struct test_arg_ops<migraphx::op::argmax, 1, 0>;
+template struct test_arg_ops<migraphx::op::argmax, 2, 0>;
+template struct test_arg_ops<migraphx::op::argmax, 3, 0>;
+template struct test_arg_ops<migraphx::op::argmax, -1, 0>;
+template struct test_arg_ops<migraphx::op::argmax, -2, 0>;
+// transpose argmin tests
+template struct test_arg_ops<migraphx::op::argmin, 0, 0>;
+template struct test_arg_ops<migraphx::op::argmin, 1, 0>;
+template struct test_arg_ops<migraphx::op::argmin, 2, 0>;
+template struct test_arg_ops<migraphx::op::argmin, 3, 0>;
+template struct test_arg_ops<migraphx::op::argmin, -3, 0>;
+template struct test_arg_ops<migraphx::op::argmin, -4, 0>;
+// broadcast argmax tests
+template struct test_arg_ops<migraphx::op::argmax, 0, 1>;
+template struct test_arg_ops<migraphx::op::argmax, 1, 1>;
+template struct test_arg_ops<migraphx::op::argmax, 2, 1>;
+template struct test_arg_ops<migraphx::op::argmax, 3, 1>;
+template struct test_arg_ops<migraphx::op::argmax, -1, 1>;
+template struct test_arg_ops<migraphx::op::argmax, -2, 1>;
+// broadcast argmin tests
+template struct test_arg_ops<migraphx::op::argmin, 0, 1>;
+template struct test_arg_ops<migraphx::op::argmin, 1, 1>;
+template struct test_arg_ops<migraphx::op::argmin, 2, 1>;
+template struct test_arg_ops<migraphx::op::argmin, 3, 1>;
+template struct test_arg_ops<migraphx::op::argmin, -3, 1>;
+template struct test_arg_ops<migraphx::op::argmin, -4, 1>;
+// slice argmax tests
+template struct test_arg_ops<migraphx::op::argmax, 0, 2>;
+template struct test_arg_ops<migraphx::op::argmax, 1, 2>;
+template struct test_arg_ops<migraphx::op::argmax, 2, 2>;
+template struct test_arg_ops<migraphx::op::argmax, 3, 2>;
+template struct test_arg_ops<migraphx::op::argmax, -1, 2>;
+template struct test_arg_ops<migraphx::op::argmax, -2, 2>;
+// slice argmin tests
+template struct test_arg_ops<migraphx::op::argmin, 0, 2>;
+template struct test_arg_ops<migraphx::op::argmin, 1, 2>;
+template struct test_arg_ops<migraphx::op::argmin, 2, 2>;
+template struct test_arg_ops<migraphx::op::argmin, 3, 2>;
+template struct test_arg_ops<migraphx::op::argmin, -3, 2>;
+template struct test_arg_ops<migraphx::op::argmin, -4, 2>;
+// default case, standard shape argmax tests
+template struct test_arg_ops<migraphx::op::argmax, 0, 3>;
+template struct test_arg_ops<migraphx::op::argmax, 1, 3>;
+template struct test_arg_ops<migraphx::op::argmax, 2, 3>;
+template struct test_arg_ops<migraphx::op::argmax, 3, 3>;
+template struct test_arg_ops<migraphx::op::argmax, -1, 3>;
+template struct test_arg_ops<migraphx::op::argmax, -2, 3>;
+// default case, standard shape argmin tests
+template struct test_arg_ops<migraphx::op::argmin, 0, 3>;
+template struct test_arg_ops<migraphx::op::argmin, 1, 3>;
+template struct test_arg_ops<migraphx::op::argmin, 2, 3>;
+template struct test_arg_ops<migraphx::op::argmin, 3, 3>;
+template struct test_arg_ops<migraphx::op::argmin, -3, 3>;
+template struct test_arg_ops<migraphx::op::argmin, -4, 3>;

test/verify/test_conv_bias_clipped_relu.cpp

@@ -28,9 +28,9 @@ struct test_conv_bias_clipped_relu : verify_program<test_conv_bias_clipped_relu>
         auto min_val  = mm->add_literal(0.0f);
         auto max_val  = mm->add_literal(6.0f);
         min_val       = mm->add_instruction(
-            migraphx::make_op("multibroadcast", {{"out_lens", input_lens}}), min_val);
+            migraphx::make_op("multibroadcast", {{"out_lens", conv->get_shape().lens()}}), min_val);
         max_val = mm->add_instruction(
-            migraphx::make_op("multibroadcast", {{"out_lens", input_lens}}), max_val);
+            migraphx::make_op("multibroadcast", {{"out_lens", conv->get_shape().lens()}}), max_val);
         mm->add_instruction(migraphx::make_op("clip"), bias_add, min_val, max_val);
         return p;
     }

test/verify/test_hsqrt.cpp

+
+#include "verify_program.hpp"
+#include <migraphx/program.hpp>
+#include <migraphx/generate.hpp>
+#include <migraphx/make_op.hpp>
+
+struct test_hsqrt : verify_program<test_hsqrt>
+{
+    migraphx::program create_program() const
+    {
+        migraphx::program p;
+        auto* mm = p.get_main_module();
+        migraphx::shape s{migraphx::shape::half_type, {2, 3, 4, 6}};
+        auto param     = mm->add_parameter("x", s);
+        auto param_abs = mm->add_instruction(migraphx::make_op("abs"), param);
+        mm->add_instruction(migraphx::make_op("sqrt"), param_abs);
+        return p;
+    }
+};

test/verify/test_nonstd_gather.cpp

+
+#include "verify_program.hpp"
+#include <migraphx/program.hpp>
+#include <migraphx/generate.hpp>
+#include <migraphx/make_op.hpp>
+
+struct test_nonstd_gather : verify_program<test_nonstd_gather>
+{
+    migraphx::program create_program() const
+    {
+        migraphx::program p;
+        auto* mm = p.get_main_module();
+        migraphx::shape s{migraphx::shape::float_type, {3, 3}};
+        migraphx::shape s_indices{migraphx::shape::int32_type, {2, 2}};
+        std::vector<int> indices{1, 1, 0, 2};
+        auto d  = mm->add_parameter("data", s);
+        auto td = mm->add_instruction(migraphx::make_op("transpose", {{"permutation", {1, 0}}}), d);
+        auto ind = mm->add_literal(migraphx::literal{s_indices, indices});
+        auto tind =
+            mm->add_instruction(migraphx::make_op("transpose", {{"permutation", {1, 0}}}), ind);
+        auto r = mm->add_instruction(migraphx::make_op("gather", {{"axis", 1}}), td, tind);
+        mm->add_return({r});
+
+        return p;
+    }
+};

tools/generate.sh

 DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )" && pwd )"
 SRC_DIR=$DIR/../src
-ls -1 $DIR/include/ | xargs -n 1 -P $(nproc) -I{} -t bash -c "python3.6 $DIR/te.py $DIR/include/{} | clang-format-5.0 -style=file > $SRC_DIR/include/migraphx/{}"
+PYTHON=python3
+if type -p python3.6 > /dev/null ; then
+    PYTHON=python3.6
+fi
+if type -p python3.8 > /dev/null ; then
+    PYTHON=python3.8
+fi
+ls -1 $DIR/include/ | xargs -n 1 -P $(nproc) -I{} -t bash -c "$PYTHON $DIR/te.py $DIR/include/{} | clang-format-5.0 -style=file > $SRC_DIR/include/migraphx/{}"
 
 function api {
-    python3.6 $DIR/api.py $SRC_DIR/api/migraphx.py $1 | clang-format-5.0 -style=file > $2
+    $PYTHON $DIR/api.py $SRC_DIR/api/migraphx.py $1 | clang-format-5.0 -style=file > $2
 }
 
 api $DIR/api/migraphx.h $SRC_DIR/api/include/migraphx/migraphx.h

tools/roctx.py

@@ -21,7 +21,7 @@ def parse_args():
         description="Parser for MIGraphX ROCTX Markers")
     parser.add_argument('--json-path',
                         type=str,
-                        metavar='json_path',
+                        metavar='json-path',
                         help='Path to json file')
     parser.add_argument('--out',
                         type=str,