Merge branch 'develop' into multiply-add

src/targets/gpu/device/sqrt.cpp

+#include <migraphx/gpu/device/sqrt.hpp>
+#include <migraphx/gpu/device/nary.hpp>
+#include <migraphx/gpu/device/types.hpp>
+
+namespace migraphx {
+inline namespace MIGRAPHX_INLINE_NS {
+namespace gpu {
+namespace device {
+
+void sqrt(hipStream_t stream, const argument& result, const argument& arg)
+{
+    nary(stream, result, arg)([](auto x) { return ::sqrt(to_hip_type(x)); });
+}
+
+} // namespace device
+} // namespace gpu
+} // namespace MIGRAPHX_INLINE_NS
+} // namespace migraphx

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/device/div.hpp

+#ifndef MIGRAPHX_GUARD_RTGLIB_DEVICE_DIV_HPP
+#define MIGRAPHX_GUARD_RTGLIB_DEVICE_DIV_HPP
+
+#include <migraphx/argument.hpp>
+#include <migraphx/config.hpp>
+#include <hip/hip_runtime_api.h>
+
+namespace migraphx {
+inline namespace MIGRAPHX_INLINE_NS {
+namespace gpu {
+namespace device {
+
+void div(hipStream_t stream, const argument& result, const argument& arg1, const argument& arg2);
+
+} // namespace device
+} // namespace gpu
+} // namespace MIGRAPHX_INLINE_NS
+} // namespace migraphx
+
+#endif

src/targets/gpu/include/migraphx/gpu/device/pow.hpp

+
+#ifndef MIGRAPHX_GUARD_RTGLIB_DEVICE_POW_HPP
+#define MIGRAPHX_GUARD_RTGLIB_DEVICE_POW_HPP
+
+#include <migraphx/argument.hpp>
+#include <migraphx/config.hpp>
+#include <hip/hip_runtime_api.h>
+
+namespace migraphx {
+inline namespace MIGRAPHX_INLINE_NS {
+namespace gpu {
+namespace device {
+
+void pow(hipStream_t stream, const argument& result, const argument& arg1, const argument& arg2);
+
+} // namespace device
+} // namespace gpu
+} // namespace MIGRAPHX_INLINE_NS
+} // namespace migraphx
+
+#endif

src/targets/gpu/include/migraphx/gpu/device/reduce_mean.hpp

+#ifndef MIGRAPHX_GUARD_RTGLIB_DEVICE_REDUCE_MEAN_HPP
+#define MIGRAPHX_GUARD_RTGLIB_DEVICE_REDUCE_MEAN_HPP
+
+#include <migraphx/argument.hpp>
+#include <migraphx/config.hpp>
+#include <hip/hip_runtime_api.h>
+
+namespace migraphx {
+inline namespace MIGRAPHX_INLINE_NS {
+namespace gpu {
+namespace device {
+
+void reduce_mean(hipStream_t stream, const argument& result, const argument& arg);
+
+} // namespace device
+} // namespace gpu
+} // namespace MIGRAPHX_INLINE_NS
+} // namespace migraphx
+
+#endif

src/targets/gpu/include/migraphx/gpu/device/rsqrt.hpp

+#ifndef MIGRAPHX_GUARD_RTGLIB_DEVICE_RSQRT_HPP
+#define MIGRAPHX_GUARD_RTGLIB_DEVICE_RSQRT_HPP
+
+#include <migraphx/argument.hpp>
+#include <migraphx/config.hpp>
+#include <hip/hip_runtime_api.h>
+
+namespace migraphx {
+inline namespace MIGRAPHX_INLINE_NS {
+namespace gpu {
+namespace device {
+
+void rsqrt(hipStream_t stream, const argument& result, const argument& arg);
+
+} // namespace device
+} // namespace gpu
+} // namespace MIGRAPHX_INLINE_NS
+} // namespace migraphx
+
+#endif

src/targets/gpu/include/migraphx/gpu/device/sqdiff.hpp

+
+#ifndef MIGRAPHX_GUARD_RTGLIB_DEVICE_SQDIFF_HPP
+#define MIGRAPHX_GUARD_RTGLIB_DEVICE_SQDIFF_HPP
+
+#include <migraphx/argument.hpp>
+#include <migraphx/config.hpp>
+#include <hip/hip_runtime_api.h>
+
+namespace migraphx {
+inline namespace MIGRAPHX_INLINE_NS {
+namespace gpu {
+namespace device {
+
+void sqdiff(hipStream_t stream, const argument& result, const argument& arg1, const argument& arg2);
+
+} // namespace device
+} // namespace gpu
+} // namespace MIGRAPHX_INLINE_NS
+} // namespace migraphx
+
+#endif

src/targets/gpu/include/migraphx/gpu/device/sqrt.hpp

+#ifndef MIGRAPHX_GUARD_RTGLIB_DEVICE_SQRT_HPP
+#define MIGRAPHX_GUARD_RTGLIB_DEVICE_SQRT_HPP
+
+#include <migraphx/argument.hpp>
+#include <migraphx/config.hpp>
+#include <hip/hip_runtime_api.h>
+
+namespace migraphx {
+inline namespace MIGRAPHX_INLINE_NS {
+namespace gpu {
+namespace device {
+
+void sqrt(hipStream_t stream, const argument& result, const argument& arg);
+
+} // namespace device
+} // namespace gpu
+} // namespace MIGRAPHX_INLINE_NS
+} // namespace migraphx
+
+#endif

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

+#ifndef MIGRAPHX_GUARD_RTGLIB_DIV_HPP
+#define MIGRAPHX_GUARD_RTGLIB_DIV_HPP
+
+#include <migraphx/gpu/oper.hpp>
+#include <migraphx/gpu/device/div.hpp>
+
+namespace migraphx {
+inline namespace MIGRAPHX_INLINE_NS {
+namespace gpu {
+
+struct hip_div : binary_device<hip_div, device::div>
+{
+};
+
+} // namespace gpu
+} // namespace MIGRAPHX_INLINE_NS
+} // namespace migraphx
+
+#endif

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];
     }
 

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

+#ifndef MIGRAPHX_GUARD_RTGLIB_POW_HPP
+#define MIGRAPHX_GUARD_RTGLIB_POW_HPP
+
+#include <migraphx/gpu/oper.hpp>
+#include <migraphx/gpu/device/pow.hpp>
+
+namespace migraphx {
+inline namespace MIGRAPHX_INLINE_NS {
+namespace gpu {
+
+struct hip_pow : binary_device<hip_pow, device::pow>
+{
+};
+
+} // namespace gpu
+} // namespace MIGRAPHX_INLINE_NS
+} // namespace migraphx
+
+#endif

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

+#ifndef MIGRAPHX_GUARD_RTGLIB_REDUCE_MEAN_HPP
+#define MIGRAPHX_GUARD_RTGLIB_REDUCE_MEAN_HPP
+
+#include <migraphx/shape.hpp>
+#include <migraphx/op/reduce_mean.hpp>
+#include <migraphx/reflect.hpp>
+
+namespace migraphx {
+inline namespace MIGRAPHX_INLINE_NS {
+namespace gpu {
+
+struct context;
+
+struct hip_reduce_mean
+{
+    op::reduce_mean op;
+
+    template <class Self, class F>
+    static auto reflect(Self& self, F f)
+    {
+        return migraphx::reflect(self.op, f);
+    }
+
+    std::string name() const { return "gpu::reduce_mean"; }
+    shape compute_shape(std::vector<shape> inputs) const;
+    argument
+    compute(context& ctx, const shape& output_shape, const std::vector<argument>& args) const;
+    std::ptrdiff_t output_alias(const std::vector<shape>& shapes) const
+    {
+        return shapes.size() - 1;
+    }
+};
+
+} // namespace gpu
+} // namespace MIGRAPHX_INLINE_NS
+} // namespace migraphx
+
+#endif

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

+#ifndef MIGRAPHX_GUARD_RTGLIB_RSQRT_HPP
+#define MIGRAPHX_GUARD_RTGLIB_RSQRT_HPP
+
+#include <migraphx/gpu/oper.hpp>
+#include <migraphx/gpu/device/rsqrt.hpp>
+
+namespace migraphx {
+inline namespace MIGRAPHX_INLINE_NS {
+namespace gpu {
+
+struct hip_rsqrt : unary_device<hip_rsqrt, device::rsqrt>
+{
+};
+
+} // namespace gpu
+} // namespace MIGRAPHX_INLINE_NS
+} // namespace migraphx
+
+#endif

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

+#ifndef MIGRAPHX_GUARD_RTGLIB_SQDIFF_HPP
+#define MIGRAPHX_GUARD_RTGLIB_SQDIFF_HPP
+
+#include <migraphx/gpu/oper.hpp>
+#include <migraphx/gpu/device/sqdiff.hpp>
+
+namespace migraphx {
+inline namespace MIGRAPHX_INLINE_NS {
+namespace gpu {
+
+struct hip_sqdiff : binary_device<hip_sqdiff, device::sqdiff>
+{
+};
+
+} // namespace gpu
+} // namespace MIGRAPHX_INLINE_NS
+} // namespace migraphx
+
+#endif

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

+#ifndef MIGRAPHX_GUARD_RTGLIB_SQRT_HPP
+#define MIGRAPHX_GUARD_RTGLIB_SQRT_HPP
+
+#include <migraphx/gpu/oper.hpp>
+#include <migraphx/gpu/device/sqrt.hpp>
+
+namespace migraphx {
+inline namespace MIGRAPHX_INLINE_NS {
+namespace gpu {
+
+struct hip_sqrt : unary_device<hip_sqrt, device::sqrt>
+{
+};
+
+} // namespace gpu
+} // namespace MIGRAPHX_INLINE_NS
+} // namespace migraphx
+
+#endif

src/targets/gpu/lowering.cpp

@@ -26,6 +26,7 @@
 #include <migraphx/gpu/logsoftmax.hpp>
 #include <migraphx/gpu/add.hpp>
 #include <migraphx/gpu/sub.hpp>
+#include <migraphx/gpu/div.hpp>
 #include <migraphx/gpu/exp.hpp>
 #include <migraphx/gpu/erf.hpp>
 #include <migraphx/gpu/log.hpp>
@@ -51,6 +52,11 @@
 #include <migraphx/gpu/convert.hpp>
 #include <migraphx/gpu/clip.hpp>
 #include <migraphx/gpu/reduce_sum.hpp>
+#include <migraphx/gpu/rsqrt.hpp>
+#include <migraphx/gpu/sqrt.hpp>
+#include <migraphx/gpu/reduce_mean.hpp>
+#include <migraphx/gpu/pow.hpp>
+#include <migraphx/gpu/sqdiff.hpp>
 #include <utility>
 #include <functional>
 #include <algorithm>
@@ -97,9 +103,14 @@ struct miopen_apply
         add_generic_op<hip_asin>("asin");
         add_generic_op<hip_acos>("acos");
         add_generic_op<hip_atan>("atan");
+        add_generic_op<hip_sqrt>("sqrt");
         add_generic_op<hip_mul>("mul");
+        add_generic_op<hip_div>("div");
         add_generic_op<hip_max>("max");
         add_generic_op<hip_min>("min");
+        add_generic_op<hip_rsqrt>("rsqrt");
+        add_generic_op<hip_pow>("pow");
+        add_generic_op<hip_sqdiff>("sqdiff");
 
         add_extend_op<miopen_gemm, op::dot>("dot");
         add_extend_op<miopen_contiguous, op::contiguous>("contiguous");
@@ -113,6 +124,7 @@ struct miopen_apply
         add_extend_op<hip_convert, op::convert>("convert");
         add_extend_op<hip_clip, op::clip>("clip");
         add_extend_op<hip_reduce_sum, op::reduce_sum>("reduce_sum");
+        add_extend_op<hip_reduce_mean, op::reduce_mean>("reduce_mean");
 
         add_lrn_op();
         add_convolution_op();

src/targets/gpu/reduce_mean.cpp

+#include <migraphx/gpu/reduce_mean.hpp>
+#include <migraphx/gpu/context.hpp>
+#include <migraphx/gpu/device/reduce_mean.hpp>
+
+namespace migraphx {
+inline namespace MIGRAPHX_INLINE_NS {
+namespace gpu {
+
+shape hip_reduce_mean::compute_shape(std::vector<shape> inputs) const
+{
+    inputs.pop_back();
+    return op.compute_shape(inputs);
+}
+
+argument
+hip_reduce_mean::compute(context& ctx, const shape&, const std::vector<argument>& args) const
+{
+    device::reduce_mean(ctx.get_stream().get(), args.back(), args.front());
+    return args.back();
+}
+
+} // namespace gpu
+} // namespace MIGRAPHX_INLINE_NS
+} // namespace migraphx

test/cpu_ops_test.cpp

@@ -542,6 +542,21 @@ TEST_CASE(erf_test)
     EXPECT(migraphx::verify_range(results_vector, gold));
 }
 
+TEST_CASE(sqrt_test)
+{
+    migraphx::program p;
+    migraphx::shape s{migraphx::shape::float_type, {5}};
+    auto l = p.add_literal(
+        migraphx::literal{s, {1.02481645, 0.85643062, 0.03404123, 0.92791926, 0.10569184}});
+    p.add_instruction(migraphx::op::sqrt{}, l);
+    p.compile(migraphx::cpu::target{});
+    auto result = p.eval({});
+    std::vector<float> results_vector;
+    result.visit([&](auto output) { results_vector.assign(output.begin(), output.end()); });
+    std::vector<float> gold = {1.01233218, 0.92543537, 0.18450265, 0.96328566, 0.32510282};
+    EXPECT(migraphx::verify_range(results_vector, gold));
+}
+
 TEST_CASE(log_test)
 {
     migraphx::program p;
@@ -556,6 +571,21 @@ TEST_CASE(log_test)
     EXPECT(migraphx::verify_range(results_vector, gold));
 }
 
+TEST_CASE(pow_test)
+{
+    migraphx::program p;
+    migraphx::shape s{migraphx::shape::float_type, {3}};
+    auto b = p.add_literal(migraphx::literal{s, {1, 2, 3}});
+    auto e = p.add_literal(migraphx::literal{s, {1, 2, 3}});
+    p.add_instruction(migraphx::op::pow{}, b, e);
+    p.compile(migraphx::cpu::target{});
+    auto result = p.eval({});
+    std::vector<float> results_vector;
+    result.visit([&](auto output) { results_vector.assign(output.begin(), output.end()); });
+    std::vector<float> gold = {1.0f, 4.0f, 27.0f};
+    EXPECT(migraphx::verify_range(results_vector, gold));
+}
+
 TEST_CASE(sin_test)
 {
     migraphx::program p;
@@ -1703,7 +1733,7 @@ TEST_CASE(clip_test)
     EXPECT(migraphx::verify_range(results_vector, gold));
 }
 
-TEST_CASE(reduce_sum_test0)
+TEST_CASE(reduce_sum_axis0)
 {
     migraphx::program p;
     migraphx::shape s{migraphx::shape::float_type, {3, 2, 2}};
@@ -1718,7 +1748,7 @@ TEST_CASE(reduce_sum_test0)
     EXPECT(results_vector == gold);
 }
 
-TEST_CASE(reduce_sum_test1)
+TEST_CASE(reduce_sum_axis1)
 {
     migraphx::program p;
     migraphx::shape s{migraphx::shape::float_type, {3, 2, 2}};
@@ -1733,7 +1763,7 @@ TEST_CASE(reduce_sum_test1)
     EXPECT(results_vector == gold);
 }
 
-TEST_CASE(reduce_sum_test2)
+TEST_CASE(reduce_sum_axis2)
 {
     migraphx::program p;
     migraphx::shape s{migraphx::shape::float_type, {3, 2, 2}};
@@ -1748,7 +1778,7 @@ TEST_CASE(reduce_sum_test2)
     EXPECT(results_vector == gold);
 }
 
-TEST_CASE(reduce_sum_test02)
+TEST_CASE(reduce_sum_axis02)
 {
     migraphx::program p;
     migraphx::shape s{migraphx::shape::float_type, {3, 2, 2}};
@@ -1763,7 +1793,7 @@ TEST_CASE(reduce_sum_test02)
     EXPECT(results_vector == gold);
 }
 
-TEST_CASE(reduce_sum_test12)
+TEST_CASE(reduce_sum_axis12)
 {
     migraphx::program p;
     migraphx::shape s{migraphx::shape::float_type, {3, 2, 2}};
@@ -1778,4 +1808,108 @@ TEST_CASE(reduce_sum_test12)
     EXPECT(results_vector == gold);
 }
 
+TEST_CASE(rsqrt_test)
+{
+    migraphx::program p;
+    migraphx::shape s{migraphx::shape::float_type, {3}};
+    auto l = p.add_literal(migraphx::literal{s, {4.0, 16.0, 64.0}});
+    p.add_instruction(migraphx::op::rsqrt{}, l);
+    p.compile(migraphx::cpu::target{});
+    auto result = p.eval({});
+    std::vector<float> results_vector(3);
+    result.visit([&](auto output) { results_vector.assign(output.begin(), output.end()); });
+    std::vector<float> gold = {0.5, 0.25, 0.125};
+    EXPECT(migraphx::verify_range(results_vector, gold));
+}
+
+TEST_CASE(reduce_mean_axis1)
+{
+    migraphx::program p;
+    migraphx::shape s{migraphx::shape::float_type, {3, 2, 2}};
+    auto input = migraphx::literal{s, {1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12}};
+    auto l0    = p.add_literal(input);
+    p.add_instruction(migraphx::op::reduce_mean{{1}}, l0);
+    p.compile(migraphx::cpu::target{});
+    auto result = p.eval({});
+    std::vector<float> results_vector;
+    result.visit([&](auto output) { results_vector.assign(output.begin(), output.end()); });
+    std::vector<float> gold{2, 3, 6, 7, 10, 11};
+    EXPECT(results_vector == gold);
+}
+
+TEST_CASE(reduce_mean_axis2)
+{
+    migraphx::program p;
+    migraphx::shape s{migraphx::shape::float_type, {3, 2, 2}};
+    auto input = migraphx::literal{s, {1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12}};
+    auto l0    = p.add_literal(input);
+    p.add_instruction(migraphx::op::reduce_mean{{2}}, l0);
+    p.compile(migraphx::cpu::target{});
+    auto result = p.eval({});
+    std::vector<float> results_vector;
+    result.visit([&](auto output) { results_vector.assign(output.begin(), output.end()); });
+    std::vector<float> gold{1.5f, 3.5f, 5.5f, 7.5f, 9.5f, 11.5f};
+    EXPECT(results_vector == gold);
+}
+
+TEST_CASE(reduce_mean_axis02)
+{
+    migraphx::program p;
+    migraphx::shape s{migraphx::shape::float_type, {3, 2, 2}};
+    auto input = migraphx::literal{s, {1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12}};
+    auto l0    = p.add_literal(input);
+    p.add_instruction(migraphx::op::reduce_mean{{0, 2}}, l0);
+    p.compile(migraphx::cpu::target{});
+    auto result = p.eval({});
+    std::vector<float> results_vector;
+    result.visit([&](auto output) { results_vector.assign(output.begin(), output.end()); });
+    std::vector<float> gold{5.5, 7.5};
+    EXPECT(results_vector == gold);
+}
+
+TEST_CASE(reduce_mean_axis12)
+{
+    migraphx::program p;
+    migraphx::shape s{migraphx::shape::float_type, {3, 2, 2}};
+    auto input = migraphx::literal{s, {1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12}};
+    auto l0    = p.add_literal(input);
+    p.add_instruction(migraphx::op::reduce_mean{{1, 2}}, l0);
+    p.compile(migraphx::cpu::target{});
+    auto result = p.eval({});
+    std::vector<float> results_vector;
+    result.visit([&](auto output) { results_vector.assign(output.begin(), output.end()); });
+    std::vector<float> gold{2.5f, 6.5f, 10.5f};
+    EXPECT(results_vector == gold);
+}
+
+TEST_CASE(reduce_mean_int)
+{
+    migraphx::program p;
+    migraphx::shape s{migraphx::shape::int32_type, {3, 2, 2}};
+    auto input = migraphx::literal{s, {1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12}};
+    auto l0    = p.add_literal(input);
+    p.add_instruction(migraphx::op::reduce_mean{{1, 2}}, l0);
+    p.compile(migraphx::cpu::target{});
+    auto result = p.eval({});
+    std::vector<int> results_vector;
+    result.visit([&](auto output) { results_vector.assign(output.begin(), output.end()); });
+    std::vector<int> gold{2, 6, 10};
+    EXPECT(results_vector == gold);
+}
+
+TEST_CASE(sqdiff_test)
+{
+    migraphx::program p;
+    migraphx::shape s{migraphx::shape::float_type, {3}};
+    auto l1 = p.add_literal(migraphx::literal{s, {-1, 0, 1}});
+    auto l2 = p.add_literal(migraphx::literal{s, {1, 2, 3}});
+    p.add_instruction(migraphx::op::sqdiff{}, l1, l2);
+    p.compile(migraphx::cpu::target{});
+    auto result = p.eval({});
+    std::vector<float> results_vector(3);
+    result.visit([&](auto output) { results_vector.assign(output.begin(), output.end()); });
+    std::vector<float> gold = {4, 4, 4};
+    EXPECT(migraphx::verify_range(results_vector, gold));
+}
+
 int main(int argc, const char* argv[]) { test::run(argc, argv); }

test/gpu/miopen.cpp

@@ -255,6 +255,19 @@ struct test_erf : verify_program<test_erf>
     }
 };
 
+struct test_sqrt : verify_program<test_sqrt>
+{
+    migraphx::program create_program() const
+    {
+        migraphx::program p;
+        migraphx::shape s{migraphx::shape::float_type, {2, 3, 4, 6}};
+        auto param     = p.add_parameter("x", s);
+        auto param_abs = p.add_instruction(migraphx::op::abs{}, param);
+        p.add_instruction(migraphx::op::sqrt{}, param_abs);
+        return p;
+    }
+};
+
 struct test_log : verify_program<test_log>
 {
     migraphx::program create_program() const
@@ -267,6 +280,20 @@ struct test_log : verify_program<test_log>
     }
 };
 
+struct test_pow : verify_program<test_pow>
+{
+    migraphx::program create_program() const
+    {
+        migraphx::program p;
+        migraphx::shape s{migraphx::shape::float_type, {6}};
+        std::vector<float> vec_e(s.elements(), 2.0f);
+        auto b = p.add_parameter("x", s);
+        auto e = p.add_literal(migraphx::literal(s, vec_e));
+        p.add_instruction(migraphx::op::pow{}, b, e);
+        return p;
+    }
+};
+
 struct test_sin : verify_program<test_sin>
 {
     migraphx::program create_program() const
@@ -581,6 +608,38 @@ struct test_sub2 : verify_program<test_sub2>
     }
 };
 
+struct test_div : verify_program<test_div>
+{
+    migraphx::program create_program() const
+    {
+        migraphx::program p;
+        migraphx::shape s{migraphx::shape::float_type, {3}};
+        auto x    = p.add_parameter("x", s);
+        auto y    = p.add_parameter("y", s);
+        auto z    = p.add_parameter("z", s);
+        auto diff = p.add_instruction(migraphx::op::div{}, x, y);
+        p.add_instruction(migraphx::op::div{}, diff, z);
+        return p;
+    }
+};
+
+struct test_div2 : verify_program<test_div2>
+{
+    migraphx::program create_program() const
+    {
+        migraphx::program p;
+        migraphx::shape s{migraphx::shape::float_type, {2, 3}};
+        migraphx::shape b{migraphx::shape::float_type, {3}};
+        auto x    = p.add_parameter("x", s);
+        auto y    = p.add_parameter("y", s);
+        auto z    = p.add_parameter("z", b);
+        auto zb   = p.add_instruction(migraphx::op::broadcast{1, s.lens()}, z);
+        auto diff = p.add_instruction(migraphx::op::div{}, x, y);
+        p.add_instruction(migraphx::op::div{}, diff, zb);
+        return p;
+    }
+};
+
 struct test_softmax1 : verify_program<test_softmax1>
 {
     migraphx::program create_program() const
@@ -3511,4 +3570,53 @@ struct test_reduce_sum_half : verify_program<test_reduce_sum_half>
     };
 };
 
+struct test_rsqrt : verify_program<test_rsqrt>
+{
+    migraphx::program create_program() const
+    {
+        migraphx::program p;
+        migraphx::shape s{migraphx::shape::float_type, {1, 3, 16, 16}};
+        auto x  = p.add_parameter("x", s);
+        auto l0 = p.add_instruction(migraphx::op::clip{std::numeric_limits<float>::max(), 1.0}, x);
+        p.add_instruction(migraphx::op::rsqrt{}, l0);
+        return p;
+    };
+};
+
+struct test_reduce_mean : verify_program<test_reduce_mean>
+{
+    migraphx::program create_program() const
+    {
+        migraphx::program p;
+        migraphx::shape s{migraphx::shape::float_type, {3, 9, 4, 3}};
+        auto x = p.add_parameter("x", s);
+        p.add_instruction(migraphx::op::reduce_mean{{1}}, x);
+        return p;
+    };
+};
+
+struct test_reduce_mean_int : verify_program<test_reduce_mean_int>
+{
+    migraphx::program create_program() const
+    {
+        migraphx::program p;
+        migraphx::shape s{migraphx::shape::int32_type, {3, 1024, 8, 8}};
+        auto x = p.add_parameter("x", s);
+        p.add_instruction(migraphx::op::reduce_mean{{1}}, x);
+        return p;
+    };
+};
+
+struct test_reduce_mean_half : verify_program<test_reduce_mean_half>
+{
+    migraphx::program create_program() const
+    {
+        migraphx::program p;
+        migraphx::shape s{migraphx::shape::half_type, {3, 1024, 8, 8}};
+        auto x = p.add_parameter("x", s);
+        p.add_instruction(migraphx::op::reduce_mean{{2}}, x);
+        return p;
+    };
+};
+
 int main(int argc, const char* argv[]) { test::run(argc, argv); }