Merge branch 'develop' into tf-transpose

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

@@ -10,7 +10,7 @@ inline namespace MIGRAPHX_INLINE_NS {
 namespace gpu {
 namespace device {
 
-argument logsoftmax(hipStream_t stream, argument result, argument arg, int axis);
+void logsoftmax(hipStream_t stream, const argument& result, const argument& arg, int axis);
 
 } // namespace device
 } // namespace gpu

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

+#ifndef MIGRAPHX_GUARD_RTGLIB_DEVICE_REDUCE_SUM_HPP
+#define MIGRAPHX_GUARD_RTGLIB_DEVICE_REDUCE_SUM_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_sum(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/softmax.hpp

@@ -10,7 +10,7 @@ inline namespace MIGRAPHX_INLINE_NS {
 namespace gpu {
 namespace device {
 
-argument softmax(hipStream_t stream, argument result, argument arg, int axis);
+void softmax(hipStream_t stream, const argument& result, const argument& arg, int axis);
 
 } // namespace device
 } // namespace gpu

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

+#ifndef MIGRAPHX_GUARD_RTGLIB_REDUCE_SUM_HPP
+#define MIGRAPHX_GUARD_RTGLIB_REDUCE_SUM_HPP
+
+#include <migraphx/shape.hpp>
+#include <migraphx/op/reduce_sum.hpp>
+#include <migraphx/reflect.hpp>
+
+namespace migraphx {
+inline namespace MIGRAPHX_INLINE_NS {
+namespace gpu {
+
+struct context;
+
+struct hip_reduce_sum
+{
+    op::reduce_sum 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_sum"; }
+    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/logsoftmax.cpp

@@ -18,7 +18,8 @@ shape hip_logsoftmax::compute_shape(const std::vector<shape>& inputs) const
 argument
 hip_logsoftmax::compute(context& ctx, const shape&, const std::vector<argument>& args) const
 {
-    return device::logsoftmax(ctx.get_stream().get(), args[1], args[0], op.axis);
+    device::logsoftmax(ctx.get_stream().get(), args.back(), args.front(), op.axis);
+    return args.back();
 }
 
 } // namespace gpu

src/targets/gpu/lowering.cpp

@@ -47,6 +47,7 @@
 #include <migraphx/gpu/lrn.hpp>
 #include <migraphx/gpu/convert.hpp>
 #include <migraphx/gpu/clip.hpp>
+#include <migraphx/gpu/reduce_sum.hpp>
 #include <utility>
 #include <functional>
 #include <algorithm>
@@ -105,6 +106,7 @@ struct miopen_apply
         add_extend_op<hip_pad, op::pad>("pad");
         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_lrn_op();
         add_convolution_op();

src/targets/gpu/reduce_sum.cpp

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

src/targets/gpu/softmax.cpp

@@ -39,7 +39,8 @@ shape hip_softmax::compute_shape(const std::vector<shape>& inputs) const
 
 argument hip_softmax::compute(context& ctx, const shape&, const std::vector<argument>& args) const
 {
-    return device::softmax(ctx.get_stream().get(), args[1], args[0], op.axis);
+    device::softmax(ctx.get_stream().get(), args.back(), args.front(), op.axis);
+    return args.back();
 }
 
 } // namespace gpu

test/cpu_ops_test.cpp

@@ -1583,4 +1583,79 @@ TEST_CASE(clip_test)
     EXPECT(migraphx::verify_range(results_vector, gold));
 }
 
+TEST_CASE(reduce_sum_test0)
+{
+    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_sum{{0}}, 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{15, 18, 21, 24};
+    EXPECT(results_vector == gold);
+}
+
+TEST_CASE(reduce_sum_test1)
+{
+    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_sum{{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{4, 6, 12, 14, 20, 22};
+    EXPECT(results_vector == gold);
+}
+
+TEST_CASE(reduce_sum_test2)
+{
+    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_sum{{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{3, 7, 11, 15, 19, 23};
+    EXPECT(results_vector == gold);
+}
+
+TEST_CASE(reduce_sum_test02)
+{
+    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_sum{{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{33, 45};
+    EXPECT(results_vector == gold);
+}
+
+TEST_CASE(reduce_sum_test12)
+{
+    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_sum{{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{10, 26, 42};
+    EXPECT(results_vector == gold);
+}
+
 int main(int argc, const char* argv[]) { test::run(argc, argv); }

test/gpu/miopen.cpp

@@ -592,13 +592,13 @@ struct test_softmax2 : verify_program<test_softmax2>
     }
 };
 
-template <int Axis>
-struct test_softmax : verify_program<test_softmax<Axis>>
+template <int Axis, migraphx::shape::type_t T>
+struct test_softmax : verify_program<test_softmax<Axis, T>>
 {
     migraphx::program create_program() const
     {
         migraphx::program p;
-        migraphx::shape s{migraphx::shape::float_type, {3, 4, 5, 6}};
+        migraphx::shape s{T, {512, 4, 1067, 6}};
         auto param = p.add_parameter("0", s);
         p.add_instruction(migraphx::op::softmax{Axis}, param);
 
@@ -606,10 +606,14 @@ struct test_softmax : verify_program<test_softmax<Axis>>
     }
 };
 
-template struct test_softmax<0>;
-template struct test_softmax<1>;
-template struct test_softmax<2>;
-template struct test_softmax<3>;
+template struct test_softmax<0, migraphx::shape::float_type>;
+template struct test_softmax<2, migraphx::shape::float_type>;
+template struct test_softmax<1, migraphx::shape::double_type>;
+template struct test_softmax<3, migraphx::shape::double_type>;
+template struct test_softmax<0, migraphx::shape::half_type>;
+template struct test_softmax<1, migraphx::shape::half_type>;
+template struct test_softmax<2, migraphx::shape::half_type>;
+template struct test_softmax<3, migraphx::shape::half_type>;
 
 struct test_conv : verify_program<test_conv>
 {
@@ -3345,12 +3349,12 @@ struct test_lstm_bidirct_default_actv2 : verify_program<test_lstm_bidirct_defaul
 };
 
 template <int Axis>
-struct test_logsoftmax : verify_program<test_logsoftmax<Axis>>
+struct test_logsoftmax_1 : verify_program<test_logsoftmax_1<Axis>>
 {
     migraphx::program create_program() const
     {
         migraphx::program p;
-        migraphx::shape s{migraphx::shape::float_type, {3, 4, 5, 6}};
+        migraphx::shape s{migraphx::shape::float_type, {3}};
         auto param = p.add_parameter("0", s);
         p.add_instruction(migraphx::op::logsoftmax{Axis}, param);
 
@@ -3358,18 +3362,15 @@ struct test_logsoftmax : verify_program<test_logsoftmax<Axis>>
     }
 };
 
-template struct test_logsoftmax<0>;
-template struct test_logsoftmax<1>;
-template struct test_logsoftmax<2>;
-template struct test_logsoftmax<3>;
+template struct test_logsoftmax_1<0>;
 
-template <int Axis>
-struct test_logsoftmax_1 : verify_program<test_logsoftmax_1<Axis>>
+template <int Axis, migraphx::shape::type_t T>
+struct test_logsoftmax : verify_program<test_logsoftmax<Axis, T>>
 {
     migraphx::program create_program() const
     {
         migraphx::program p;
-        migraphx::shape s{migraphx::shape::float_type, {3}};
+        migraphx::shape s{T, {10, 4, 2080, 6}};
         auto param = p.add_parameter("0", s);
         p.add_instruction(migraphx::op::logsoftmax{Axis}, param);
 
@@ -3377,7 +3378,16 @@ struct test_logsoftmax_1 : verify_program<test_logsoftmax_1<Axis>>
     }
 };
 
-template struct test_logsoftmax_1<0>;
+template struct test_logsoftmax<0, migraphx::shape::float_type>;
+template struct test_logsoftmax<1, migraphx::shape::float_type>;
+template struct test_logsoftmax<2, migraphx::shape::float_type>;
+template struct test_logsoftmax<3, migraphx::shape::float_type>;
+template struct test_logsoftmax<1, migraphx::shape::double_type>;
+template struct test_logsoftmax<3, migraphx::shape::double_type>;
+template struct test_logsoftmax<1, migraphx::shape::half_type>;
+template struct test_logsoftmax<0, migraphx::shape::half_type>;
+template struct test_logsoftmax<2, migraphx::shape::half_type>;
+template struct test_logsoftmax<3, migraphx::shape::half_type>;
 
 struct test_fp32_fp16_lall : verify_program<test_fp32_fp16_lall>
 {
@@ -3445,4 +3455,40 @@ struct test_fp32_fp16_sub : verify_program<test_fp32_fp16_sub>
     };
 };
 
+struct test_reduce_sum : verify_program<test_reduce_sum>
+{
+    migraphx::program create_program() const
+    {
+        migraphx::program p;
+        migraphx::shape s{migraphx::shape::float_type, {3, 1026, 4, 3}};
+        auto x = p.add_parameter("x", s);
+        p.add_instruction(migraphx::op::reduce_sum{{1}}, x);
+        return p;
+    };
+};
+
+struct test_reduce_sum_int : verify_program<test_reduce_sum_int>
+{
+    migraphx::program create_program() const
+    {
+        migraphx::program p;
+        migraphx::shape s{migraphx::shape::int32_type, {3, 4, 8, 8}};
+        auto x = p.add_parameter("x", s);
+        p.add_instruction(migraphx::op::reduce_sum{{1}}, x);
+        return p;
+    };
+};
+
+struct test_reduce_sum_half : verify_program<test_reduce_sum_half>
+{
+    migraphx::program create_program() const
+    {
+        migraphx::program p;
+        migraphx::shape s{migraphx::shape::half_type, {3, 4, 8, 8}};
+        auto x = p.add_parameter("x", s);
+        p.add_instruction(migraphx::op::reduce_sum{{1}}, x);
+        return p;
+    };
+};
+
 int main(int argc, const char* argv[]) { test::run(argc, argv); }

test/onnx/implicit_bcast_test.onnx

-implicit_bcast-example:q
-
+add2:u
+
 
0
-
1
2"Addtest-multi_bcastZ
+
1out"Addsubtraction2Z
 
0
 

 
 
 
-Z
-
1
-

+Z
+
1
+

 
-b
-
2
+
+
b
+out
 

 
 
 
-B
+B
\ No newline at end of file

test/onnx/implicit_sub_bcast_test.onnx

-subtraction2:q
+add2:q
 
 
0
 
1out"Subsubtraction2Z
@@ -10,11 +10,11 @@
 Z
 
1
 

-
-b
+
+b
 out
 

 
 
 
-B
+B
\ No newline at end of file

test/onnx/onnx_test.cpp

@@ -350,7 +350,7 @@ TEST_CASE(implicit_add_bcast_test)
 {
     migraphx::program p;
     auto l0 = p.add_parameter("0", migraphx::shape{migraphx::shape::float_type, {2, 3, 4, 5}});
-    auto l1 = p.add_parameter("1", migraphx::shape{migraphx::shape::float_type, {3, 4}});
+    auto l1 = p.add_parameter("1", migraphx::shape{migraphx::shape::float_type, {3, 4, 1}});
     auto l2 = p.add_instruction(migraphx::op::multibroadcast{{2, 3, 4, 5}}, l0);
     auto l3 = p.add_instruction(migraphx::op::multibroadcast{{2, 3, 4, 5}}, l1);
     p.add_instruction(migraphx::op::add{}, l2, l3);
@@ -377,7 +377,7 @@ TEST_CASE(implicit_sub_bcast_test)
 {
     migraphx::program p;
     auto l0 = p.add_parameter("0", migraphx::shape{migraphx::shape::float_type, {2, 3, 4, 5}});
-    auto l1 = p.add_parameter("1", migraphx::shape{migraphx::shape::float_type, {3, 4}});
+    auto l1 = p.add_parameter("1", migraphx::shape{migraphx::shape::float_type, {4, 5}});
     auto l2 = p.add_instruction(migraphx::op::multibroadcast{{2, 3, 4, 5}}, l0);
     auto l3 = p.add_instruction(migraphx::op::multibroadcast{{2, 3, 4, 5}}, l1);
     p.add_instruction(migraphx::op::sub{}, l2, l3);

test/op_shape_test.cpp

@@ -227,6 +227,16 @@ TEST_CASE(multibroadcast)
         migraphx::shape input{migraphx::shape::float_type, {}};
         throws_shape(migraphx::op::multibroadcast{lens}, input);
     }
+    {
+        std::vector<std::size_t> lens{2, 3, 4, 5};
+        migraphx::shape input{migraphx::shape::float_type, {3, 4}};
+        throws_shape(migraphx::op::multibroadcast{lens}, input);
+    }
+    {
+        std::vector<std::size_t> lens{2, 3, 4, 5};
+        migraphx::shape input{migraphx::shape::float_type, {2, 3, 4}};
+        throws_shape(migraphx::op::multibroadcast{lens}, input);
+    }
 }
 
 TEST_CASE(broadcast)