merge changes from develop branch

src/include/migraphx/op/atan.hpp

@@ -17,9 +17,12 @@ namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {
 namespace op {
 
-struct atan : unary
+struct atan : unary<atan>
 {
-    std::string name() const { return "atan"; }
+    auto apply() const
+    {
+        return [](auto x) { return std::atan(x); };
+    }
 };
 
 } // namespace op

src/include/migraphx/op/binary.hpp

 #ifndef MIGRAPHX_GUARD_OPERATORS_BINARY_HPP
 #define MIGRAPHX_GUARD_OPERATORS_BINARY_HPP
 
-#include <array>
-#include <migraphx/operation.hpp>
-#include <migraphx/check_shapes.hpp>
-#include <migraphx/stringutils.hpp>
-#include <migraphx/streamutils.hpp>
-#include <migraphx/literal.hpp>
-#include <migraphx/shape_for_each.hpp>
-#include <migraphx/config.hpp>
-#include <cmath>
-#include <utility>
+#include <migraphx/op/name.hpp>
 
 namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {
 namespace op {
 
-struct binary
+template <class Derived>
+struct binary : op_name<Derived>
 {
     shape compute_shape(std::vector<shape> inputs) const
     {
         check_shapes{inputs}.has(2).same_type().same_dims();
-        auto t    = inputs.at(0).type();
-        auto lens = inputs.at(0).lens();
-        return {t, lens};
+        auto s0 = inputs.at(0);
+        auto s1 = inputs.at(1);
+        if(s0 == s1 and s0.packed())
+        {
+            return s0;
+        }
+        else
+        {
+            return {s0.type(), s0.lens()};
+        }
+    }
+    argument compute(const shape& output_shape, std::vector<argument> args) const
+    {
+        argument result{output_shape};
+        visit_all(result, args[0], args[1])([&](auto output, auto input1, auto input2) {
+            if(input1.get_shape().standard() and input2.get_shape().standard())
+            {
+                std::transform(input1.begin(),
+                               input1.end(),
+                               input2.begin(),
+                               output.begin(),
+                               static_cast<const Derived&>(*this).apply());
+            }
+            else
+            {
+                shape_for_each(output.get_shape(), [&](const auto& idx) {
+                    output(idx.begin(), idx.end()) = static_cast<const Derived&>(*this).apply()(
+                        input1(idx.begin(), idx.end()), input2(idx.begin(), idx.end()));
+                });
+            }
+        });
+        return result;
     }
 };
 

src/include/migraphx/op/broadcast.hpp

@@ -27,45 +27,43 @@ namespace op {
 struct broadcast
 {
     uint64_t axis = 0;
+    std::vector<std::size_t> broadcast_lens;
 
     template <class Self, class F>
     static auto reflect(Self& self, F f)
     {
-        return pack(f(self.axis, "axis"));
+        return pack(f(self.axis, "axis"), f(self.broadcast_lens, "dims"));
     }
 
-    shape broadcast_shape;
     std::string name() const { return "broadcast"; }
     shape compute_shape(std::vector<shape> inputs) const
     {
         auto t     = inputs.at(0).type();
         auto input = inputs.at(0);
 
-        std::vector<size_t> bcast_strides(broadcast_shape.lens().size(), 0);
+        std::vector<size_t> bcast_strides(broadcast_lens.size(), 0);
 
-        if(std::all_of(broadcast_shape.lens().cbegin(), broadcast_shape.lens().cend(), [&](auto x) {
-               return x == 1;
-           }))
+        if(std::all_of(
+               broadcast_lens.cbegin(), broadcast_lens.cend(), [&](auto x) { return x == 1; }))
         {
             if(axis != 0)
-                MIGRAPHX_THROW("when broadcasting tensor of size 1, axis should be 0");
-            return {t, broadcast_shape.lens(), std::move(bcast_strides)};
+                MIGRAPHX_THROW("BROADCAST: when broadcasting tensor of size 1, axis should be 0");
+            return {t, broadcast_lens, std::move(bcast_strides)};
         }
         else
         {
-            assert(broadcast_shape.lens().size() - axis >= input.lens().size());
-            if(!std::equal(
-                   input.lens().begin(), input.lens().end(), broadcast_shape.lens().begin() + axis))
-                MIGRAPHX_THROW("when broadcasting success sizes must match");
+            assert(broadcast_lens.size() - axis >= input.lens().size());
+            if(!std::equal(input.lens().begin(), input.lens().end(), broadcast_lens.begin() + axis))
+                MIGRAPHX_THROW("BROADCAST: when broadcasting success sizes must match");
             std::copy(input.strides().begin(), input.strides().end(), bcast_strides.begin() + axis);
-            return {t, broadcast_shape.lens(), std::move(bcast_strides)};
+            return {t, broadcast_lens, std::move(bcast_strides)};
         }
     }
     argument compute(shape output_shape, std::vector<argument> args) const
     {
         return {std::move(output_shape), std::move(args.at(0).data)};
     }
-    int output_alias(const std::vector<shape>&) const { return 0; }
+    std::ptrdiff_t output_alias(const std::vector<shape>&) const { return 0; }
 };
 
 } // namespace op

src/include/migraphx/op/common.hpp

@@ -31,6 +31,8 @@ enum class rnn_direction
     bidirectional,
 };
 
+std::ostream& operator<<(std::ostream& os, rnn_direction v);
+
 } // namespace op
 } // namespace MIGRAPHX_INLINE_NS
 } // namespace migraphx

src/include/migraphx/op/cos.hpp

@@ -17,9 +17,12 @@ namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {
 namespace op {
 
-struct cos : unary
+struct cos : unary<cos>
 {
-    std::string name() const { return "cos"; }
+    auto apply() const
+    {
+        return [](auto x) { return std::cos(x); };
+    }
 };
 
 } // namespace op

src/include/migraphx/op/cosh.hpp

@@ -17,9 +17,12 @@ namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {
 namespace op {
 
-struct cosh : unary
+struct cosh : unary<cosh>
 {
-    std::string name() const { return "cosh"; }
+    auto apply() const
+    {
+        return [](auto x) { return std::cosh(x); };
+    }
 };
 
 } // namespace op

src/include/migraphx/op/div.hpp

@@ -17,9 +17,12 @@ namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {
 namespace op {
 
-struct div : binary
+struct div : binary<div>
 {
-    std::string name() const { return "div"; }
+    auto apply() const
+    {
+        return [](auto x, auto y) { return x / y; };
+    }
 };
 
 } // namespace op

src/include/migraphx/op/exp.hpp

@@ -17,9 +17,12 @@ namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {
 namespace op {
 
-struct exp : unary
+struct exp : unary<exp>
 {
-    std::string name() const { return "exp"; }
+    auto apply() const
+    {
+        return [](auto x) { return std::exp(x); };
+    }
 };
 
 } // namespace op

src/include/migraphx/op/flatten.hpp

@@ -46,7 +46,7 @@ struct flatten
     {
         return {std::move(output_shape), std::move(args.front().data)};
     }
-    int output_alias(const std::vector<shape>&) const { return 0; }
+    std::ptrdiff_t output_alias(const std::vector<shape>&) const { return 0; }
 };
 
 } // namespace op

src/include/migraphx/op/gather.hpp

@@ -19,11 +19,18 @@ namespace op {
 struct gather
 {
     int axis = 0;
+
+    template <class Self, class F>
+    static auto reflect(Self& self, F f)
+    {
+        return pack(f(self.axis, "axis"));
+    }
+
     std::string name() const { return "gather"; }
 
     shape compute_shape(std::vector<shape> inputs) const
     {
-        check_shapes{inputs, *this}.has(2);
+        check_shapes{inputs, *this}.has(2).standard();
         auto lens = inputs[0].lens();
         int n_dim = static_cast<int>(lens.size());
         if(axis >= n_dim || axis < -n_dim)

src/include/migraphx/op/gru.hpp

@@ -27,6 +27,16 @@ struct gru
     float clip              = 0.0f;
     int linear_before_reset = 0;
 
+    template <class Self, class F>
+    static auto reflect(Self& self, F f)
+    {
+        return pack(f(self.hidden_size, "hidden_size"),
+                    f(self.actv_funcs, "actv_func"),
+                    f(self.direction, "direction"),
+                    f(self.clip, "clip"),
+                    f(self.linear_before_reset, "linear_before_reset"));
+    }
+
     std::string name() const { return "gru"; }
     shape compute_shape(std::vector<shape> inputs) const
     {

src/include/migraphx/op/identity.hpp

@@ -24,7 +24,7 @@ struct identity
     {
         return {std::move(output_shape), std::move(args.at(0).data)};
     }
-    int output_alias(const std::vector<shape>&) const { return 0; }
+    std::ptrdiff_t output_alias(const std::vector<shape>&) const { return 0; }
 };
 
 } // namespace op

src/include/migraphx/op/load.hpp

@@ -39,7 +39,7 @@ struct load
             MIGRAPHX_THROW("Load access is out of bounds");
         return {s, args[0].data() + offset};
     }
-    int output_alias(const std::vector<shape>&) const { return 0; }
+    std::ptrdiff_t output_alias(const std::vector<shape>&) const { return 0; }
 
     friend std::ostream& operator<<(std::ostream& os, const load& op)
     {

src/include/migraphx/op/log.hpp

@@ -17,9 +17,12 @@ namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {
 namespace op {
 
-struct log : unary
+struct log : unary<log>
 {
-    std::string name() const { return "log"; }
+    auto apply() const
+    {
+        return [](auto x) { return std::log(x); };
+    }
 };
 
 } // namespace op

src/include/migraphx/op/logsoftmax.hpp

@@ -19,10 +19,17 @@ namespace op {
 struct logsoftmax
 {
     int axis = 1;
+
+    template <class Self, class F>
+    static auto reflect(Self& self, F f)
+    {
+        return pack(f(self.axis, "axis"));
+    }
+
     std::string name() const { return "logsoftmax"; }
     shape compute_shape(std::vector<shape> inputs) const
     {
-        check_shapes{inputs}.has(1);
+        check_shapes{inputs}.has(1).standard();
         if(axis < 0 || axis > inputs[0].lens().size())
         {
             MIGRAPHX_THROW("LogSoftMax: input axis value " + std::to_string(axis) +

src/include/migraphx/op/lstm.hpp

@@ -25,6 +25,15 @@ struct lstm
     float clip              = 0.0f;
     int input_forget        = 0;
 
+    template <class Self, class F>
+    static auto reflect(Self& self, F f)
+    {
+        return pack(f(self.hidden_size, "hidden_size"),
+                    f(self.actv_funcs, "actv_func"),
+                    f(self.direction, "direction"),
+                    f(self.input_forget, "input_forget"));
+    }
+
     std::string name() const { return "lstm"; }
     shape compute_shape(std::vector<shape> inputs) const
     {

src/include/migraphx/op/max.hpp

@@ -17,9 +17,12 @@ namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {
 namespace op {
 
-struct max : binary
+struct max : binary<max>
 {
-    std::string name() const { return "max"; }
+    auto apply() const
+    {
+        return [](auto x, auto y) { return std::max(x, y); };
+    }
 };
 
 } // namespace op

src/include/migraphx/op/min.hpp

@@ -17,9 +17,12 @@ namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {
 namespace op {
 
-struct min : binary
+struct min : binary<min>
 {
-    std::string name() const { return "min"; }
+    auto apply() const
+    {
+        return [](auto x, auto y) { return std::min(x, y); };
+    }
 };
 
 } // namespace op

src/include/migraphx/op/mul.hpp

@@ -17,9 +17,12 @@ namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {
 namespace op {
 
-struct mul : binary
+struct mul : binary<mul>
 {
-    std::string name() const { return "mul"; }
+    auto apply() const
+    {
+        return [](auto x, auto y) { return x * y; };
+    }
 };
 
 } // namespace op

src/include/migraphx/op/multibroadcast.hpp

@@ -42,7 +42,7 @@ struct multibroadcast
 
         std::vector<size_t> bcast_strides(output_lens.size(), 0);
         auto offset = output_lens.size() - input.lens().size();
-        for(int i = input.lens().size() - 1; i >= 0; i--)
+        for(std::ptrdiff_t i = input.lens().size() - 1; i >= 0; i--)
         {
             if(output_lens[i + offset] == input.lens()[i])
             {
@@ -55,7 +55,7 @@ struct multibroadcast
     {
         return {std::move(output_shape), std::move(args.at(0).data)};
     }
-    int output_alias(const std::vector<shape>&) const { return 0; }
+    std::ptrdiff_t output_alias(const std::vector<shape>&) const { return 0; }
 };
 
 } // namespace op