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

src/targets/gpu/clip.cpp

-/*
- * The MIT License (MIT)
- *
- * Copyright (c) 2015-2022 Advanced Micro Devices, Inc. All rights reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to deal
- * in the Software without restriction, including without limitation the rights
- * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
- * copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
- * THE SOFTWARE.
- */
-#include <migraphx/gpu/clip.hpp>
-#include <migraphx/gpu/context.hpp>
-#include <migraphx/gpu/device/clip.hpp>
-
-namespace migraphx {
-inline namespace MIGRAPHX_INLINE_NS {
-namespace gpu {
-
-shape hip_clip::compute_shape(std::vector<shape> inputs) const
-{
-    inputs.pop_back();
-    return op.compute_shape(inputs);
-}
-
-argument hip_clip::compute(context& ctx, const shape&, const std::vector<argument>& args) const
-{
-    device::clip(ctx.get_stream().get(), args.back(), args.front(), args.at(1), args.at(2));
-    return args.back();
-}
-
-} // namespace gpu
-} // namespace MIGRAPHX_INLINE_NS
-} // namespace migraphx

src/targets/gpu/compile_gen.cpp

@@ -22,6 +22,7 @@
  * THE SOFTWARE.
  */
 #include <migraphx/gpu/compile_gen.hpp>
+#include <migraphx/gpu/context.hpp>
 #include <migraphx/shape.hpp>
 #include <migraphx/permutation.hpp>
 #include <migraphx/stringutils.hpp>
@@ -48,12 +49,13 @@ static std::vector<std::size_t> vector_sizes(const std::vector<shape>& inputs)
     return {4, 2};
 }
 
-vectorize vectorize::elements(std::size_t axis, const std::vector<shape>& inputs)
+vectorize vectorize::elements(std::size_t axis,
+                              const std::vector<shape>& inputs,
+                              const std::vector<std::size_t>& sizes)
 {
     if(std::all_of(
            inputs.begin(), inputs.end(), [&](const auto& s) { return s.lens()[axis] == 1; }))
         return {1, axis};
-    auto sizes = vector_sizes(inputs);
     std::vector<std::size_t> max_vec_size;
     std::transform(inputs.begin(),
                    inputs.end(),
@@ -81,6 +83,33 @@ vectorize vectorize::elements(std::size_t axis, const std::vector<shape>& inputs
     return {*std::min_element(max_vec_size.begin(), max_vec_size.end()), axis};
 }
 
+vectorize vectorize::elements(context& ctx, std::size_t axis, const std::vector<shape>& inputs)
+{
+    if(inputs.empty())
+        return {1, axis};
+    std::size_t n = std::max_element(inputs.begin(),
+                                     inputs.end(),
+                                     by(std::less<>{}, [](const auto& s) { return s.elements(); }))
+                        ->elements();
+    std::size_t max_global = ctx.get_current_device().get_cu_count() *
+                             ctx.get_current_device().get_max_workitems_per_cu();
+    std::size_t over = n / max_global;
+    bool broadcasted =
+        std::any_of(inputs.begin(), inputs.end(), [](const auto& s) { return s.broadcasted(); });
+    std::vector<std::size_t> sizes;
+    if(broadcasted and over > 8)
+        sizes.push_back(8);
+    if(over > 4)
+        sizes.push_back(4);
+    sizes.push_back(2);
+    return elements(axis, inputs, sizes);
+}
+
+vectorize vectorize::elements(std::size_t axis, const std::vector<shape>& inputs)
+{
+    return elements(axis, inputs, vector_sizes(inputs));
+}
+
 std::string vectorize::str() const
 {
     return "vectorize<" + to_string(size) + ", " + to_string(axis) + ">()";
@@ -102,7 +131,7 @@ preload preload::broadcasts(std::size_t axis, const std::vector<shape>& inputs)
     std::size_t bytes = 0;
     for(auto i : preloaded)
     {
-        auto input = inputs[i];
+        const auto& input = inputs[i];
         bytes += input.bytes();
         if(bytes > max_lds_bytes)
             break;

src/targets/gpu/concat.cpp

-/*
- * The MIT License (MIT)
- *
- * Copyright (c) 2015-2022 Advanced Micro Devices, Inc. All rights reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to deal
- * in the Software without restriction, including without limitation the rights
- * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
- * copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
- * THE SOFTWARE.
- */
-#include <migraphx/gpu/concat.hpp>
-#include <migraphx/gpu/context.hpp>
-#include <migraphx/gpu/device/concat.hpp>
-
-namespace migraphx {
-inline namespace MIGRAPHX_INLINE_NS {
-namespace gpu {
-
-shape hip_concat::compute_shape(std::vector<shape> inputs) const
-{
-    inputs.pop_back();
-    return op.normalize_compute_shape(inputs);
-}
-
-argument hip_concat::compute(context& ctx,
-                             const shape& output_shape,
-                             const std::vector<argument>& args) const
-{
-    std::vector<std::size_t> offsets = op.compute_offsets(output_shape, args);
-    return device::concat(ctx.get_stream().get(), output_shape, args, offsets);
-}
-
-} // namespace gpu
-} // namespace MIGRAPHX_INLINE_NS
-} // namespace migraphx

src/targets/gpu/convert.cpp

-/*
- * The MIT License (MIT)
- *
- * Copyright (c) 2015-2022 Advanced Micro Devices, Inc. All rights reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to deal
- * in the Software without restriction, including without limitation the rights
- * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
- * copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
- * THE SOFTWARE.
- */
-#include <migraphx/gpu/convert.hpp>
-#include <migraphx/gpu/context.hpp>
-#include <migraphx/gpu/device/convert.hpp>
-
-namespace migraphx {
-inline namespace MIGRAPHX_INLINE_NS {
-namespace gpu {
-
-shape hip_convert::compute_shape(std::vector<shape> inputs) const
-{
-    inputs.pop_back();
-    check_shapes{inputs, *this}.packed();
-    return op.compute_shape(inputs);
-}
-
-argument hip_convert::compute(context& ctx, const shape&, const std::vector<argument>& args) const
-{
-    device::convert(ctx.get_stream().get(), args[1], args[0]);
-    return args[1];
-}
-
-} // namespace gpu
-} // namespace MIGRAPHX_INLINE_NS
-} // namespace migraphx

src/targets/gpu/convolution.cpp

@@ -24,6 +24,7 @@
 #include <migraphx/gpu/convolution.hpp>
 #include <migraphx/gpu/context.hpp>
 #include <migraphx/generate.hpp>
+#include <miopen/miopen.h>
 
 namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {
@@ -55,14 +56,40 @@ argument miopen_convolution::compute(context& ctx,
                                      const shape& output_shape,
                                      const std::vector<argument>& args) const
 {
-    auto x_desc = make_tensor(reshape_if_1d(args[0].get_shape()));
-    auto w_desc = make_tensor(reshape_if_1d(args[1].get_shape()));
-    auto y_desc = make_tensor(reshape_if_1d(output_shape));
+    auto x_desc                = make_tensor(reshape_if_1d(args[0].get_shape()));
+    auto w_desc                = make_tensor(reshape_if_1d(args[1].get_shape()));
+    auto y_desc                = make_tensor(reshape_if_1d(output_shape));
+    auto* miopen_stream_handle = ctx.get_stream().get_miopen();
+    auto workspace_size        = args[2].get_shape().bytes();
 
+#ifdef MIGRAPHX_HAS_FIND_2_API
+    {
+        const miopenTensorArgument_t tensor_args[3] = {
+            {miopenTensorConvolutionX, nullptr, args[0].implicit()},
+            {miopenTensorConvolutionW, nullptr, args[1].implicit()},
+            {miopenTensorConvolutionY, nullptr, args[3].implicit()},
+        };
+
+        if(solution_ptr.get() == nullptr)
+            MIGRAPHX_THROW("MIOpen Convolution : Load MIOpen Solution before running it");
+
+        auto status = miopenRunSolution(miopen_stream_handle,
+                                        solution_ptr.get(),
+                                        3,
+                                        tensor_args,
+                                        args[2].implicit(),
+                                        workspace_size);
+        if(status != miopenStatusSuccess)
+            MIGRAPHX_THROW("MIOpen Convolution: running convolution using find_2.0 failed");
+
+        return args[3];
+    }
+#else
+    // else use immediate mode
     if(solution_id == 0)
         MIGRAPHX_THROW("MIOpen Convolution: invalid solution ID");
 
-    auto status = miopenConvolutionForwardImmediate(ctx.get_stream().get_miopen(),
+    auto status = miopenConvolutionForwardImmediate(miopen_stream_handle,
                                                     w_desc.get(),
                                                     args[1].implicit(),
                                                     x_desc.get(),
@@ -71,29 +98,66 @@ argument miopen_convolution::compute(context& ctx,
                                                     y_desc.get(),
                                                     args[3].implicit(),
                                                     args[2].implicit(),
-                                                    args[2].get_shape().bytes(),
+                                                    workspace_size,
                                                     solution_id);
 
     if(status != miopenStatusSuccess)
         MIGRAPHX_THROW("MIOpen Convolution: running convolution failed");
     return args[3];
+#endif
 }
 
 shape miopen_convolution::find(context& ctx, const shape& output_shape, std::vector<shape> inputs)
 {
     shape workspace_shape{};
+    auto x_desc                = make_tensor(reshape_if_1d(inputs[0]));
+    auto w_desc                = make_tensor(reshape_if_1d(inputs[1]));
+    auto y_desc                = make_tensor(reshape_if_1d(output_shape));
+    std::size_t workspace_size = 0;
 
-    auto x_desc = make_tensor(reshape_if_1d(inputs[0]));
-    auto w_desc = make_tensor(reshape_if_1d(inputs[1]));
-    auto y_desc = make_tensor(reshape_if_1d(output_shape));
+#ifdef MIGRAPHX_HAS_FIND_2_API
+    {
+        auto conv_problem = make_obj<miopen_problem>(
+            &miopenCreateConvProblem, cd.get(), miopenProblemDirectionForward);
+
+        set_tensor_descriptor(miopenTensorConvolutionX, x_desc, conv_problem);
+        set_tensor_descriptor(miopenTensorConvolutionW, w_desc, conv_problem);
+        set_tensor_descriptor(miopenTensorConvolutionY, y_desc, conv_problem);
+
+        auto* miopen_stream_handle = ctx.get_stream().get_miopen();
+
+        solution_ptr = find_solution(miopen_stream_handle, conv_problem.get());
+
+        auto status = miopenGetSolutionWorkspaceSize(solution_ptr.get(), &workspace_size);
+        if(status != miopenStatusSuccess)
+            MIGRAPHX_THROW("MIOpen Convolution : failed to get solution's workspace size");
+
+        std::size_t solution_size;
+        status = miopenGetSolutionSize(solution_ptr.get(), &solution_size);
+        if(status != miopenStatusSuccess)
+            MIGRAPHX_THROW("MIOpen Convolution: Failed to fetch solution size");
+
+        auto solution_binary = std::vector<char>{};
+        solution_binary.resize(solution_size);
+
+        status = miopenSaveSolution(solution_ptr.get(), solution_binary.data());
+        if(status != miopenStatusSuccess)
+            MIGRAPHX_THROW("MIOpen Convolution: Saving solution failed");
+        solution_object = value::binary{solution_binary.data(), solution_size};
+
+        return shape{shape::int8_type, {workspace_size}};
+    }
+#else
+    // else use immediate find mode
+    auto status = miopenConvolutionForwardGetWorkSpaceSize(ctx.get_stream().get_miopen(),
+                                                           w_desc.get(),
+                                                           x_desc.get(),
+                                                           cd.get(),
+                                                           y_desc.get(),
+                                                           &workspace_size);
+    if(status != miopenStatusSuccess)
+        MIGRAPHX_THROW("MIOpen Convolution: Failed to get forward workspace size");
 
-    std::size_t workspace_size = 0;
-    miopenConvolutionForwardGetWorkSpaceSize(ctx.get_stream().get_miopen(),
-                                             w_desc.get(),
-                                             x_desc.get(),
-                                             cd.get(),
-                                             y_desc.get(),
-                                             &workspace_size);
     workspace_shape = shape{shape::int8_type, {workspace_size}};
 
     auto x         = to_gpu(generate_argument(inputs[0]));
@@ -103,20 +167,20 @@ shape miopen_convolution::find(context& ctx, const shape& output_shape, std::vec
 
     int algo_count = 1;
     miopenConvAlgoPerf_t perf;
-    auto status = miopenFindConvolutionForwardAlgorithm(ctx.get_stream().get_miopen(),
-                                                        x_desc.get(),
-                                                        x.implicit(),
-                                                        w_desc.get(),
-                                                        w.implicit(),
-                                                        cd.get(),
-                                                        y_desc.get(),
-                                                        y.implicit(),
-                                                        1,
-                                                        &algo_count,
-                                                        &perf,
-                                                        workspace.implicit(),
-                                                        workspace_size,
-                                                        false);
+    status = miopenFindConvolutionForwardAlgorithm(ctx.get_stream().get_miopen(),
+                                                   x_desc.get(),
+                                                   x.implicit(),
+                                                   w_desc.get(),
+                                                   w.implicit(),
+                                                   cd.get(),
+                                                   y_desc.get(),
+                                                   y.implicit(),
+                                                   1,
+                                                   &algo_count,
+                                                   &perf,
+                                                   workspace.implicit(),
+                                                   workspace_size,
+                                                   false);
     if(status != miopenStatusSuccess)
         MIGRAPHX_THROW("MIOpen Convolution: find convolution failed");
     algo = perf.fwd_algo;
@@ -148,35 +212,58 @@ shape miopen_convolution::find(context& ctx, const shape& output_shape, std::vec
     solution_id = solutions.front().solution_id;
 
     return shape{shape::int8_type, {perf.memory}};
+#endif
 }
 
 void miopen_convolution::finalize(context& ctx,
                                   const shape& output_shape,
-                                  std::vector<shape> inputs)
+                                  const std::vector<shape>& inputs)
 {
-    if(cd == nullptr)
-        cd = make_conv(op);
-    if(solution_id == 0)
+#ifdef MIGRAPHX_HAS_FIND_2_API
     {
-        // Check that workspace hasn't changed
-        auto size = inputs.at(2).bytes();
-        auto ws   = find(ctx, output_shape, inputs);
-        if(ws.bytes() > size)
-            MIGRAPHX_THROW("MIOpen Convolution: workspace has changed during finalization.");
+        (void)(ctx); // avoid warnings
+        (void)(output_shape);
+        (void)(inputs);
+        // load solution
+        if(solution_ptr == nullptr)
+        {
+            miopenSolution_t ptr;
+            auto status  = miopenLoadSolution(&ptr,
+                                             reinterpret_cast<const char*>(solution_object.data()),
+                                             solution_object.size());
+            solution_ptr = miopen_solution{ptr};
+            if(status != miopenStatusSuccess)
+                MIGRAPHX_THROW("MIOpen Convolution: loading convolution solution failed");
+        }
     }
+#else
+    // Use immediate mode API
+    {
+        if(cd == nullptr)
+            cd = make_conv(op);
+        if(solution_id == 0)
+        {
+            // Check that workspace hasn't changed
+            auto size = inputs.at(2).bytes();
+            auto ws   = find(ctx, output_shape, inputs);
+            if(ws.bytes() > size)
+                MIGRAPHX_THROW("MIOpen Convolution: workspace has changed during finalization.");
+        }
 
-    auto x_desc = make_tensor(reshape_if_1d(inputs[0]));
-    auto w_desc = make_tensor(reshape_if_1d(inputs[1]));
-    auto y_desc = make_tensor(reshape_if_1d(output_shape));
+        auto x_desc = make_tensor(reshape_if_1d(inputs[0]));
+        auto w_desc = make_tensor(reshape_if_1d(inputs[1]));
+        auto y_desc = make_tensor(reshape_if_1d(output_shape));
 
-    auto status = miopenConvolutionForwardCompileSolution(ctx.get_stream().get_miopen(),
-                                                          w_desc.get(),
-                                                          x_desc.get(),
-                                                          cd.get(),
-                                                          y_desc.get(),
-                                                          solution_id);
-    if(status != miopenStatusSuccess)
-        MIGRAPHX_THROW("MIOpen Convolution: compile solution failed");
+        auto status = miopenConvolutionForwardCompileSolution(ctx.get_stream().get_miopen(),
+                                                              w_desc.get(),
+                                                              x_desc.get(),
+                                                              cd.get(),
+                                                              y_desc.get(),
+                                                              solution_id);
+        if(status != miopenStatusSuccess)
+            MIGRAPHX_THROW("MIOpen Convolution: compile solution failed");
+    }
+#endif
 }
 
 } // namespace gpu

src/targets/gpu/device/acos.cpp

-/*
- * The MIT License (MIT)
- *
- * Copyright (c) 2015-2022 Advanced Micro Devices, Inc. All rights reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to deal
- * in the Software without restriction, including without limitation the rights
- * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
- * copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
- * THE SOFTWARE.
- */
-#include <migraphx/gpu/device/acos.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 acos(hipStream_t stream, const argument& result, const argument& arg)
-{
-    nary(stream, result, arg)([](auto x) __device__ { return ::acos(to_hip_type(x)); });
-}
-
-} // namespace device
-} // namespace gpu
-} // namespace MIGRAPHX_INLINE_NS
-} // namespace migraphx

src/targets/gpu/device/acosh.cpp

-/*
- * The MIT License (MIT)
- *
- * Copyright (c) 2015-2022 Advanced Micro Devices, Inc. All rights reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to deal
- * in the Software without restriction, including without limitation the rights
- * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
- * copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
- * THE SOFTWARE.
- */
-#include <migraphx/gpu/device/acosh.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 acosh(hipStream_t stream, const argument& result, const argument& arg)
-{
-    nary(stream, result, arg)([](auto x) { return ::acosh(to_hip_type(x)); });
-}
-
-} // namespace device
-} // namespace gpu
-} // namespace MIGRAPHX_INLINE_NS
-} // namespace migraphx

src/targets/gpu/device/add.cpp

-/*
- * The MIT License (MIT)
- *
- * Copyright (c) 2015-2022 Advanced Micro Devices, Inc. All rights reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to deal
- * in the Software without restriction, including without limitation the rights
- * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
- * copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
- * THE SOFTWARE.
- */
-#include <migraphx/gpu/device/add.hpp>
-#include <migraphx/gpu/device/nary.hpp>
-
-namespace migraphx {
-inline namespace MIGRAPHX_INLINE_NS {
-namespace gpu {
-namespace device {
-
-void add(hipStream_t stream, const argument& result, const argument& arg1, const argument& arg2)
-{
-    nary(stream, result, arg1, arg2)([](auto x, auto y) __device__ { return x + y; });
-}
-
-void add(hipStream_t stream,
-         const argument& result,
-         const argument& arg1,
-         const argument& arg2,
-         const argument& arg3)
-{
-    nary(stream, result, arg1, arg2, arg3)([](auto x, auto y, auto z)
-                                               __device__ { return x + y + z; });
-}
-
-} // namespace device
-} // namespace gpu
-} // namespace MIGRAPHX_INLINE_NS
-} // namespace migraphx

src/targets/gpu/device/add_clip.cpp

-/*
- * The MIT License (MIT)
- *
- * Copyright (c) 2015-2022 Advanced Micro Devices, Inc. All rights reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to deal
- * in the Software without restriction, including without limitation the rights
- * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
- * copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
- * THE SOFTWARE.
- */
-#include <migraphx/gpu/device/add_clip.hpp>
-#include <migraphx/gpu/device/nary.hpp>
-
-namespace migraphx {
-inline namespace MIGRAPHX_INLINE_NS {
-namespace gpu {
-namespace device {
-
-void add_clip(hipStream_t stream,
-              const argument& result,
-              const argument& arg1,
-              const argument& arg2,
-              const argument& min_arg,
-              const argument& max_arg)
-{
-    nary(stream, result, arg1, arg2, min_arg, max_arg)(
-        [](auto x, auto y, auto min, auto max)
-            __device__ { return ::min<decltype(x + y)>(::max<decltype(x)>(min, x + y), max); });
-}
-
-void add_clip(hipStream_t stream,
-              const argument& result,
-              const argument& arg1,
-              const argument& arg2,
-              const argument& arg3,
-              const argument& min_arg,
-              const argument& max_arg)
-{
-    nary(stream, result, arg1, arg2, arg3, min_arg, max_arg)(
-        [](auto x, auto y, auto z, auto min, auto max) __device__ {
-            return ::min<decltype(x + y + z)>(::max<decltype(x)>(min, x + y + z), max);
-        });
-}
-
-} // namespace device
-} // namespace gpu
-} // namespace MIGRAPHX_INLINE_NS
-} // namespace migraphx

src/targets/gpu/device/add_relu.cpp

-/*
- * The MIT License (MIT)
- *
- * Copyright (c) 2015-2022 Advanced Micro Devices, Inc. All rights reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to deal
- * in the Software without restriction, including without limitation the rights
- * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
- * copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
- * THE SOFTWARE.
- */
-#include <migraphx/gpu/device/add_relu.hpp>
-#include <migraphx/gpu/device/nary.hpp>
-
-namespace migraphx {
-inline namespace MIGRAPHX_INLINE_NS {
-namespace gpu {
-namespace device {
-
-void add_relu(hipStream_t stream,
-              const argument& result,
-              const argument& arg1,
-              const argument& arg2)
-{
-    nary(stream, result, arg1, arg2)([](auto x, auto y)
-                                         __device__ { return ::max<decltype(x + y)>(0, x + y); });
-}
-
-void add_relu(hipStream_t stream,
-              const argument& result,
-              const argument& arg1,
-              const argument& arg2,
-              const argument& arg3)
-{
-    nary(stream, result, arg1, arg2, arg3)(
-        [](auto x, auto y, auto z) __device__ { return ::max<decltype(x + y + z)>(0, x + y + z); });
-}
-
-} // namespace device
-} // namespace gpu
-} // namespace MIGRAPHX_INLINE_NS
-} // namespace migraphx

src/targets/gpu/device/add_sigmoid.cpp

-/*
- * The MIT License (MIT)
- *
- * Copyright (c) 2015-2022 Advanced Micro Devices, Inc. All rights reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to deal
- * in the Software without restriction, including without limitation the rights
- * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
- * copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
- * THE SOFTWARE.
- */
-#include <migraphx/gpu/device/add_sigmoid.hpp>
-#include <migraphx/gpu/device/nary.hpp>
-
-namespace migraphx {
-inline namespace MIGRAPHX_INLINE_NS {
-namespace gpu {
-namespace device {
-
-void add_sigmoid(hipStream_t stream,
-                 const argument& result,
-                 const argument& arg1,
-                 const argument& arg2)
-{
-    nary(stream, result, arg1, arg2)(
-        [](auto x, auto y) __device__ { return 1.f / (1.f + ::exp(to_hip_type(-(x + y)))); });
-}
-
-void add_sigmoid(hipStream_t stream,
-                 const argument& result,
-                 const argument& arg1,
-                 const argument& arg2,
-                 const argument& arg3)
-{
-    nary(stream, result, arg1, arg2, arg3)([](auto x, auto y, auto z) __device__ {
-        return 1.f / (1.f + ::exp(to_hip_type(-(x + y + z))));
-    });
-}
-
-} // namespace device
-} // namespace gpu
-} // namespace MIGRAPHX_INLINE_NS
-} // namespace migraphx

src/targets/gpu/device/add_tanh.cpp

-/*
- * The MIT License (MIT)
- *
- * Copyright (c) 2015-2022 Advanced Micro Devices, Inc. All rights reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to deal
- * in the Software without restriction, including without limitation the rights
- * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
- * copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
- * THE SOFTWARE.
- */
-#include <migraphx/gpu/device/add_tanh.hpp>
-#include <migraphx/gpu/device/nary.hpp>
-
-namespace migraphx {
-inline namespace MIGRAPHX_INLINE_NS {
-namespace gpu {
-namespace device {
-
-void add_tanh(hipStream_t stream,
-              const argument& result,
-              const argument& arg1,
-              const argument& arg2)
-{
-    nary(stream, result, arg1, arg2)([](auto x, auto y)
-                                         __device__ { return ::tanh(to_hip_type(x + y)); });
-}
-
-void add_tanh(hipStream_t stream,
-              const argument& result,
-              const argument& arg1,
-              const argument& arg2,
-              const argument& arg3)
-{
-    nary(stream, result, arg1, arg2, arg3)(
-        [](auto x, auto y, auto z) __device__ { return ::tanh(to_hip_type(x + y + z)); });
-}
-
-} // namespace device
-} // namespace gpu
-} // namespace MIGRAPHX_INLINE_NS
-} // namespace migraphx

src/targets/gpu/device/asin.cpp

-/*
- * The MIT License (MIT)
- *
- * Copyright (c) 2015-2022 Advanced Micro Devices, Inc. All rights reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to deal
- * in the Software without restriction, including without limitation the rights
- * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
- * copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
- * THE SOFTWARE.
- */
-#include <migraphx/gpu/device/asin.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 asin(hipStream_t stream, const argument& result, const argument& arg)
-{
-    nary(stream, result, arg)([](auto x) __device__ { return ::asin(to_hip_type(x)); });
-}
-
-} // namespace device
-} // namespace gpu
-} // namespace MIGRAPHX_INLINE_NS
-} // namespace migraphx

src/targets/gpu/device/asinh.cpp

-/*
- * The MIT License (MIT)
- *
- * Copyright (c) 2015-2022 Advanced Micro Devices, Inc. All rights reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to deal
- * in the Software without restriction, including without limitation the rights
- * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
- * copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
- * THE SOFTWARE.
- */
-#include <migraphx/gpu/device/asinh.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 asinh(hipStream_t stream, const argument& result, const argument& arg)
-{
-    nary(stream, result, arg)([](auto x) { return ::asinh(to_hip_type(x)); });
-}
-
-} // namespace device
-} // namespace gpu
-} // namespace MIGRAPHX_INLINE_NS
-} // namespace migraphx

src/targets/gpu/device/atan.cpp

-/*
- * The MIT License (MIT)
- *
- * Copyright (c) 2015-2022 Advanced Micro Devices, Inc. All rights reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to deal
- * in the Software without restriction, including without limitation the rights
- * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
- * copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
- * THE SOFTWARE.
- */
-#include <migraphx/gpu/device/atan.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 atan(hipStream_t stream, const argument& result, const argument& arg)
-{
-    nary(stream, result, arg)([](auto x) __device__ { return ::atan(to_hip_type(x)); });
-}
-
-} // namespace device
-} // namespace gpu
-} // namespace MIGRAPHX_INLINE_NS
-} // namespace migraphx

src/targets/gpu/device/atanh.cpp

-/*
- * The MIT License (MIT)
- *
- * Copyright (c) 2015-2022 Advanced Micro Devices, Inc. All rights reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to deal
- * in the Software without restriction, including without limitation the rights
- * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
- * copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
- * THE SOFTWARE.
- */
-#include <migraphx/gpu/device/atanh.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 atanh(hipStream_t stream, const argument& result, const argument& arg)
-{
-    nary(stream, result, arg)([](auto x) { return ::atanh(to_hip_type(x)); });
-}
-
-} // namespace device
-} // namespace gpu
-} // namespace MIGRAPHX_INLINE_NS
-} // namespace migraphx

src/targets/gpu/device/ceil.cpp

-/*
- * The MIT License (MIT)
- *
- * Copyright (c) 2015-2022 Advanced Micro Devices, Inc. All rights reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to deal
- * in the Software without restriction, including without limitation the rights
- * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
- * copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
- * THE SOFTWARE.
- */
-#include <migraphx/gpu/device/ceil.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 ceil(hipStream_t stream, const argument& result, const argument& arg)
-{
-    nary(stream, result, arg)([](auto x) __device__ { return ::ceil(to_hip_type(x)); });
-}
-
-} // namespace device
-} // namespace gpu
-} // namespace MIGRAPHX_INLINE_NS
-} // namespace migraphx

src/targets/gpu/device/clip.cpp

-/*
- * The MIT License (MIT)
- *
- * Copyright (c) 2015-2022 Advanced Micro Devices, Inc. All rights reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to deal
- * in the Software without restriction, including without limitation the rights
- * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
- * copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
- * THE SOFTWARE.
- */
-#include <migraphx/gpu/device/clip.hpp>
-#include <migraphx/gpu/device/nary.hpp>
-
-namespace migraphx {
-inline namespace MIGRAPHX_INLINE_NS {
-namespace gpu {
-namespace device {
-
-void clip(hipStream_t stream,
-          const argument& result,
-          const argument& arg1,
-          const argument& min_val,
-          const argument& max_val)
-{
-
-    nary(stream, result, arg1, min_val, max_val)([](auto x, auto min, auto max) __device__ {
-        return ::min<decltype(x)>(::max<decltype(x)>(min, x), max);
-    });
-}
-
-} // namespace device
-} // namespace gpu
-} // namespace MIGRAPHX_INLINE_NS
-} // namespace migraphx

src/targets/gpu/device/concat.cpp

-/*
- * The MIT License (MIT)
- *
- * Copyright (c) 2015-2022 Advanced Micro Devices, Inc. All rights reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to deal
- * in the Software without restriction, including without limitation the rights
- * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
- * copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
- * THE SOFTWARE.
- */
-#include <migraphx/shape.hpp>
-#include <migraphx/argument.hpp>
-#include <migraphx/gpu/device/concat.hpp>
-#include <migraphx/gpu/device/contiguous.hpp>
-
-namespace migraphx {
-inline namespace MIGRAPHX_INLINE_NS {
-namespace gpu {
-namespace device {
-
-argument concat(hipStream_t stream,
-                const migraphx::shape&,
-                std::vector<migraphx::argument> args,
-                std::vector<std::size_t> offsets)
-{
-    auto ninputs = args.size() - 1;
-    for(std::size_t j = 0; j < ninputs; j++)
-    {
-        auto&& arg        = args[j];
-        auto offset       = offsets[j];
-        auto byte_offset  = offset * arg.get_shape().type_size();
-        auto output_shape = shape{
-            arg.get_shape().type(), arg.get_shape().lens(), args.back().get_shape().strides()};
-        auto output = argument{output_shape, args.back().data() + byte_offset};
-        contiguous(stream, output, arg);
-    }
-    return args.back();
-}
-
-} // namespace device
-} // namespace gpu
-} // namespace MIGRAPHX_INLINE_NS
-} // namespace migraphx

src/targets/gpu/device/convert.cpp

-/*
- * The MIT License (MIT)
- *
- * Copyright (c) 2015-2022 Advanced Micro Devices, Inc. All rights reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to deal
- * in the Software without restriction, including without limitation the rights
- * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
- * copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
- * THE SOFTWARE.
- */
-#include <migraphx/gpu/device/convert.hpp>
-#include <migraphx/gpu/device/nary.hpp>
-
-namespace migraphx {
-inline namespace MIGRAPHX_INLINE_NS {
-namespace gpu {
-namespace device {
-
-void convert(hipStream_t stream, const argument& result, const argument& arg)
-{
-    result.visit([&](auto output) {
-        arg.visit([&](auto input) {
-            const auto* input_ptr = device_cast(input.data());
-            auto* output_ptr      = device_cast(output.data());
-            gs_launch(stream, result.get_shape().elements())(
-                [=](auto i) __device__ { output_ptr[i] = input_ptr[i]; });
-        });
-    });
-}
-
-} // namespace device
-} // namespace gpu
-} // namespace MIGRAPHX_INLINE_NS
-} // namespace migraphx