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

.github/workflows/ci.yaml

@@ -268,7 +268,9 @@ jobs:
         lcov --directory . --capture --output-file $(pwd)/coverage.info
         lcov --remove $(pwd)/coverage.info '/usr/*' --output-file $(pwd)/coverage.info
         lcov --list $(pwd)/coverage.info
-        curl -s https://codecov.io/bash | bash
+        curl -Os https://uploader.codecov.io/latest/linux/codecov
+        chmod +x codecov
+        ./codecov -t ${CODECOV_TOKEN}
         echo "Uploaded"
 
   linux-fpga:
@@ -364,5 +366,7 @@ jobs:
     #    lcov --directory . --capture --output-file $(pwd)/coverage.info
     #    lcov --remove $(pwd)/coverage.info '/usr/*' --output-file $(pwd)/coverage.info
     #    lcov --list $(pwd)/coverage.info
-    #    curl -s https://codecov.io/bash | bash
-    #    echo "Uploaded"
\ No newline at end of file
+    #    curl -Os https://uploader.codecov.io/latest/linux/codecov
+    #    chmod +x codecov
+    #    ./codecov -t ${CODECOV_TOKEN}
+    #    echo "Uploaded"

.github/workflows/performance.yaml

@@ -26,6 +26,8 @@ on:
         required: true
         default: '-s'
 
+concurrency: "perftest-${{ github.head_ref ||  github.base_ref || 'schedule' }}"
+
 jobs:
   release:
     uses: rocmsoftwareplatform/migraphx-benchmark/.github/workflows/perf-test.yml@main

CMakeLists.txt

@@ -212,6 +212,7 @@ rocm_enable_cppcheck(
         ConfigurationNotChecked
         unmatchedSuppression
         unusedFunction
+        ctuPointerArith
         noExplicitConstructor
         passedByValue
         unusedStructMember

dev-requirements.txt

@@ -25,6 +25,6 @@ pfultz2/rocm-recipes
 facebook/zstd@v1.4.5 -X subdir -DCMAKE_DIR=build/cmake
 ccache@v4.1
 pcre,pfultz2/pcre@8.45 -H sha256:d6f7182602a775a7d500a0cedca6449af0400c6493951513046d17615ed0bf11
-danmar/cppcheck@2.8 -DHAVE_RULES=1
+danmar/cppcheck@2.9 -DHAVE_RULES=1
 RadeonOpenCompute/rocm-cmake@1ebf7e7bc61bb5e949c171562b421264065230a7 --build
 -f requirements.txt

src/include/migraphx/match/layernorm.hpp

@@ -50,8 +50,8 @@ struct layernorm_matcher
     {
         return f("div")(arg(0)(x_minus_mean()),
 
-                        arg(1)(skip_broadcasts(f("sqrt")(
-                            arg(0)(f("add")(either_arg(0, 1)(variance(), has_value(1e-12f))))))));
+                        arg(1)(skip_broadcasts(f("sqrt")(arg(0)(
+                            f("add")(either_arg(0, 1)(variance(), is_constant().bind("eps"))))))));
     }
 
     auto matcher() const { return layernorm_onnx(); }

src/opt/memory_coloring_impl.cpp

@@ -72,7 +72,7 @@ bool memory_coloring_impl::allocate(interval_ptr interval)
 
     if(conflict_table.find(vn) != conflict_table.end())
     {
-        std::set<int>& vn_set = conflict_table[vn];
+        const std::set<int>& vn_set = conflict_table[vn];
         for(const auto& iter : vn_set)
         {
             live_range* range = live_ranges[iter];
@@ -267,8 +267,8 @@ void memory_coloring_impl::verify()
     {
         for(int i = 0; i < num_of_lives; ++i)
         {
-            live_interval& interval = live_intervals[i];
-            live_range& segment     = interval.segment;
+            const live_interval& interval = live_intervals[i];
+            const live_range& segment     = interval.segment;
 
             if(segment.begin == invalid_offset)
             {
@@ -284,7 +284,7 @@ void memory_coloring_impl::verify()
             int vn = segment.vn;
             if(conflict_table.find(vn) != conflict_table.end())
             {
-                std::set<int>& vn_set = conflict_table[vn];
+                const std::set<int>& vn_set = conflict_table[vn];
                 for(const auto& iter : vn_set)
                 {
                     live_range* range = live_ranges[iter];
@@ -319,8 +319,8 @@ void memory_coloring_impl::dump_intervals()
         {
             std::cout << " segment:" << i;
             std::cout << " =>";
-            std::set<int>& table = conflict_table[i];
-            for(auto& iter : table)
+            const std::set<int>& table = conflict_table[i];
+            for(const auto& iter : table)
             {
                 std::cout << (iter) << ",";
             }
@@ -357,7 +357,7 @@ void live_interval::dump()
     std::cout << "id:" << id;
     segment.dump();
     std::cout << " uses:";
-    for(auto& iter : use_points)
+    for(const auto& iter : use_points)
     {
         std::cout << " " << get_ins_enum(iter) << ",";
     }

src/replace_allocate.cpp

@@ -73,7 +73,7 @@ void insert_submod_allocations(instruction_ref ins, module& mod, const allocatio
         name_shapes.insert(ps.begin(), ps.end());
     }
 
-    for(auto& pn : name_shapes)
+    for(const auto& pn : name_shapes)
     {
         const auto& s = pn.second;
         instruction_ref output{};

src/simplify_algebra.cpp

@@ -57,12 +57,14 @@ auto conv_const_weights()
 
 auto reduction() { return match::name_contains("reduce"); }
 
+// conv(x, w) * a => conv(x, a * w)
 struct find_mul_conv
 {
     auto matcher() const
     {
-        return match::name("mul")(match::either_arg(0, 1)(conv_const_weights().bind("conv"),
-                                                          match::name("broadcast").bind("a")));
+        return match::name("mul")(
+            match::either_arg(0, 1)(conv_const_weights().bind("conv"),
+                                    match::name("broadcast", "multibroadcast").bind("a")));
     }
 
     void apply(module& m, const match::matcher_result& r) const
@@ -72,14 +74,35 @@ struct find_mul_conv
         auto a_ins    = r.instructions["a"];
         auto w_ins    = r.instructions["w"];
 
-        auto broadcast_op = any_cast<op::broadcast>(a_ins->get_operator());
-        if(broadcast_op.axis != 1)
+        const auto& a_input_lens = a_ins->inputs().front()->get_shape().lens();
+
+        std::size_t num_not_one_dims = std::count_if(
+            a_input_lens.cbegin(), a_input_lens.cend(), [](auto dim) { return dim != 1; });
+        if(num_not_one_dims > 1)
+            return;
+
+        // check broadcasted along channels
+        const auto& a_lens    = a_ins->get_shape().lens();
+        const auto& a_strides = a_ins->get_shape().strides();
+
+        auto is_broadcasted_axis = [](auto len, auto stride) { return len == 1 or stride == 0; };
+
+        if(a_strides.at(1) != 1)
             return;
 
+        if(not is_broadcasted_axis(a_lens.front(), a_strides.front()))
+            return;
+
+        if(not std::equal(a_lens.begin() + 2,
+                          a_lens.end(),
+                          a_strides.begin() + 2,
+                          a_strides.end(),
+                          is_broadcasted_axis))
+            return;
+
+        auto sq    = m.insert_instruction(ins, make_op("squeeze"), a_ins->inputs().front());
         auto new_a = m.insert_instruction(
-            ins,
-            make_op("broadcast", {{"axis", 0}, {"out_lens", w_ins->get_shape().lens()}}),
-            a_ins->inputs().front());
+            ins, make_op("broadcast", {{"axis", 0}, {"out_lens", w_ins->get_shape().lens()}}), sq);
         auto new_mul  = m.insert_instruction(ins, make_op("mul"), new_a, w_ins);
         auto new_conv = m.insert_instruction(
             ins, conv_ins->get_operator(), conv_ins->inputs().front(), new_mul);

src/targets/gpu/CMakeLists.txt

@@ -39,81 +39,9 @@ file(GLOB KERNEL_FILES ${CONFIGURE_DEPENDS}
 message(STATUS "KERNEL_FILES: ${KERNEL_FILES}")
 add_embed_library(migraphx_kernels ${KERNEL_FILES})
 
-add_library(migraphx_device
-    device/acos.cpp
-    device/acosh.cpp
-    device/add.cpp
-    device/add_clip.cpp
-    device/add_relu.cpp
-    device/add_sigmoid.cpp
-    device/add_tanh.cpp
-    device/argmax.cpp
-    device/argmin.cpp
-    device/asin.cpp
-    device/asinh.cpp
-    device/atan.cpp
-    device/atanh.cpp
-    device/ceil.cpp
-    device/clip.cpp
-    device/concat.cpp
-    device/contiguous.cpp
-    device/convert.cpp
-    device/cos.cpp
-    device/cosh.cpp
-    device/div.cpp
-    device/equal.cpp
-    device/erf.cpp
-    device/exp.cpp
-    device/fill.cpp
-    device/floor.cpp
-    device/gather.cpp
-    device/gelu.cpp
-    device/greater.cpp
-    device/int8_gemm_pack.cpp
-    device/layernorm.cpp
-    device/less.cpp
-    device/log.cpp
-    device/logical_and.cpp
-    device/logical_or.cpp
-    device/logical_xor.cpp
-    device/logsoftmax.cpp
-    device/max.cpp
-    device/min.cpp
-    device/mul.cpp
-    device/mul_add.cpp
-    device/mul_add_relu.cpp
-    device/multinomial.cpp
-    device/nonzero.cpp
-    device/pad.cpp
-    device/pow.cpp
-    device/prelu.cpp
-    device/prefix_scan_sum.cpp
-    device/recip.cpp
-    device/reduce_max.cpp
-    device/reduce_mean.cpp
-    device/reduce_min.cpp
-    device/reduce_sum.cpp
-    device/reduce_prod.cpp
-    device/relu.cpp
-    device/reverse.cpp
-    device/rnn_variable_seq_lens.cpp
-    device/round.cpp
-    device/rsqrt.cpp
-    device/scatter.cpp
-    device/sigmoid.cpp
-    device/sign.cpp
-    device/sin.cpp
-    device/sinh.cpp
-    device/softmax.cpp
-    device/sqdiff.cpp
-    device/sqrt.cpp
-    device/sub.cpp
-    device/tan.cpp
-    device/tanh.cpp
-    device/topk.cpp
-    device/unary_not.cpp
-    device/where.cpp
-)
+file(GLOB DEVICE_GPU_SRCS ${CONFIGURE_DEPENDS} ${CMAKE_CURRENT_SOURCE_DIR}/device/*.cpp)
+add_library(migraphx_device ${DEVICE_GPU_SRCS})
+
 add_library(compile_for_gpu INTERFACE)
 target_compile_options(compile_for_gpu INTERFACE -std=c++17 -fno-gpu-rdc -Wno-cuda-compat -Wno-unused-command-line-argument -Xclang -fallow-half-arguments-and-returns)
 target_link_libraries(compile_for_gpu INTERFACE hip::device -fno-gpu-rdc -Wno-invalid-command-line-argument -Wno-unused-command-line-argument -Wno-option-ignored)
@@ -151,15 +79,12 @@ add_library(migraphx_gpu
     argmax.cpp
     argmin.cpp
     batch_norm_inference.cpp
-    clip.cpp
     code_object_op.cpp
     compile_ops.cpp
     compile_gen.cpp
     compile_hip.cpp
     compile_hip_code_object.cpp
     compiler.cpp
-    concat.cpp
-    convert.cpp
     convolution.cpp
     deconvolution.cpp
     device_name.cpp
@@ -192,7 +117,6 @@ add_library(migraphx_gpu
     rocblas.cpp
     scatter.cpp
     schedule_model.cpp
-    softmax.cpp
     sync_device.cpp
     target.cpp
     topk.cpp
@@ -207,68 +131,18 @@ function(register_migraphx_gpu_ops PREFIX)
     endforeach()
 endfunction()
 register_migraphx_gpu_ops(hip_
-    acosh
-    acos
-    add
     argmax
     argmin
-    asinh
-    asin
-    atanh
-    atan
-    ceil
-    clip
-    concat
-    convert
-    cosh
-    cos
-    div
-    equal
-    erf
-    exp
-    floor
     gather
-    greater
-    less
-    log
     logsoftmax
-    logical_and
-    logical_or
-    logical_xor
     loop
-    max
-    min
-    mul
     multinomial
     nonzero
     pad
-    pow
-    prelu
     prefix_scan_sum
-    recip
-    reduce_max
-    reduce_mean
-    reduce_min
-    reduce_prod
-    reduce_sum
-    relu
     reverse
-    round
-    rsqrt
     scatter
-    sigmoid
-    sign
-    sinh
-    sin
-    softmax
-    sqdiff
-    sqrt
-    sub
-    tanh
-    tan
     topk
-    unary_not
-    where
 )
 register_migraphx_gpu_ops(miopen_
     abs

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/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/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