Skip to content

GitLab

Commit 4f63c691 authored by charlie's avatar charlie
Browse files

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

parents f02f5d98 e19f78ae
Hide whitespace changes
Inline Side-by-side
Showing with 25 additions and 616 deletions
src/targets/gpu/include/migraphx/gpu/rsqrt.hpp deleted 100644 → 0
View file @ f02f5d98
/*
* 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.
*/
#ifndef MIGRAPHX_GUARD_RTGLIB_RSQRT_HPP
#define MIGRAPHX_GUARD_RTGLIB_RSQRT_HPP
#include <migraphx/gpu/oper.hpp>
#include <migraphx/gpu/device/rsqrt.hpp>
namespace migraphx {
inline namespace MIGRAPHX_INLINE_NS {
namespace gpu {
struct hip_rsqrt : unary_device<hip_rsqrt, device::rsqrt>
{
};
} // namespace gpu
} // namespace MIGRAPHX_INLINE_NS
} // namespace migraphx
#endif
src/targets/gpu/include/migraphx/gpu/sigmoid.hpp deleted 100644 → 0
View file @ f02f5d98
/*
* 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.
*/
#ifndef MIGRAPHX_GUARD_RTGLIB_SIGMOID_HPP
#define MIGRAPHX_GUARD_RTGLIB_SIGMOID_HPP
#include <migraphx/gpu/oper.hpp>
#include <migraphx/gpu/device/sigmoid.hpp>
namespace migraphx {
inline namespace MIGRAPHX_INLINE_NS {
namespace gpu {
struct hip_sigmoid : unary_device<hip_sigmoid, device::sigmoid>
{
};
} // namespace gpu
} // namespace MIGRAPHX_INLINE_NS
} // namespace migraphx
#endif
src/targets/gpu/include/migraphx/gpu/sign.hpp deleted 100644 → 0
View file @ f02f5d98
/*
* 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.
*/
#ifndef MIGRAPHX_GUARD_RTGLIB_SIGN_HPP
#define MIGRAPHX_GUARD_RTGLIB_SIGN_HPP
#include <migraphx/gpu/oper.hpp>
#include <migraphx/gpu/device/sign.hpp>
namespace migraphx {
inline namespace MIGRAPHX_INLINE_NS {
namespace gpu {
struct hip_sign : unary_device<hip_sign, device::sign>
{
};
} // namespace gpu
} // namespace MIGRAPHX_INLINE_NS
} // namespace migraphx
#endif
src/targets/gpu/include/migraphx/gpu/sin.hpp deleted 100644 → 0
View file @ f02f5d98
/*
* 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.
*/
#ifndef MIGRAPHX_GUARD_RTGLIB_SIN_HPP
#define MIGRAPHX_GUARD_RTGLIB_SIN_HPP
#include <migraphx/gpu/oper.hpp>
#include <migraphx/gpu/device/sin.hpp>
namespace migraphx {
inline namespace MIGRAPHX_INLINE_NS {
namespace gpu {
struct hip_sin : unary_device<hip_sin, device::sin>
{
};
} // namespace gpu
} // namespace MIGRAPHX_INLINE_NS
} // namespace migraphx
#endif
src/targets/gpu/include/migraphx/gpu/sinh.hpp deleted 100644 → 0
View file @ f02f5d98
/*
* 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.
*/
#ifndef MIGRAPHX_GUARD_RTGLIB_SINH_HPP
#define MIGRAPHX_GUARD_RTGLIB_SINH_HPP
#include <migraphx/gpu/oper.hpp>
#include <migraphx/gpu/device/sinh.hpp>
namespace migraphx {
inline namespace MIGRAPHX_INLINE_NS {
namespace gpu {
struct hip_sinh : unary_device<hip_sinh, device::sinh>
{
};
} // namespace gpu
} // namespace MIGRAPHX_INLINE_NS
} // namespace migraphx
#endif
src/targets/gpu/include/migraphx/gpu/softmax.hpp deleted 100644 → 0
View file @ f02f5d98
/*
* 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.
*/
#ifndef MIGRAPHX_GUARD_RTGLIB_SOFTMAX_HPP
#define MIGRAPHX_GUARD_RTGLIB_SOFTMAX_HPP
#include <migraphx/op/softmax.hpp>
#include <migraphx/shape.hpp>
#include <migraphx/reflect.hpp>
#include <migraphx/gpu/context.hpp>
namespace migraphx {
inline namespace MIGRAPHX_INLINE_NS {
namespace gpu {
struct context;
struct hip_softmax
{
op::softmax op;
template <class Self, class F>
static auto reflect(Self& self, F f)
{
return migraphx::reflect(self.op, f);
}
std::string name() const { return "gpu::softmax"; }
shape compute_shape(const std::vector<shape>& inputs) const;
argument
compute(context& ctx, const shape& output_shape, const std::vector<argument>& args) const;
std::ptrdiff_t output_alias(const std::vector<shape>& shapes) const
{
return shapes.size() - 1;
}
};
} // namespace gpu
} // namespace MIGRAPHX_INLINE_NS
} // namespace migraphx
#endif
src/targets/gpu/include/migraphx/gpu/sqdiff.hpp deleted 100644 → 0
View file @ f02f5d98
/*
* 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.
*/
#ifndef MIGRAPHX_GUARD_RTGLIB_SQDIFF_HPP
#define MIGRAPHX_GUARD_RTGLIB_SQDIFF_HPP
#include <migraphx/gpu/oper.hpp>
#include <migraphx/gpu/device/sqdiff.hpp>
namespace migraphx {
inline namespace MIGRAPHX_INLINE_NS {
namespace gpu {
struct hip_sqdiff : binary_device<hip_sqdiff, device::sqdiff>
{
};
} // namespace gpu
} // namespace MIGRAPHX_INLINE_NS
} // namespace migraphx
#endif
src/targets/gpu/include/migraphx/gpu/sqrt.hpp deleted 100644 → 0
View file @ f02f5d98
/*
* 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.
*/
#ifndef MIGRAPHX_GUARD_RTGLIB_SQRT_HPP
#define MIGRAPHX_GUARD_RTGLIB_SQRT_HPP
#include <migraphx/gpu/oper.hpp>
#include <migraphx/gpu/device/sqrt.hpp>
namespace migraphx {
inline namespace MIGRAPHX_INLINE_NS {
namespace gpu {
struct hip_sqrt : unary_device<hip_sqrt, device::sqrt>
{
};
} // namespace gpu
} // namespace MIGRAPHX_INLINE_NS
} // namespace migraphx
#endif
src/targets/gpu/include/migraphx/gpu/sub.hpp deleted 100644 → 0
View file @ f02f5d98
/*
* 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.
*/
#ifndef MIGRAPHX_GUARD_RTGLIB_SUB_HPP
#define MIGRAPHX_GUARD_RTGLIB_SUB_HPP
#include <migraphx/gpu/oper.hpp>
#include <migraphx/gpu/device/sub.hpp>
namespace migraphx {
inline namespace MIGRAPHX_INLINE_NS {
namespace gpu {
struct hip_sub : binary_device<hip_sub, device::sub>
{
};
} // namespace gpu
} // namespace MIGRAPHX_INLINE_NS
} // namespace migraphx
#endif
src/targets/gpu/include/migraphx/gpu/tan.hpp deleted 100644 → 0
View file @ f02f5d98
/*
* 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.
*/
#ifndef MIGRAPHX_GUARD_RTGLIB_TAN_HPP
#define MIGRAPHX_GUARD_RTGLIB_TAN_HPP
#include <migraphx/gpu/oper.hpp>
#include <migraphx/gpu/device/tan.hpp>
namespace migraphx {
inline namespace MIGRAPHX_INLINE_NS {
namespace gpu {
struct hip_tan : unary_device<hip_tan, device::tan>
{
};
} // namespace gpu
} // namespace MIGRAPHX_INLINE_NS
} // namespace migraphx
#endif
src/targets/gpu/include/migraphx/gpu/tanh.hpp deleted 100644 → 0
View file @ f02f5d98
/*
* 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.
*/
#ifndef MIGRAPHX_GUARD_RTGLIB_TANH_HPP
#define MIGRAPHX_GUARD_RTGLIB_TANH_HPP
#include <migraphx/gpu/oper.hpp>
#include <migraphx/gpu/device/tanh.hpp>
namespace migraphx {
inline namespace MIGRAPHX_INLINE_NS {
namespace gpu {
struct hip_tanh : unary_device<hip_tanh, device::tanh>
{
};
} // namespace gpu
} // namespace MIGRAPHX_INLINE_NS
} // namespace migraphx
#endif
src/targets/gpu/include/migraphx/gpu/unary_not.hpp deleted 100644 → 0
View file @ f02f5d98
/*
* 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.
*/
#ifndef MIGRAPHX_GUARD_RTGLIB_UNARY_NOT_HPP
#define MIGRAPHX_GUARD_RTGLIB_UNARY_NOT_HPP
#include <migraphx/gpu/oper.hpp>
#include <migraphx/gpu/device/unary_not.hpp>
namespace migraphx {
inline namespace MIGRAPHX_INLINE_NS {
namespace gpu {
struct hip_unary_not : unary_device<hip_unary_not, device::unary_not>
{
std::string name() const { return "gpu::not"; }
};
} // namespace gpu
} // namespace MIGRAPHX_INLINE_NS
} // namespace migraphx
#endif
src/targets/gpu/include/migraphx/gpu/where.hpp deleted 100644 → 0
View file @ f02f5d98
/*
* 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.
*/
#ifndef MIGRAPHX_GUARD_RTGLIB_WHERE_HPP
#define MIGRAPHX_GUARD_RTGLIB_WHERE_HPP
#include <migraphx/gpu/oper.hpp>
#include <migraphx/gpu/device/where.hpp>
namespace migraphx {
inline namespace MIGRAPHX_INLINE_NS {
namespace gpu {
struct hip_where : ternary_device<hip_where, device::where>
{
shape compute_shape(const std::vector<shape>& inputs) const
{
check_shapes{inputs, *this}.has(4).same_dims();
auto s1 = inputs.at(1);
auto s2 = inputs.at(2);
if(s1 == s2 and s1.packed())
{
return s1;
}
else if(s1.packed() != s2.packed())
{
return s1.packed() ? s1 : s2;
}
else if(s1.broadcasted() != s2.broadcasted())
{
return s1.broadcasted() ? s2.with_lens(s1.lens()) : s1.with_lens(s1.lens());
}
else
{
return {s1.type(), s1.lens()};
}
}
};
} // namespace gpu
} // namespace MIGRAPHX_INLINE_NS
} // namespace migraphx
#endif
src/targets/gpu/jit/concat.cpp
View file @ 4f63c691
...@@ -74,7 +74,7 @@ struct concat_compiler : compiler<concat_compiler> ...@@ -74,7 +74,7 @@ struct concat_compiler : compiler<concat_compiler>
options.output = inputs.back(); options.output = inputs.back();
options.params = "-Wno-float-equal"; options.params = "-Wno-float-equal";
auto axis = find_fast_axis(options.inputs); auto axis = find_fast_axis(options.inputs);
auto vec = vectorize::elements(axis, options.inputs); auto vec = vectorize::elements(ctx, axis, options.inputs);
options.kernel_name = v.get("kernel", "concat_kernel"); options.kernel_name = v.get("kernel", "concat_kernel");
options.set_launch_params( options.set_launch_params(
v, compute_global_for(ctx, get_concat_elements(options.inputs) / vec.size, 256)); v, compute_global_for(ctx, get_concat_elements(options.inputs) / vec.size, 256));
......
src/targets/gpu/jit/gathernd.cpp
View file @ 4f63c691
...@@ -65,7 +65,7 @@ struct gathernd_compiler : compiler<gathernd_compiler> ...@@ -65,7 +65,7 @@ struct gathernd_compiler : compiler<gathernd_compiler>
operation compile_op(context& ctx, const std::vector<shape>& inputs, const value& v) const operation compile_op(context& ctx, const std::vector<shape>& inputs, const value& v) const
{ {
hip_compile_options options; hip_compile_options options;
auto out_s = inputs.back(); const auto& out_s = inputs.back();
options.set_launch_params(v, compute_global_for(ctx, out_s.elements())); options.set_launch_params(v, compute_global_for(ctx, out_s.elements()));
options.inputs = inputs; options.inputs = inputs;
options.output = out_s; options.output = out_s;
......
src/targets/gpu/jit/layernorm.cpp
View file @ 4f63c691
...@@ -50,9 +50,8 @@ ${preamble} ...@@ -50,9 +50,8 @@ ${preamble}
extern "C" { extern "C" {
__global__ void ${kernel}(${params}) __global__ void ${kernel}(${params})
{ {
auto idx = make_index();
transform_args(make_tensors(), rotate_last(), ${transformers})(${args})([](auto... xs) { transform_args(make_tensors(), rotate_last(), ${transformers})(${args})([](auto... xs) {
${layernorm}<${axis}>(${post}, xs...); ${layernorm}<${axis}>(${post}, ${eps}, xs...);
}); });
} }
...@@ -78,9 +77,8 @@ struct layernorm_compiler : compiler<layernorm_compiler> ...@@ -78,9 +77,8 @@ struct layernorm_compiler : compiler<layernorm_compiler>
// Vectorize if the axis is a reduction axis // Vectorize if the axis is a reduction axis
if(axis == faxis) if(axis == faxis)
{ {
vec = vectorize::elements(faxis, inputs); vec = vectorize::elements(ctx, faxis, inputs);
} }
auto preloads = preload::broadcasts(axis, inputs);
auto relements = inputs[0].lens()[axis] / vec.size; auto relements = inputs[0].lens()[axis] / vec.size;
auto nelements = (inputs.back().elements() / inputs[0].lens()[axis]); auto nelements = (inputs.back().elements() / inputs[0].lens()[axis]);
auto block_size = compute_block_size(relements, 256); auto block_size = compute_block_size(relements, 256);
...@@ -90,16 +88,18 @@ struct layernorm_compiler : compiler<layernorm_compiler> ...@@ -90,16 +88,18 @@ struct layernorm_compiler : compiler<layernorm_compiler>
options.output = inputs.back(); options.output = inputs.back();
options.inputs = inputs; options.inputs = inputs;
options.kernel_name = v.get("kernel", "layernorm_kernel"); options.kernel_name = v.get("kernel", "layernorm_kernel");
auto eps = v.get("epsilon", 1e-12f);
auto src = interpolate_string(layernorm_kernel, auto src = interpolate_string(layernorm_kernel,
{{"kernel", options.kernel_name}, {{"kernel", options.kernel_name},
{"params", enum_params(inputs.size(), "void * private_p")}, {"params", enum_params(inputs.size(), "void * private_p")},
{"args", enum_params(inputs.size(), "private_p")}, {"args", enum_params(inputs.size(), "private_p")},
{"transformers", make_transformer_args(preloads, vec)}, {"transformers", make_transformer_args(vec)},
{"post", v.get("post", std::string{"op::id{}"})}, {"post", v.get("post", std::string{"op::id{}"})},
{"preamble", v.get("preamble", std::string{})}, {"preamble", v.get("preamble", std::string{})},
{"layernorm", v.get("layernorm", std::string{"layernorm"})}, {"layernorm", v.get("layernorm", std::string{"layernorm"})},
{"axis", to_string(axis)}}); {"axis", to_string(axis)},
{"eps", to_string(eps)}});
return compile_hip_code_object(src, options); return compile_hip_code_object(src, options);
} }
......
src/targets/gpu/jit/pointwise.cpp
View file @ 4f63c691
...@@ -75,20 +75,16 @@ struct pointwise_compiler : compiler<pointwise_compiler> ...@@ -75,20 +75,16 @@ struct pointwise_compiler : compiler<pointwise_compiler>
options.virtual_inputs = reduce_dims(inputs); options.virtual_inputs = reduce_dims(inputs);
options.params = "-Wno-float-equal"; options.params = "-Wno-float-equal";
auto axis = find_fast_axis(options.virtual_inputs); auto axis = find_fast_axis(options.virtual_inputs);
auto vec = vectorize::elements(axis, options.virtual_inputs); auto vec = vectorize::elements(ctx, axis, options.virtual_inputs);
auto preloads = preload::broadcasts(axis, options.virtual_inputs);
options.kernel_name = v.get("kernel", "kernel"); options.kernel_name = v.get("kernel", "kernel");
options.set_launch_params( options.set_launch_params(
v, v, compute_global_for(ctx, options.output.elements() / vec.size, 256));
compute_global_for(ctx,
options.output.elements() / vec.size,
oversubscribe_if(not preloads.is_preloading())));
auto src = interpolate_string(pointwise_kernel, auto src = interpolate_string(pointwise_kernel,
{{"kernel", options.kernel_name}, {{"kernel", options.kernel_name},
{"params", enum_params(inputs.size(), "void * private_p")}, {"params", enum_params(inputs.size(), "void * private_p")},
{"args", enum_params(inputs.size(), "private_p")}, {"args", enum_params(inputs.size(), "private_p")},
{"lambda", v.at("lambda").to<std::string>()}, {"lambda", v.at("lambda").to<std::string>()},
{"transformers", make_transformer_args(preloads, vec)}, {"transformers", make_transformer_args(vec)},
{"preamble", v.get("preamble", std::string{})}}); {"preamble", v.get("preamble", std::string{})}});
return compile_hip_code_object(src, options); return compile_hip_code_object(src, options);
} }
......
src/targets/gpu/jit/reduce.cpp
View file @ 4f63c691
...@@ -121,7 +121,7 @@ struct reduce_compiler : compiler<reduce_compiler> ...@@ -121,7 +121,7 @@ struct reduce_compiler : compiler<reduce_compiler>
// Vectorize if the axis is a reduction axis // Vectorize if the axis is a reduction axis
if(options.virtual_inputs.back().lens()[faxis] == 1) if(options.virtual_inputs.back().lens()[faxis] == 1)
{ {
vec = vectorize::elements(faxis, options.virtual_inputs); vec = vectorize::elements(ctx, faxis, options.virtual_inputs);
} }
auto relements = get_reduce_elements(options.virtual_inputs) / vec.size; auto relements = get_reduce_elements(options.virtual_inputs) / vec.size;
auto nelements = options.virtual_inputs.back().elements(); auto nelements = options.virtual_inputs.back().elements();
......
src/targets/gpu/jit/softmax.cpp
View file @ 4f63c691
...@@ -69,7 +69,7 @@ struct softmax_compiler : compiler<softmax_compiler> ...@@ -69,7 +69,7 @@ struct softmax_compiler : compiler<softmax_compiler>
// Vectorize if the axis is a reduction axis // Vectorize if the axis is a reduction axis
if(faxis == axis) if(faxis == axis)
{ {
vec = vectorize::elements(faxis, inputs); vec = vectorize::elements(ctx, faxis, inputs);
} }
auto relements = inputs[0].lens()[axis] / vec.size; auto relements = inputs[0].lens()[axis] / vec.size;
auto nelements = (inputs.back().elements() / inputs[0].lens()[axis]); auto nelements = (inputs.back().elements() / inputs[0].lens()[axis]);
......
src/targets/gpu/kernels/include/migraphx/kernels/layernorm.hpp
View file @ 4f63c691
...@@ -43,7 +43,7 @@ template <index_int Axis, ...@@ -43,7 +43,7 @@ template <index_int Axis,
class Input2, class Input2,
class... Inputs> class... Inputs>
__device__ void generic_binary_layernorm( __device__ void generic_binary_layernorm(
F compute, BinOp op, Output output, Input1 input1, Input2 input2, Inputs... inputs) F compute, BinOp op, float eps, Output output, Input1 input1, Input2 input2, Inputs... inputs)
{ {
using reduce_output = reduce::with_axis<Input1, Axis>; using reduce_output = reduce::with_axis<Input1, Axis>;
reduce::block::run<reduce_output>([&](auto, auto r) { reduce::block::run<reduce_output>([&](auto, auto r) {
...@@ -55,32 +55,34 @@ __device__ void generic_binary_layernorm( ...@@ -55,32 +55,34 @@ __device__ void generic_binary_layernorm(
return make_array(x, x * x) * vec_type<value_type>{1.0 / relements}; return make_array(x, x * x) * vec_type<value_type>{1.0 / relements};
})(input1, input2); })(input1, input2);
auto mean_x = means[0]; auto mean_x = means[0];
auto mean_x2 = means[1]; auto mean_x2 = means[1];
auto variance = mean_x2 - (mean_x * mean_x); auto variance = mean_x2 - (mean_x * mean_x);
value_type eps_val = eps; // implicit conversion for eps
r.inner([&](auto& y, auto x1, auto x2, auto... xs) { r.inner([&](auto& y, auto x1, auto x2, auto... xs) {
auto x = op(x1, x2); auto x = op(x1, x2);
auto m = x - mean_x; auto m = x - mean_x;
// m * rsqrt(mean(m ^ 2) + 1e-12)
y = compute(m * rsqrt(variance + value_type{1e-12}), xs...); // m * rsqrt(mean(m ^ 2) + epsilon)
y = compute(m * rsqrt(variance + eps_val), xs...);
})(output, input1, input2, inputs...); })(output, input1, input2, inputs...);
}); });
} }
template <index_int Axis, class F, class Output, class Input, class... Inputs> template <index_int Axis, class F, class Output, class Input, class... Inputs>
__device__ void layernorm(F compute, Output output, Input input, Inputs... inputs) __device__ void layernorm(F compute, float eps, Output output, Input input, Inputs... inputs)
{ {
generic_binary_layernorm<Axis>( generic_binary_layernorm<Axis>(
compute, [](auto x, auto) { return x; }, output, input, input, inputs...); compute, [](auto x, auto) { return x; }, eps, output, input, input, inputs...);
} }
template <index_int Axis, class F, class Output, class Input1, class Input2, class... Inputs> template <index_int Axis, class F, class Output, class Input1, class Input2, class... Inputs>
__device__ void __device__ void
add_layernorm(F compute, Output output, Input1 input1, Input2 input2, Inputs... inputs) add_layernorm(F compute, float eps, Output output, Input1 input1, Input2 input2, Inputs... inputs)
{ {
generic_binary_layernorm<Axis>( generic_binary_layernorm<Axis>(
compute, [](auto x1, auto x2) { return x1 + x2; }, output, input1, input2, inputs...); compute, [](auto x1, auto x2) { return x1 + x2; }, eps, output, input1, input2, inputs...);
} }
} // namespace migraphx } // namespace migraphx
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment