aboutsummaryrefslogtreecommitdiff
path: root/src/cpu/kernels/elementwise_unary/generic/neon/impl.h
diff options
context:
space:
mode:
authorDana Zlotnik <dana.zlotnik@arm.com>2022-02-21 13:12:41 +0200
committerDana Zlotnik <dana.zlotnik@arm.com>2022-03-01 11:19:23 +0000
commita538ae583c8816f69d05b98c62a9d3092f88f798 (patch)
tree33ffb4611e9d660223b50d805ac60babbaba4ebb /src/cpu/kernels/elementwise_unary/generic/neon/impl.h
parentee9050089e391e598cd58e05bc7a07597a6d1db0 (diff)
downloadComputeLibrary-a538ae583c8816f69d05b98c62a9d3092f88f798.tar.gz
Multi ISA Technical Debt
* Update json struct meet multi-ISA updates * Add impl.cpp in kernels where we only have impl.h Resolves COMPMID-5173 Change-Id: I5da3c4b016a5d0115c4ba46cbfefde7bce518ac1 Signed-off-by: Dana Zlotnik <dana.zlotnik@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/7191 Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Giorgio Arena <giorgio.arena@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/cpu/kernels/elementwise_unary/generic/neon/impl.h')
-rw-r--r--src/cpu/kernels/elementwise_unary/generic/neon/impl.h79
1 files changed, 2 insertions, 77 deletions
diff --git a/src/cpu/kernels/elementwise_unary/generic/neon/impl.h b/src/cpu/kernels/elementwise_unary/generic/neon/impl.h
index fd930ae7cf..66b8b5fe45 100644
--- a/src/cpu/kernels/elementwise_unary/generic/neon/impl.h
+++ b/src/cpu/kernels/elementwise_unary/generic/neon/impl.h
@@ -24,6 +24,7 @@
#ifndef SRC_CORE_NEON_KERNELS_ELEMENTWISE_UNARY_LIST_H
#define SRC_CORE_NEON_KERNELS_ELEMENTWISE_UNARY_LIST_H
+#include "arm_compute/core/Helpers.h"
#include "arm_compute/core/Types.h"
#include "src/core/NEON/wrapper/intrinsics/intrinsics.h"
@@ -32,83 +33,7 @@ namespace arm_compute
namespace cpu
{
template <typename ScalarType>
-inline ScalarType elementwise_op_scalar_imp(ElementWiseUnary op, const ScalarType &a)
-{
- switch(op)
- {
- case ElementWiseUnary::RSQRT:
- return 1 / sqrt(a);
- case ElementWiseUnary::EXP:
- return std::exp(a);
- case ElementWiseUnary::NEG:
- return -a;
- case ElementWiseUnary::LOG:
- return std::log(a);
- case ElementWiseUnary::ABS:
- return std::abs(a);
- case ElementWiseUnary::ROUND:
- return support::cpp11::nearbyint(a);
- case ElementWiseUnary::SIN:
- return std::sin(a);
- default:
- ARM_COMPUTE_ERROR("NOT_SUPPORTED!");
- }
-}
-
-template <typename ScalarType, typename VectorType>
-inline VectorType elementwise_op_imp(ElementWiseUnary op, const VectorType &a)
-{
- switch(op)
- {
- case ElementWiseUnary::RSQRT:
- return wrapper::vinvsqrt(a);
- case ElementWiseUnary::EXP:
- return wrapper::vexpq(a);
- case ElementWiseUnary::NEG:
- return wrapper::vneg(a);
- case ElementWiseUnary::LOG:
- return wrapper::vlog(a);
- case ElementWiseUnary::ABS:
- return wrapper::vabs(a);
- case ElementWiseUnary::ROUND:
- return wrapper::vround(a);
- case ElementWiseUnary::SIN:
- return wrapper::vsin(a);
- default:
- ARM_COMPUTE_ERROR("NOT_SUPPORTED!");
- }
-}
-
-template <typename ScalarType>
-void elementwise_op(const ITensor *in, ITensor *out, const Window &window, ElementWiseUnary op)
-{
- const int window_step_x = 16 / sizeof(ScalarType);
- const auto window_start_x = static_cast<int>(window.x().start());
- const auto window_end_x = static_cast<int>(window.x().end());
-
- Window win = window;
- win.set(Window::DimX, Window::Dimension(0, 1, 1));
-
- Iterator input(in, win);
- Iterator output(out, win);
-
- execute_window_loop(win, [&](const Coordinates &)
- {
- auto output_ptr = reinterpret_cast<ScalarType *>(output.ptr());
- const auto input_ptr = reinterpret_cast<const ScalarType *>(input.ptr());
-
- int x = window_start_x;
- for(; x <= window_end_x - window_step_x; x += window_step_x)
- {
- wrapper::vstore(output_ptr + x, elementwise_op_imp<ScalarType>(op, wrapper::vloadq(input_ptr + x)));
- }
- for(; x < window_end_x; ++x)
- {
- *(output_ptr + x) = elementwise_op_scalar_imp(op, *(input_ptr + x));
- }
- },
- input, output);
-}
+void elementwise_op(const ITensor *in, ITensor *out, const Window &window, ElementWiseUnary op);
} // namespace cpu
} // namespace arm_compute