aboutsummaryrefslogtreecommitdiff
path: root/src/cpu/kernels/elementwise_unary/generic/neon
diff options
context:
space:
mode:
Diffstat (limited to 'src/cpu/kernels/elementwise_unary/generic/neon')
-rw-r--r--src/cpu/kernels/elementwise_unary/generic/neon/impl.cpp115
-rw-r--r--src/cpu/kernels/elementwise_unary/generic/neon/impl.h79
2 files changed, 117 insertions, 77 deletions
diff --git a/src/cpu/kernels/elementwise_unary/generic/neon/impl.cpp b/src/cpu/kernels/elementwise_unary/generic/neon/impl.cpp
new file mode 100644
index 0000000000..30caa4ebeb
--- /dev/null
+++ b/src/cpu/kernels/elementwise_unary/generic/neon/impl.cpp
@@ -0,0 +1,115 @@
+/*
+ * Copyright (c) 2018-2022 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * 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 "src/cpu/kernels/elementwise_unary/generic/neon/impl.h"
+
+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);
+}
+#if defined(__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) && defined(ENABLE_FP16_KERNELS)
+template void elementwise_op<__fp16>(const ITensor *in, ITensor *out, const Window &window, ElementWiseUnary op);
+#endif //defined(__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) && defined(ENABLE_FP16_KERNELS)
+template void elementwise_op<float>(const ITensor *in, ITensor *out, const Window &window, ElementWiseUnary op);
+template void elementwise_op<int32_t>(const ITensor *in, ITensor *out, const Window &window, ElementWiseUnary op);
+
+} // namespace cpu
+} // namespace arm_compute
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
generated by cgit v1.2.1 (git 2.23.0) at 2024-06-17 14:52:16 +0000