arm_compute v18.02

[platform/upstream/armcl.git] / src / core / NEON / kernels / NEBitwiseAndKernel.cpp

diff --git a/src/core/NEON/kernels/NEBitwiseAndKernel.cpp b/src/core/NEON/kernels/NEBitwiseAndKernel.cpp
index f6eca89..c1e3e1f 100644 (file)
--- a/src/core/NEON/kernels/NEBitwiseAndKernel.cpp
+++ b/src/core/NEON/kernels/NEBitwiseAndKernel.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016, 2017 ARM Limited.
+ * Copyright (c) 2016-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -25,6 +25,7 @@
 
 #include "arm_compute/core/Helpers.h"
 #include "arm_compute/core/ITensor.h"
+#include "arm_compute/core/NEON/wrapper/wrapper.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/core/Validate.h"
 
@@ -32,6 +33,7 @@
 #include <cstdint>
 
 using namespace arm_compute;
+using namespace arm_compute::wrapper;
 
 namespace arm_compute
 {
@@ -40,12 +42,14 @@ class Coordinates;
 
 namespace
 {
-inline void bitwise_and_U8_U8_U8(const uint8_t *__restrict input1, const uint8_t *__restrict input2, uint8_t *__restrict output)
+template <typename T, int S>
+inline void bitwise_and(const T *__restrict input1, const T *__restrict input2, T *__restrict output)
 {
-    const uint8x16_t val1 = vld1q_u8(input1);
-    const uint8x16_t val2 = vld1q_u8(input2);
+    using type      = typename wrapper::traits::neon_vector<T, S>::type;
+    const type val1 = vloadq(static_cast<const T *>(input1));
+    const type val2 = vloadq(static_cast<const T *>(input2));
 
-    vst1q_u8(output, vandq_u8(val1, val2));
+    vstore(static_cast<T *>(output), vand(val1, val2));
 }
 } // namespace
 
@@ -56,23 +60,33 @@ NEBitwiseAndKernel::NEBitwiseAndKernel()
 
 void NEBitwiseAndKernel::configure(const ITensor *input1, const ITensor *input2, ITensor *output)
 {
+    ARM_COMPUTE_ERROR_ON_NULLPTR(input1, input2, output);
+
+    set_shape_if_empty(*output->info(), input1->info()->tensor_shape());
+
+    set_format_if_unknown(*output->info(), Format::U8);
+    set_format_if_unknown(*input1->info(), Format::U8);
+    set_format_if_unknown(*input2->info(), Format::U8);
+
+    ARM_COMPUTE_ERROR_ON_MISMATCHING_SHAPES(input1, input2, output);
     ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input1, 1, DataType::U8);
     ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input2, 1, DataType::U8);
     ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::U8);
+    ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(input1, input2, output);
 
     _input1 = input1;
     _input2 = input2;
     _output = output;
 
-    const unsigned int processed_elements = 16;
+    constexpr unsigned int num_elems_processed_per_iteration = 16;
 
     // Configure kernel window
-    Window                 win = calculate_max_window(*input1->info(), Steps(processed_elements));
-    AccessWindowHorizontal output_access(output->info(), 0, processed_elements);
+    Window                 win = calculate_max_window(*input1->info(), Steps(num_elems_processed_per_iteration));
+    AccessWindowHorizontal output_access(output->info(), 0, num_elems_processed_per_iteration);
 
     update_window_and_padding(win,
-                              AccessWindowHorizontal(input1->info(), 0, processed_elements),
-                              AccessWindowHorizontal(input2->info(), 0, processed_elements),
+                              AccessWindowHorizontal(input1->info(), 0, num_elems_processed_per_iteration),
+                              AccessWindowHorizontal(input2->info(), 0, num_elems_processed_per_iteration),
                               output_access);
 
     const ValidRegion valid_region = intersect_valid_regions(input1->info()->valid_region(),
@@ -83,8 +97,9 @@ void NEBitwiseAndKernel::configure(const ITensor *input1, const ITensor *input2,
     INEKernel::configure(win);
 }
 
-void NEBitwiseAndKernel::run(const Window &window)
+void NEBitwiseAndKernel::run(const Window &window, const ThreadInfo &info)
 {
+    ARM_COMPUTE_UNUSED(info);
     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
     ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(INEKernel::window(), window);
     Iterator input1(_input1, window);
@@ -93,7 +108,7 @@ void NEBitwiseAndKernel::run(const Window &window)
 
     execute_window_loop(window, [&](const Coordinates & id)
     {
-        bitwise_and_U8_U8_U8(input1.ptr(), input2.ptr(), output.ptr());
+        bitwise_and<uint8_t, 16>(input1.ptr(), input2.ptr(), output.ptr());
     },
     input1, input2, output);
 }