diff --git a/ARM.CMSIS-NN.pdsc b/ARM.CMSIS-NN.pdsc index 48cc75c..e967fcc 100644 --- a/ARM.CMSIS-NN.pdsc +++ b/ARM.CMSIS-NN.pdsc @@ -149,6 +149,7 @@ + diff --git a/Include/arm_nnfunctions.h b/Include/arm_nnfunctions.h index 26b6142..2d89094 100644 --- a/Include/arm_nnfunctions.h +++ b/Include/arm_nnfunctions.h @@ -21,8 +21,8 @@ * Title: arm_nnfunctions.h * Description: Public header file for CMSIS NN Library * - * $Date: 08 October 2024 - * $Revision: V.17.1.0 + * $Date: 17 October 2024 + * $Revision: V.17.2.0 * * Target : Arm(R) M-Profile Architecture * -------------------------------------------------------------------- */ @@ -2780,6 +2780,31 @@ arm_cmsis_nn_status arm_batch_matmul_s16(const cmsis_nn_context *ctx, const cmsis_nn_dims *output_dims, int16_t *output); +/** + * @defgroup Pad Pad Layer Functions: + * + */ + +/** + * @brief Expands the size of the input by adding constant values before and after the data, in all dimensions. + * + * @param[in] input Pointer to input data + * @param[out] output Pointer to output data + * @param[in] pad_value Value to pad with + * @param[in] input_size Input tensor dimensions + * @param[in] pre_pad Padding to apply before data in each dimension + * @param[in] post_pad Padding to apply after data in each dimension + * + * @return The function returns ARM_CMSIS_NN_SUCCESS + * + */ +arm_cmsis_nn_status arm_pad_s8(const int8_t *input, + int8_t *output, + const int8_t pad_value, + const cmsis_nn_dims *input_size, + const cmsis_nn_dims *pre_pad, + const cmsis_nn_dims *post_pad); + /** * @brief Elementwise binary minimum with 8bit data. * diff --git a/README.md b/README.md index eb059b0..cba13fd 100644 --- a/README.md +++ b/README.md @@ -40,6 +40,7 @@ Examples are Cortex-M55 or Cortex-M85 configured with MVE. | Softmax | Yes | Yes | N/A | Yes | Yes | N/A | Yes | No | N/A | | LSTM | Yes | Yes | No | Yes | Yes | No | Yes | Yes | No | | SVDF | Yes | No | No | Yes | No | No | Yes | No | No | +| Pad | Yes | No | N/A | No | No | N/A | Yes | No | N/A | * int4 weights + int8 activations @@ -91,7 +92,7 @@ cmake .. -DCMAKE_TOOLCHAIN_FILE=/cmake/toolchain ``` ### Compiler Options -Default optimization level is set at Ofast. This can be overwritten with CMake on command line by using *"-DCMSIS_OPTIMIZATION_LEVEL"*. Please change according to project needs. +Default optimization level is set at Ofast. This can be overwritten with CMake on command line by using *"-DCMSIS_OPTIMIZATION_LEVEL"*. Please change according to project needs. Just bear in mind this can impact performance. With only optimization level -O0, *ARM_MATH_AUTOVECTORIZE* needs to be defined for processors with Helium Technology. diff --git a/Source/CMakeLists.txt b/Source/CMakeLists.txt index a12f385..440aa61 100644 --- a/Source/CMakeLists.txt +++ b/Source/CMakeLists.txt @@ -32,6 +32,7 @@ option(BASICMATHSNN "Basic Maths for NN" ON) option(RESHAPE "Reshape" ON) option(SVDF "SVDF" ON) option(LSTM "LSTM" ON) +option(PAD "Pad" ON) # Always needed if any other module above is on. option(NNSUPPORT "NN Support" ON) @@ -81,6 +82,10 @@ if (RESHAPE) add_subdirectory(ReshapeFunctions) endif() +if (PAD) + add_subdirectory(PadFunctions) +endif() + # Keep NNSUPPORT at the end if (NNSUPPORT) add_subdirectory(NNSupportFunctions) diff --git a/Source/PadFunctions/CMakeLists.txt b/Source/PadFunctions/CMakeLists.txt new file mode 100644 index 0000000..5da8af9 --- /dev/null +++ b/Source/PadFunctions/CMakeLists.txt @@ -0,0 +1,20 @@ +# +# SPDX-FileCopyrightText: Copyright 2010-2024 Arm Limited and/or its affiliates +# +# SPDX-License-Identifier: Apache-2.0 +# +# Licensed under the Apache License, Version 2.0 (the License); you may +# not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an AS IS BASIS, WITHOUT +# WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. +# + +file(GLOB SRC "./*_s8.c") +target_sources(cmsis-nn PRIVATE ${SRC}) diff --git a/Source/PadFunctions/arm_pad_s8.c b/Source/PadFunctions/arm_pad_s8.c new file mode 100644 index 0000000..5f71ae2 --- /dev/null +++ b/Source/PadFunctions/arm_pad_s8.c @@ -0,0 +1,117 @@ + +/* + * SPDX-FileCopyrightText: Copyright 2024 Arm Limited and/or its affiliates + * + * SPDX-License-Identifier: Apache-2.0 + * + * Licensed under the Apache License, Version 2.0 (the License); you may + * not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an AS IS BASIS, WITHOUT + * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +/* ---------------------------------------------------------------------- + * Project: CMSIS NN Library + * Title: arm_pad_s8.c + * Description: Pad a s8 vector + * + * $Date: 19 Sep 2024 + * $Revision: V.1.0.0 + * + * Target : Arm(R) M-Profile Architecture + * + * -------------------------------------------------------------------- */ + +#include "arm_nn_types.h" +#include "arm_nnfunctions.h" +#include "arm_nnsupportfunctions.h" +/** + * @ingroup Public + */ + +/** + * @addtogroup Pad + * @{ + */ + +/* + * Basic s8 pad function. + * + * Refer header file for details. + * + */ + +arm_cmsis_nn_status arm_pad_s8(const int8_t *input, + int8_t *output, + const int8_t pad_value, + const cmsis_nn_dims *input_size, + const cmsis_nn_dims *pre_pad, + const cmsis_nn_dims *post_pad) +{ + + const cmsis_nn_dims output_size = {pre_pad->n + input_size->n + post_pad->n, + pre_pad->h + input_size->h + post_pad->h, + pre_pad->w + input_size->w + post_pad->w, + pre_pad->c + input_size->c + post_pad->c}; + + const int32_t batch_block_size = output_size.h * output_size.w * output_size.c; + const int32_t row_block_size = output_size.w * output_size.c; + const int32_t col_block_size = output_size.c; + + arm_memset_s8(output, pad_value, batch_block_size * pre_pad->n); + output += batch_block_size * pre_pad->n; + for (int32_t b = 0; b < input_size->n; b++) + { + + arm_memset_s8(output, pad_value, row_block_size * pre_pad->h); + output += row_block_size * pre_pad->h; + for (int32_t y = 0; y < input_size->h; y++) + { + + arm_memset_s8(output, pad_value, col_block_size * pre_pad->w); + output += col_block_size * pre_pad->w; + if (input_size->c == output_size.c) + { + arm_memcpy_s8(output, input, input_size->w * input_size->c); + output += input_size->w * input_size->c; + input += input_size->w * input_size->c; + } + else + { + for (int32_t x = 0; x < input_size->w; x++) + { + + arm_memset_s8(output, pad_value, pre_pad->c); + output += pre_pad->c; + + arm_memcpy_s8(output, input, input_size->c); + output += input_size->c; + input += input_size->c; + + arm_memset_s8(output, pad_value, post_pad->c); + output += post_pad->c; + } + } + + arm_memset_s8(output, pad_value, col_block_size * post_pad->w); + output += col_block_size * post_pad->w; + } + + arm_memset_s8(output, pad_value, row_block_size * post_pad->h); + output += row_block_size * post_pad->h; + } + arm_memset_s8(output, pad_value, batch_block_size * post_pad->n); + + return ARM_CMSIS_NN_SUCCESS; +} + +/** + * @} end of Pad group + */ diff --git a/Tests/UnitTest/CMakeLists.txt b/Tests/UnitTest/CMakeLists.txt index a333bcc..495d17d 100644 --- a/Tests/UnitTest/CMakeLists.txt +++ b/Tests/UnitTest/CMakeLists.txt @@ -109,6 +109,7 @@ add_subdirectory(TestCases/test_arm_transpose_conv_s8) add_subdirectory(TestCases/test_arm_lstm_unidirectional_s16) add_subdirectory(TestCases/test_arm_batch_matmul_s8) add_subdirectory(TestCases/test_arm_batch_matmul_s16) +add_subdirectory(TestCases/test_arm_pad_s8) set(MAKE_CMD "python3") set(MAKE_CMD_SCRIPT "${CMAKE_CURRENT_SOURCE_DIR}/unittest_targets.py") diff --git a/Tests/UnitTest/README.md b/Tests/UnitTest/README.md index 01ff3eb..427fbfa 100644 --- a/Tests/UnitTest/README.md +++ b/Tests/UnitTest/README.md @@ -23,7 +23,7 @@ If in a virtual environment just start by upgrading pip. pip install --upgrade pip ``` -After upgrading pip, the requirements file found in Tests/UnitTests can be installed. This contains all +After upgrading pip, the requirements file found in Tests/UnitTests can be installed. This contains all python modules required to run all of the scripts. This will install tensorflow and keras to allow the use of the generate_test_data.py script. If you have version specific requirements, it is reccomended to install this requirements.txt in a virtual environment. @@ -74,11 +74,11 @@ The easiest way to run the unit tests on Corstone-300 is to use the build_and_ru Sample usage: ``` -./build_and_run_tests.sh -c cortex-m3,cortex-m7,cortex-m55 -o '-Ofast' +./build_and_run_tests.sh -c cortex-m3,cortex-m7,cortex-m55 -o '-Ofast' ``` By default the script will download and target gcc. To use arm compiler ensure that arm compilers folder is located in path, export CC and use the -a option on the script. -Downloaded dependencies including python venv can be found in Tests/UnitTests/downloads. Test elfs can be found in Tests/UnitTests/build-($cpu) directories. +Downloaded dependencies including python venv can be found in Tests/UnitTests/downloads. Test elfs can be found in Tests/UnitTests/build-($cpu) directories. Otherwise, you can build it manually: @@ -150,6 +150,7 @@ Operator bit-exactness compability: | add | x | x | | mul | x | x | | batch matmul | x | x | +| pad | x | x | | minimum | x | x | | maximum | x | x | @@ -178,6 +179,7 @@ Current progress: | add | x | | | mul | x | | | batch matmul | | x | +| pad | | x | | minimum | | x | | maximum | | x | diff --git a/Tests/UnitTest/RefactoredTestGen/Lib/op_pad.py b/Tests/UnitTest/RefactoredTestGen/Lib/op_pad.py new file mode 100644 index 0000000..0ee5e6a --- /dev/null +++ b/Tests/UnitTest/RefactoredTestGen/Lib/op_pad.py @@ -0,0 +1,69 @@ +# SPDX-FileCopyrightText: Copyright 2024 Arm Limited and/or its affiliates +# +# SPDX-License-Identifier: Apache-2.0 +# +# Licensed under the Apache License, Version 2.0 (the License); you may +# not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an AS IS BASIS, WITHOUT +# WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. +# +import Lib.op_utils +import tensorflow as tf +import math +import numpy as np + +from tensorflow.lite.python.interpreter import Interpreter +from tensorflow.lite.python.interpreter import OpResolverType +import tf_keras as keras + +class Op_pad(Lib.op_utils.Op_type): + + def get_shapes(params): + shapes = {} + shapes["input_tensor"] = (params["input_n"], params["input_h"], params["input_w"], params["input_c"]) + shapes["representational_dataset"] = shapes["input_tensor"] + + return shapes + + def generate_keras_model(shapes, params): + + model = keras.models.Sequential() + model.add(keras.layers.InputLayer(input_shape=shapes["input_tensor"][1:])) + + if (params["pre_pad_n"] == params["post_pad_n"] == params["pre_pad_h"] == params["post_pad_h"] == 0): + model.add(keras.layers.ZeroPadding2D(padding=((params["pre_pad_w"], params["post_pad_w"]), (params["pre_pad_c"], params["post_pad_c"])), data_format="channels_first")) + elif (params["pre_pad_n"] == params["post_pad_n"] == params["pre_pad_c"] == params["post_pad_c"] == 0): + model.add(keras.layers.ZeroPadding2D(padding=((params["pre_pad_h"], params["post_pad_h"]), (params["pre_pad_w"], params["post_pad_w"])), data_format="channels_last")) + else: + raise ValueError(f"Keras can only generate padding for (h,w) or (w,c), the others must be zero.") + + return model + + def generate_data_tflite(tflite_fname, params): + tensors = {} + effective_scales = {} + scales = {} + generated_params = {} + + generated_params["pad_value"] = -128 + + interpreter = Interpreter(str(tflite_fname), experimental_op_resolver_type=OpResolverType.BUILTIN_REF) + interpreter.allocate_tensors() + + output_details = interpreter.get_output_details() + output_n = output_details[0]['shape'][3] + output_h = output_details[0]['shape'][2] + output_w = output_details[0]['shape'][1] + output_c = output_details[0]['shape'][0] + + generated_params["output_size"] = output_n * output_h * output_w * output_c; + + return Lib.op_utils.Generated_data(generated_params, tensors, scales, effective_scales) + diff --git a/Tests/UnitTest/RefactoredTestGen/Lib/test.py b/Tests/UnitTest/RefactoredTestGen/Lib/test.py index e1a72cc..f9eda30 100644 --- a/Tests/UnitTest/RefactoredTestGen/Lib/test.py +++ b/Tests/UnitTest/RefactoredTestGen/Lib/test.py @@ -20,6 +20,7 @@ import Lib.op_batch_matmul import Lib.op_fully_connected import Lib.op_pooling +import Lib.op_pad import Lib.op_maximum_minimum import tensorflow as tf import numpy as np @@ -186,6 +187,8 @@ def get_op_type(op_type_string): return Lib.op_fully_connected.Op_fully_connected elif op_type_string == "avgpool" or op_type_string == "maxpool": return Lib.op_pooling.Op_pooling + if op_type_string == "pad": + return Lib.op_pad.Op_pad elif op_type_string == "maximum_minimum": return Lib.op_maximum_minimum.Op_maximum_minimum else: diff --git a/Tests/UnitTest/RefactoredTestGen/test_plan.json b/Tests/UnitTest/RefactoredTestGen/test_plan.json index dd26901..5c15387 100644 --- a/Tests/UnitTest/RefactoredTestGen/test_plan.json +++ b/Tests/UnitTest/RefactoredTestGen/test_plan.json @@ -820,7 +820,7 @@ "out_ch" : 22, "generate_bias": true, "per_channel_quant": true - } + } ] }, { @@ -841,8 +841,8 @@ "stride_w": 9, "stride_h": 5, "pad" : "SAME", - "activation_max": 127, - "activation_min": -128 + "activation_max": 127, + "activation_min": -128 }, {"name" : "avgpooling_1", "batch_size" : 1, @@ -855,8 +855,8 @@ "stride_w": 1, "stride_h": 2, "pad" : "VALID", - "activation_max": 127, - "activation_min": -128 + "activation_max": 127, + "activation_min": -128 }, {"name" : "avgpooling_2", "batch_size" : 1, @@ -869,8 +869,8 @@ "stride_w": 1, "stride_h": 2, "pad" : "SAME", - "activation_max": 127, - "activation_min": -128 + "activation_max": 127, + "activation_min": -128 }, {"name" : "avgpooling_3", "batch_size" : 1, @@ -883,8 +883,8 @@ "stride_w": 2, "stride_h": 1, "pad" : "VALID", - "activation_max": 127, - "activation_min": -128 + "activation_max": 127, + "activation_min": -128 }, {"name" : "avgpooling_4", "batch_size" : 3, @@ -897,8 +897,8 @@ "stride_w": 1, "stride_h": 3, "pad" : "SAME", - "activation_max": 127, - "activation_min": -128 + "activation_max": 127, + "activation_min": -128 }, {"name" : "avgpooling_5", "batch_size" : 1, @@ -911,8 +911,8 @@ "stride_w": 1, "stride_h": 1, "pad" : "SAME", - "activation_max": 6, - "activation_min": 0 + "activation_max": 6, + "activation_min": 0 } ] }, @@ -934,8 +934,8 @@ "stride_w": 2, "stride_h": 1, "pad" : "SAME", - "activation_max": 32767, - "activation_min": -32768 + "activation_max": 32767, + "activation_min": -32768 }, {"name" : "avgpooling_int16_1", "batch_size" : 3, @@ -948,8 +948,8 @@ "stride_w": 2, "stride_h": 1, "pad" : "VALID", - "activation_max": 32767, - "activation_min": -32768 + "activation_max": 32767, + "activation_min": -32768 }, {"name" : "avgpooling_int16_2", "batch_size" : 1, @@ -962,8 +962,8 @@ "stride_w": 2, "stride_h": 1, "pad" : "VALID", - "activation_max": 32767, - "activation_min": -32768 + "activation_max": 32767, + "activation_min": -32768 }, {"name" : "avgpooling_int16_3", "batch_size" : 2, @@ -976,8 +976,8 @@ "stride_w": 9, "stride_h": 5, "pad" : "SAME", - "activation_max": 32767, - "activation_min": -32768 + "activation_max": 32767, + "activation_min": -32768 } ] }, @@ -999,8 +999,8 @@ "stride_w": 9, "stride_h": 5, "pad" : "SAME", - "activation_max": 127, - "activation_min": -128 + "activation_max": 127, + "activation_min": -128 }, {"name" : "maxpooling_1", "batch_size" : 1, @@ -1013,8 +1013,8 @@ "stride_w": 1, "stride_h": 2, "pad" : "VALID", - "activation_max": 127, - "activation_min": -128 + "activation_max": 127, + "activation_min": -128 }, {"name" : "maxpooling_2", "batch_size" : 1, @@ -1027,8 +1027,8 @@ "stride_w": 1, "stride_h": 2, "pad" : "SAME", - "activation_max": 127, - "activation_min": -128 + "activation_max": 127, + "activation_min": -128 }, {"name" : "maxpooling_3", "batch_size" : 1, @@ -1041,8 +1041,8 @@ "stride_w": 2, "stride_h": 1, "pad" : "VALID", - "activation_max": 127, - "activation_min": -128 + "activation_max": 127, + "activation_min": -128 }, {"name" : "maxpooling_4", "batch_size" : 1, @@ -1055,8 +1055,8 @@ "stride_w": 1, "stride_h": 3, "pad" : "SAME", - "activation_max": 127, - "activation_min": -128 + "activation_max": 127, + "activation_min": -128 }, {"name" : "maxpooling_5", "batch_size" : 1, @@ -1069,10 +1069,10 @@ "stride_w": 1, "stride_h": 1, "pad" : "SAME", - "activation_max": 127, - "activation_min": -128 + "activation_max": 127, + "activation_min": -128 }, - {"name" : "maxpooling_6", + {"name" : "maxpooling_6", "batch_size" : 1, "input_n" : 1, "input_w" : 1, @@ -1083,10 +1083,10 @@ "stride_w": 1, "stride_h": 3, "pad" : "SAME", - "activation_max": 127, - "activation_min": -128 + "activation_max": 127, + "activation_min": -128 }, - {"name" : "maxpooling_7", + {"name" : "maxpooling_7", "batch_size" : 1, "input_n" : 1, "input_w" : 4, @@ -1097,8 +1097,8 @@ "stride_w": 2, "stride_h": 2, "pad" : "VALID", - "activation_max": 6, - "activation_min": 0 + "activation_max": 6, + "activation_min": 0 } ] }, @@ -1120,8 +1120,8 @@ "stride_w": 2, "stride_h": 2, "pad" : "VALID", - "activation_max": 32767, - "activation_min": -32768 + "activation_max": 32767, + "activation_min": -32768 }, {"name" : "maxpool_int16_1", "batch_size" : 2, @@ -1134,8 +1134,8 @@ "stride_w": 2, "stride_h": 1, "pad" : "SAME", - "activation_max": 30000, - "activation_min": -30000 + "activation_max": 30000, + "activation_min": -30000 }, {"name" : "maxpool_int16_2", "batch_size" : 1, @@ -1148,8 +1148,45 @@ "stride_w": 1, "stride_h": 1, "pad" : "VALID", - "activation_max": 30000, - "activation_min": -30000 + "activation_max": 30000, + "activation_min": -30000 + } + ] +}, +{ + "suite_name" : "test_arm_pad_s8", + "op_type" : "pad", + "input_data_type": "int8_t", + "interpreter": "tensorflow", + "tflite_generator": "keras", + "tests" : [ + {"name" : "pad_int8_1", + "input_n" : 1, + "input_w" : 2, + "input_h" : 2, + "input_c" : 2, + "pre_pad_n": 0, + "pre_pad_h": 0, + "pre_pad_w": 1, + "pre_pad_c": 1, + "post_pad_n": 0, + "post_pad_h": 0, + "post_pad_w": 2, + "post_pad_c": 2 + }, + {"name" : "pad_int8_2", + "input_n" : 1, + "input_w" : 2, + "input_h" : 2, + "input_c" : 2, + "pre_pad_n": 0, + "pre_pad_h": 2, + "pre_pad_w": 2, + "pre_pad_c": 0, + "post_pad_n": 0, + "post_pad_h": 1, + "post_pad_w": 1, + "post_pad_c": 0 } ] }, diff --git a/Tests/UnitTest/TestCases/TestData/pad_int8_1/config_data.h b/Tests/UnitTest/TestCases/TestData/pad_int8_1/config_data.h new file mode 100644 index 0000000..07e7311 --- /dev/null +++ b/Tests/UnitTest/TestCases/TestData/pad_int8_1/config_data.h @@ -0,0 +1,17 @@ +// Generated by generate_test_data.py using tensorflow version 2.17.0 (Keras version 3.5.0). +// Interpreter from tensorflow version 2.17.0 and revision v2.17.0-rc1-2-gad6d8cc177d. +#pragma once +#define PAD_INT8_1_INPUT_N 1 +#define PAD_INT8_1_INPUT_W 2 +#define PAD_INT8_1_INPUT_H 2 +#define PAD_INT8_1_INPUT_C 2 +#define PAD_INT8_1_PRE_PAD_N 0 +#define PAD_INT8_1_PRE_PAD_H 0 +#define PAD_INT8_1_PRE_PAD_W 1 +#define PAD_INT8_1_PRE_PAD_C 1 +#define PAD_INT8_1_POST_PAD_N 0 +#define PAD_INT8_1_POST_PAD_H 0 +#define PAD_INT8_1_POST_PAD_W 2 +#define PAD_INT8_1_POST_PAD_C 2 +#define PAD_INT8_1_PAD_VALUE -128 +#define PAD_INT8_1_OUTPUT_SIZE 50 diff --git a/Tests/UnitTest/TestCases/TestData/pad_int8_1/input_tensor.h b/Tests/UnitTest/TestCases/TestData/pad_int8_1/input_tensor.h new file mode 100644 index 0000000..502aef2 --- /dev/null +++ b/Tests/UnitTest/TestCases/TestData/pad_int8_1/input_tensor.h @@ -0,0 +1,6 @@ +// Generated by generate_test_data.py using tensorflow version 2.17.0 (Keras version 3.5.0). +// Interpreter from tensorflow version 2.17.0 and revision v2.17.0-rc1-2-gad6d8cc177d. +#pragma once +#include + +const int8_t pad_int8_1_input_tensor[8] = {-102, 10, 30, 95, 76, 9, 79, -121}; diff --git a/Tests/UnitTest/TestCases/TestData/pad_int8_1/output.h b/Tests/UnitTest/TestCases/TestData/pad_int8_1/output.h new file mode 100644 index 0000000..89558d0 --- /dev/null +++ b/Tests/UnitTest/TestCases/TestData/pad_int8_1/output.h @@ -0,0 +1,9 @@ +// Generated by generate_test_data.py using tensorflow version 2.17.0 (Keras version 3.5.0). +// Interpreter from tensorflow version 2.17.0 and revision v2.17.0-rc1-2-gad6d8cc177d. +#pragma once +#include + +const int8_t pad_int8_1_output[50] = {-128, -128, -128, -128, -128, -128, -102, 10, -128, -128, -128, 30, 95, + -128, -128, -128, -128, -128, -128, -128, -128, -128, -128, -128, -128, -128, + -128, -128, -128, -128, -128, 76, 9, -128, -128, -128, 79, -121, -128, + -128, -128, -128, -128, -128, -128, -128, -128, -128, -128, -128}; diff --git a/Tests/UnitTest/TestCases/TestData/pad_int8_1/test_data.h b/Tests/UnitTest/TestCases/TestData/pad_int8_1/test_data.h new file mode 100644 index 0000000..0e46bde --- /dev/null +++ b/Tests/UnitTest/TestCases/TestData/pad_int8_1/test_data.h @@ -0,0 +1,3 @@ +#include "config_data.h" +#include "input_tensor.h" +#include "output.h" diff --git a/Tests/UnitTest/TestCases/TestData/pad_int8_2/config_data.h b/Tests/UnitTest/TestCases/TestData/pad_int8_2/config_data.h new file mode 100644 index 0000000..c7303ef --- /dev/null +++ b/Tests/UnitTest/TestCases/TestData/pad_int8_2/config_data.h @@ -0,0 +1,17 @@ +// Generated by generate_test_data.py using tensorflow version 2.17.0 (Keras version 3.5.0). +// Interpreter from tensorflow version 2.17.0 and revision v2.17.0-rc1-2-gad6d8cc177d. +#pragma once +#define PAD_INT8_2_INPUT_N 1 +#define PAD_INT8_2_INPUT_W 2 +#define PAD_INT8_2_INPUT_H 2 +#define PAD_INT8_2_INPUT_C 2 +#define PAD_INT8_2_PRE_PAD_N 0 +#define PAD_INT8_2_PRE_PAD_H 2 +#define PAD_INT8_2_PRE_PAD_W 2 +#define PAD_INT8_2_PRE_PAD_C 0 +#define PAD_INT8_2_POST_PAD_N 0 +#define PAD_INT8_2_POST_PAD_H 1 +#define PAD_INT8_2_POST_PAD_W 1 +#define PAD_INT8_2_POST_PAD_C 0 +#define PAD_INT8_2_PAD_VALUE -128 +#define PAD_INT8_2_OUTPUT_SIZE 50 diff --git a/Tests/UnitTest/TestCases/TestData/pad_int8_2/input_tensor.h b/Tests/UnitTest/TestCases/TestData/pad_int8_2/input_tensor.h new file mode 100644 index 0000000..021612b --- /dev/null +++ b/Tests/UnitTest/TestCases/TestData/pad_int8_2/input_tensor.h @@ -0,0 +1,6 @@ +// Generated by generate_test_data.py using tensorflow version 2.17.0 (Keras version 3.5.0). +// Interpreter from tensorflow version 2.17.0 and revision v2.17.0-rc1-2-gad6d8cc177d. +#pragma once +#include + +const int8_t pad_int8_2_input_tensor[8] = {77, 94, -41, 47, 21, 61, -98, -2}; diff --git a/Tests/UnitTest/TestCases/TestData/pad_int8_2/output.h b/Tests/UnitTest/TestCases/TestData/pad_int8_2/output.h new file mode 100644 index 0000000..17f6635 --- /dev/null +++ b/Tests/UnitTest/TestCases/TestData/pad_int8_2/output.h @@ -0,0 +1,9 @@ +// Generated by generate_test_data.py using tensorflow version 2.17.0 (Keras version 3.5.0). +// Interpreter from tensorflow version 2.17.0 and revision v2.17.0-rc1-2-gad6d8cc177d. +#pragma once +#include + +const int8_t pad_int8_2_output[50] = {-128, -128, -128, -128, -128, -128, -128, -128, -128, -128, -128, -128, -128, + -128, -128, -128, -128, -128, -128, -128, -128, -128, -128, -128, 77, 94, + -41, 47, -128, -128, -128, -128, -128, -128, 21, 61, -98, -2, -128, + -128, -128, -128, -128, -128, -128, -128, -128, -128, -128, -128}; diff --git a/Tests/UnitTest/TestCases/TestData/pad_int8_2/test_data.h b/Tests/UnitTest/TestCases/TestData/pad_int8_2/test_data.h new file mode 100644 index 0000000..0e46bde --- /dev/null +++ b/Tests/UnitTest/TestCases/TestData/pad_int8_2/test_data.h @@ -0,0 +1,3 @@ +#include "config_data.h" +#include "input_tensor.h" +#include "output.h" diff --git a/Tests/UnitTest/TestCases/test_arm_pad_s8/CMakeLists.txt b/Tests/UnitTest/TestCases/test_arm_pad_s8/CMakeLists.txt new file mode 100644 index 0000000..b9c04b9 --- /dev/null +++ b/Tests/UnitTest/TestCases/test_arm_pad_s8/CMakeLists.txt @@ -0,0 +1,23 @@ +# +# SPDX-FileCopyrightText: Copyright 2010-2024 Arm Limited and/or its affiliates +# +# SPDX-License-Identifier: Apache-2.0 +# +# Licensed under the Apache License, Version 2.0 (the License); you may +# not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an AS IS BASIS, WITHOUT +# WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. +# + +add_cmsis_nn_unit_test_executable(test_arm_pad_s8) + +target_sources(test_arm_pad_s8 PRIVATE + Unity/unity_test_arm_pad_s8.c + Unity/TestRunner/unity_test_arm_pad_s8_runner.c) diff --git a/Tests/UnitTest/TestCases/test_arm_pad_s8/Unity/unity_test_arm_pad_s8.c b/Tests/UnitTest/TestCases/test_arm_pad_s8/Unity/unity_test_arm_pad_s8.c new file mode 100644 index 0000000..8dcea1e --- /dev/null +++ b/Tests/UnitTest/TestCases/test_arm_pad_s8/Unity/unity_test_arm_pad_s8.c @@ -0,0 +1,49 @@ +/* + * SPDX-FileCopyrightText: Copyright 2010-2024 Arm Limited and/or its affiliates + * + * SPDX-License-Identifier: Apache-2.0 + * + * Licensed under the Apache License, Version 2.0 (the License); you may + * not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an AS IS BASIS, WITHOUT + * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include +#include +#include +#include +#include + +#include "../test_arm_pad_s8.c" +#include "unity.h" + +#ifdef USING_FVP_CORSTONE_300 +extern void uart_init(void); +#endif + +/* This function is called from the autogenerated file. + * The name must be exactly like this + */ +void setUp(void) +{ /* This is run before EACH TEST */ +#ifdef USING_FVP_CORSTONE_300 + uart_init(); +#endif +} + +/* This function is called from the autogenerated file. + * The name must be exactly like this + */ +void tearDown(void) {} + +void test_pad_int8_1_arm_pad_s8(void) { pad_int8_1_arm_pad_s8(); } + +void test_pad_int8_2_arm_pad_s8(void) { pad_int8_2_arm_pad_s8(); } diff --git a/Tests/UnitTest/TestCases/test_arm_pad_s8/test_arm_pad_s8.c b/Tests/UnitTest/TestCases/test_arm_pad_s8/test_arm_pad_s8.c new file mode 100644 index 0000000..9114d80 --- /dev/null +++ b/Tests/UnitTest/TestCases/test_arm_pad_s8/test_arm_pad_s8.c @@ -0,0 +1,58 @@ +/* + * SPDX-FileCopyrightText: Copyright 2010-2024 Arm Limited and/or its affiliates + * + * SPDX-License-Identifier: Apache-2.0 + * + * Licensed under the Apache License, Version 2.0 (the License); you may + * not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an AS IS BASIS, WITHOUT + * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include "../TestData/pad_int8_1/test_data.h" +#include "../TestData/pad_int8_2/test_data.h" +#include "../Utils/validate.h" +#include "arm_nn_types.h" +#include "arm_nnfunctions.h" +#include "unity.h" + +void pad_int8_1_arm_pad_s8(void) +{ + const int8_t *input_ptr = pad_int8_1_input_tensor; + int8_t output_ptr[PAD_INT8_1_OUTPUT_SIZE] = {0}; + + const cmsis_nn_dims input_size = {PAD_INT8_1_INPUT_N, PAD_INT8_1_INPUT_H, PAD_INT8_1_INPUT_W, PAD_INT8_1_INPUT_C}; + const cmsis_nn_dims pre_pad = { + PAD_INT8_1_PRE_PAD_N, PAD_INT8_1_PRE_PAD_H, PAD_INT8_1_PRE_PAD_W, PAD_INT8_1_PRE_PAD_C}; + const cmsis_nn_dims post_pad = { + PAD_INT8_1_POST_PAD_N, PAD_INT8_1_POST_PAD_H, PAD_INT8_1_POST_PAD_W, PAD_INT8_1_POST_PAD_C}; + + const arm_cmsis_nn_status result = + arm_pad_s8(input_ptr, output_ptr, PAD_INT8_1_PAD_VALUE, &input_size, &pre_pad, &post_pad); + TEST_ASSERT_EQUAL(ARM_CMSIS_NN_SUCCESS, result); + TEST_ASSERT_TRUE(validate(output_ptr, pad_int8_1_output, PAD_INT8_1_OUTPUT_SIZE)); +} + +void pad_int8_2_arm_pad_s8(void) +{ + const int8_t *input_ptr = pad_int8_2_input_tensor; + int8_t output_ptr[PAD_INT8_2_OUTPUT_SIZE] = {0}; + + const cmsis_nn_dims input_size = {PAD_INT8_2_INPUT_N, PAD_INT8_2_INPUT_H, PAD_INT8_2_INPUT_W, PAD_INT8_2_INPUT_C}; + const cmsis_nn_dims pre_pad = { + PAD_INT8_2_PRE_PAD_N, PAD_INT8_2_PRE_PAD_H, PAD_INT8_2_PRE_PAD_W, PAD_INT8_2_PRE_PAD_C}; + const cmsis_nn_dims post_pad = { + PAD_INT8_2_POST_PAD_N, PAD_INT8_2_POST_PAD_H, PAD_INT8_2_POST_PAD_W, PAD_INT8_2_POST_PAD_C}; + + const arm_cmsis_nn_status result = + arm_pad_s8(input_ptr, output_ptr, PAD_INT8_2_PAD_VALUE, &input_size, &pre_pad, &post_pad); + TEST_ASSERT_EQUAL(ARM_CMSIS_NN_SUCCESS, result); + TEST_ASSERT_TRUE(validate(output_ptr, pad_int8_2_output, PAD_INT8_2_OUTPUT_SIZE)); +}