diff options
Diffstat (limited to 'tests/validation/CL/UNIT')
| -rw-r--r-- | tests/validation/CL/UNIT/CompileContext.cpp | 3 | ||||
| -rw-r--r-- | tests/validation/CL/UNIT/DynamicTensor.cpp | 5 | ||||
| -rw-r--r-- | tests/validation/CL/UNIT/Extensions.cpp | 49 | ||||
| -rw-r--r-- | tests/validation/CL/UNIT/MLGOHeuristics.cpp | 473 | ||||
| -rw-r--r-- | tests/validation/CL/UNIT/MemoryManager.cpp | 2 | ||||
| -rw-r--r-- | tests/validation/CL/UNIT/Multithreaded.cpp | 113 | ||||
| -rw-r--r-- | tests/validation/CL/UNIT/TensorAllocator.cpp | 137 | ||||
| -rw-r--r-- | tests/validation/CL/UNIT/Tuner.cpp | 78 | ||||
| -rw-r--r-- | tests/validation/CL/UNIT/WeightsRetention.cpp | 3 |
9 files changed, 763 insertions, 100 deletions
diff --git a/tests/validation/CL/UNIT/CompileContext.cpp b/tests/validation/CL/UNIT/CompileContext.cpp index 5245044323..a0541b6a99 100644 --- a/tests/validation/CL/UNIT/CompileContext.cpp +++ b/tests/validation/CL/UNIT/CompileContext.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2020 ARM Limited. + * Copyright (c) 2020 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -56,6 +56,7 @@ TEST_CASE(CompileContextCache, framework::DatasetMode::ALL) std::set<std::string> build_opts; build_opts.emplace("-DDATA_TYPE=float"); build_opts.emplace("-DVEC_SIZE=16"); + build_opts.emplace("-DVEC_SIZE_LEFTOVER=0"); compile_context.create_kernel(kernel_name, program_name, kernel_src.first, kernel_path, build_opts, kernel_src.second); // Check if the program is stored in the cache diff --git a/tests/validation/CL/UNIT/DynamicTensor.cpp b/tests/validation/CL/UNIT/DynamicTensor.cpp index f5cc1f7c9d..ac433721d8 100644 --- a/tests/validation/CL/UNIT/DynamicTensor.cpp +++ b/tests/validation/CL/UNIT/DynamicTensor.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2019-2020 ARM Limited. + * Copyright (c) 2019-2021 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -28,6 +28,9 @@ #include "arm_compute/runtime/MemoryGroup.h" #include "arm_compute/runtime/MemoryManagerOnDemand.h" #include "arm_compute/runtime/PoolManager.h" +#include "src/core/CL/kernels/CLFillBorderKernel.h" +#include "src/core/CL/kernels/CLL2NormalizeLayerKernel.h" +#include "src/core/CL/kernels/CLReductionOperationKernel.h" #include "tests/AssetsLibrary.h" #include "tests/CL/CLAccessor.h" #include "tests/Globals.h" diff --git a/tests/validation/CL/UNIT/Extensions.cpp b/tests/validation/CL/UNIT/Extensions.cpp new file mode 100644 index 0000000000..8119290d4b --- /dev/null +++ b/tests/validation/CL/UNIT/Extensions.cpp @@ -0,0 +1,49 @@ +/* + * Copyright (c) 2021 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 "arm_compute/AclVersion.h" +#include "tests/framework/Asserts.h" +#include "tests/framework/Macros.h" +#include "tests/validation/Validation.h" + +namespace arm_compute +{ +namespace test +{ +namespace validation +{ +TEST_SUITE(CL) +TEST_SUITE(UNIT) +TEST_CASE(GetClContext, framework::DatasetMode::ALL) +{ + const auto ver = AclVersionInfo(); + ARM_COMPUTE_EXPECT(ver->major == ARM_COMPUTE_LIBRARY_VERSION_MAJOR, framework::LogLevel::ERRORS); + ARM_COMPUTE_EXPECT(ver->minor == ARM_COMPUTE_LIBRARY_VERSION_MINOR, framework::LogLevel::ERRORS); + ARM_COMPUTE_EXPECT(ver->patch == ARM_COMPUTE_LIBRARY_VERSION_PATCH, framework::LogLevel::ERRORS); + ARM_COMPUTE_EXPECT(!std::string(ver->build_info).empty(), framework::LogLevel::ERRORS); +} +TEST_SUITE_END() // UNIT +TEST_SUITE_END() // CL +} // namespace validation +} // namespace test +} // namespace arm_compute diff --git a/tests/validation/CL/UNIT/MLGOHeuristics.cpp b/tests/validation/CL/UNIT/MLGOHeuristics.cpp new file mode 100644 index 0000000000..e26464f9f7 --- /dev/null +++ b/tests/validation/CL/UNIT/MLGOHeuristics.cpp @@ -0,0 +1,473 @@ +/* + * Copyright (c) 2021 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/runtime/CL/mlgo/MLGOHeuristics.h" +#include "src/runtime/CL/mlgo/Utils.h" +#include "tests/framework/Asserts.h" +#include "tests/framework/Macros.h" + +using namespace arm_compute::mlgo; + +namespace arm_compute +{ +namespace test +{ +namespace validation +{ +TEST_SUITE(CL) +TEST_SUITE(UNIT) +TEST_SUITE(MLGOHeuristics) +TEST_CASE(CorrectDotMLGOShouldLoadCorrectly, framework::DatasetMode::ALL) +{ + std::string mlgo_str = R"_( + + <header> + + gemm-version, [1,2,1] + ip-type,gpu + </header> + <heuristics-table> + 0, g76 , 8, f32, best-performance, static, gemm-type, [m,n,k,n] + + 1, g71 , 8, f16, best-performance, static, gemm-config-reshaped-only-rhs, [m,n,k,n] + 2, g76 , 8, f16, best-performance, static, gemm-config-reshaped, [m,n,k,n] + </heuristics-table> + <heuristic, 0> + b , 0, var, m, ==, num, 10., 1, 2 + l , 1, gemm-type, reshaped + b , 2, var, r_mn, >=, num, 2., 3, 6 + + b , 3, var, n, >=, num, 200., 4, 5 + l, 4, gemm-type, reshaped-only-rhs + l , 5, gemm-type, reshaped + l , 6, gemm-type, reshaped-only-rhs + </heuristic> + <heuristic, 1> + b ,0,var, n, >, num, 100., 1, 4 + b ,1,var, r_mnk, <=, num, 20., 2, 3 + + + l ,2,gemm-config-reshaped-only-rhs, [4, 4,4,2,1,0,1] + l ,3,gemm-config-reshaped-only-rhs,[ 2, 2,4,2,1,1, 1 ] + b ,4,var, n, >=, num, 199.12, 5, 6 + l ,5,gemm-config-reshaped-only-rhs, [1, 4,3,4,0,0,0] + l ,6,gemm-config-reshaped-only-rhs, [5, 4,4,5,1,1,0] + </heuristic> + + <heuristic, 2> + l ,0,gemm-config-reshaped,[4,2,4,2,8,1,0,1,0] + + </heuristic> + + )_"; + std::stringstream ss(mlgo_str); + MLGOHeuristics heuristics; + heuristics.reload_from_stream(ss); + + ARM_COMPUTE_EXPECT(heuristics.query_gemm_type(Query{ "g76", DataType::F32, 10, 1024, 20, 1 }).second == GEMMType::RESHAPED, framework::LogLevel::ERRORS); + ARM_COMPUTE_EXPECT(heuristics.query_gemm_type(Query{ "g76", DataType::F32, 400, 201, 5, 1 }).second == GEMMType::RESHAPED_ONLY_RHS, framework::LogLevel::ERRORS); + ARM_COMPUTE_EXPECT(heuristics.query_gemm_type(Query{ "g76", DataType::F32, 400, 200, 199, 16 }).second == GEMMType::RESHAPED_ONLY_RHS, framework::LogLevel::ERRORS); + ARM_COMPUTE_EXPECT(heuristics.query_gemm_type(Query{ "g76", DataType::F32, 400, 199, 512, 4 }).second == GEMMType::RESHAPED, framework::LogLevel::ERRORS); + + ARM_COMPUTE_EXPECT((heuristics.query_gemm_config_reshaped_only_rhs(Query{ "g71", DataType::F16, 100, 1024, 20, 32 }).second == GEMMConfigReshapedOnlyRHS{ 4, 4, 4, 2, true, false, true }), + framework::LogLevel::ERRORS); + ARM_COMPUTE_EXPECT((heuristics.query_gemm_config_reshaped_only_rhs(Query{ "g71", DataType::F16, 100, 1024, 20, 32 }).second == GEMMConfigReshapedOnlyRHS{ 4, 4, 4, 2, true, false, true }), + framework::LogLevel::ERRORS); + ARM_COMPUTE_EXPECT((heuristics.query_gemm_config_reshaped_only_rhs(Query{ "g71", DataType::F16, 128, 101, 20, 1 }).second == GEMMConfigReshapedOnlyRHS{ 2, 2, 4, 2, true, true, true }), + framework::LogLevel::ERRORS); + ARM_COMPUTE_EXPECT((heuristics.query_gemm_config_reshaped_only_rhs(Query{ "g71", DataType::F16, 400, 100, 512, 1 }).second == GEMMConfigReshapedOnlyRHS{ 5, 4, 4, 5, true, true, false }), + framework::LogLevel::ERRORS); + ARM_COMPUTE_EXPECT((heuristics.query_gemm_config_reshaped_only_rhs(Query{ "g71", DataType::F16, 400, 100, 512, 1 }).second == GEMMConfigReshapedOnlyRHS{ 5, 4, 4, 5, true, true, false }), + framework::LogLevel::ERRORS); + + ARM_COMPUTE_EXPECT((heuristics.query_gemm_config_reshaped(Query{ "g76", DataType::F16, 100, 100, 20, 32 }).second == GEMMConfigReshaped{ 4, 2, 4, 2, 8, true, false, true, false }), + framework::LogLevel::ERRORS); + ARM_COMPUTE_EXPECT((heuristics.query_gemm_config_reshaped(Query{ "g76", DataType::F16, 128, 512, 1024, 1 }).second == GEMMConfigReshaped{ 4, 2, 4, 2, 8, true, false, true, false }), + framework::LogLevel::ERRORS); +} + +TEST_CASE(InvalidDotmlgoSyntaxShouldReturnInvalidStatus, framework::DatasetMode::ALL) +{ + std::string mlgo_str = R"_( + <header> + gemm-version, [1,2,1] + ip-type,pu + </header> + <heuristics-table> + 0, g76 , 8, f32, best-performance, static, gemm-config-reshaped, [m,n,k,n] + + </heurist + <heuristic, 0> + l ,0,gemm-config-reshaped,[4,2,4,2,8,1,0,1,0] + </heuristic> + )_"; + std::stringstream ss(mlgo_str); + MLGOHeuristics heuristics; + ARM_COMPUTE_EXPECT(!heuristics.reload_from_stream(ss), framework::LogLevel::ERRORS); +} + +TEST_SUITE(InvalidDotmlgoSemanticsShouldReturnInvalidStatus) +// If the semantics errors are local to some trees instead of the entire heuristics, an alternative is to simply +// ignore/remove those invalid trees. However the reason why we choose to throw, thus invalidating the entire +// heuristics is that if there are some invalid trees, the quality of the dotmlgo is called into question even if +// the rest of the trees are semantically valid, and they could severely degrade the performance of GEMM. Therefore +// this "all or nothing" approach when it comes to dotmlgo correctness is safer and more defensive. + +// Also note that the semantic error of the tree only refers to those that obstruct its evaluation and thus query, +// (e.g. invalid tree structure, unsupported features etc.) instead of those affecting the desired outcome +// (usually in terms of final GEMM performance, e.g. the effectiveness of the decision tree) + +// In the future we might want to check the content of the exceptions as well. But right now it suffices to only +// know that it throws exactly when it needs to. +TEST_CASE(MismatchesBetweenHeuristicsTableEntriesAndHeuristicTrees, framework::DatasetMode::ALL) +{ + { + // Mismatching number of entries 1 + std::string mlgo_str = R"_( + <header> + gemm-version, [1,2,1] + ip-type,gpu + </header> + <heuristics-table> + + 0, g76 , 8, f32, best-performance, static, gemm-config-reshaped, [m,n,k,n] + + </heuristics-table> + )_"; + std::stringstream ss(mlgo_str); + MLGOHeuristics heuristics; + // NOTE: This case might throw an internal error as the tree inserted by the heuristics-table cannot not be checked + ARM_COMPUTE_EXPECT(!heuristics.reload_from_stream(ss), framework::LogLevel::ERRORS); + } + + { + // Mismatching number of entries 2 + std::string mlgo_str = R"_( + <header> + gemm-version, [1,2,1] + ip-type,gpu + </header> + <heuristics-table> + </heuristics-table> + <heuristic, 1> + l ,0,gemm-config-reshaped,[4,2,4,2,8,1,0,1,0] + </heuristic> + )_"; + std::stringstream ss(mlgo_str); + MLGOHeuristics heuristics; + ARM_COMPUTE_EXPECT(!heuristics.reload_from_stream(ss), framework::LogLevel::ERRORS); + } + + { + // Mismatching info + std::string mlgo_str = R"_( + <header> + gemm-version, [1,2,1] + ip-type,gpu + </header> + <heuristics-table> + 0, g76 , 8, f32, best-performance, static, gemm-type, [m,n,k,n] + </heuristics-table> + <heuristic, 0> + l ,0,gemm-config-reshaped,[4,2,4,2,8,1,0,1,0] + </heuristic> + )_"; + std::stringstream ss(mlgo_str); + MLGOHeuristics heuristics; + ARM_COMPUTE_EXPECT(!heuristics.reload_from_stream(ss), framework::LogLevel::ERRORS); + } +} + +TEST_CASE(RepeatedHeuristicsTableEntriesId, framework::DatasetMode::ALL) +{ + std::string mlgo_str = R"_( + <header> + gemm-version, [1,2,1] + ip-type,gpu + </header> + <heuristics-table> + 0, g76 , 8, f32, best-performance, static, gemm-config-reshaped, [m,n,k,n] + 0, g71 , 8, f32, best-performance, static, gemm-config-reshaped, [m,n,k,n] + </heuristics-table> + <heuristic, 0> + l ,0,gemm-config-reshaped,[4,2,4,2,8,1,0,1,0] + </heuristic> + <heuristic, 1> + l ,0,gemm-config-reshaped,[4,2,4,2,8,1,0,1,0] + </heuristic> + )_"; + std::stringstream ss(mlgo_str); + MLGOHeuristics heuristics; + ARM_COMPUTE_EXPECT(!heuristics.reload_from_stream(ss), framework::LogLevel::ERRORS); +} + +TEST_CASE(RepeatedHeuristicsTableEntriesIndex, framework::DatasetMode::ALL) +{ + std::string mlgo_str = R"_( + <header> + gemm-version, [1,2,1] + ip-type,gpu + </header> + <heuristics-table> + 0, g76 , 8, f32, best-performance, static, gemm-config-reshaped, [m,n,k,n] + 1, g76 , 8, f32, best-performance, static, gemm-config-reshaped, [m,n,k,n] + </heuristics-table> + <heuristic, 0> + l ,0,gemm-config-reshaped,[4,2,4,2,8,1,0,1,0] + </heuristic> + <heuristic, 1> + l ,0,gemm-config-reshaped,[4,2,4,2,8,1,0,1,0] + </heuristic> + )_"; + std::stringstream ss(mlgo_str); + MLGOHeuristics heuristics; + ARM_COMPUTE_EXPECT(!heuristics.reload_from_stream(ss), framework::LogLevel::ERRORS); +} + +TEST_CASE(RepeatedHeuristicTreesId, framework::DatasetMode::ALL) +{ + std::string mlgo_str = R"_( + <header> + gemm-version, [1,2,1] + ip-type,gpu + </header> + <heuristics-table> + 0, g76 , 8, f32, best-performance, static, gemm-config-reshaped, [m,n,k,n] + 1, g71 , 8, f32, best-performance, static, gemm-config-reshaped, [m,n,k,n] + </heuristics-table> + <heuristic, 0> + l ,0,gemm-config-reshaped,[4,2,4,2,8,1,0,1,0] + </heuristic> + <heuristic, 0> + l ,0,gemm-config-reshaped,[4,2,4,2,8,1,0,1,0] + </heuristic> + )_"; + std::stringstream ss(mlgo_str); + MLGOHeuristics heuristics; + ARM_COMPUTE_EXPECT(!heuristics.reload_from_stream(ss), framework::LogLevel::ERRORS); +} +TEST_CASE(EmptyTree, framework::DatasetMode::ALL) +{ + std::string mlgo_str = R"_( + <header> + gemm-version, [1,2,1] + ip-type,gpu + </header> + <heuristics-table> + 0, g76 , 8, f32, best-performance, static, gemm-config-reshaped, [m,n,k,n] + </heuristics-table> + <heuristic, 0> + </heuristic> + )_"; + std::stringstream ss(mlgo_str); + MLGOHeuristics heuristics; + ARM_COMPUTE_EXPECT(!heuristics.reload_from_stream(ss), framework::LogLevel::ERRORS); +} + +TEST_CASE(InvalidTreeMissingRoot, framework::DatasetMode::ALL) +{ + std::string mlgo_str = R"_( + <header> + gemm-version, [1,2,1] + ip-type,gpu + </header> + <heuristics-table> + 0, g76 , 8, f32, best-performance, static, gemm-config-reshaped, [m,n,k,n] + </heuristics-table> + <heuristic, 0> + b ,2, var, m, ==, num, 10., 3, 4 + l ,3,gemm-config-reshaped,[4,2,4,2,8,1,0,1,0] + l ,4,gemm-config-reshaped,[4,2,4,2,8,1,0,1,0] + </heuristic> + )_"; + std::stringstream ss(mlgo_str); + MLGOHeuristics heuristics; + ARM_COMPUTE_EXPECT(!heuristics.reload_from_stream(ss), framework::LogLevel::ERRORS); +} +TEST_CASE(InvalidTreeMissingNodes, framework::DatasetMode::ALL) +{ + std::string mlgo_str = R"_( + <header> + gemm-version, [1,2,1] + ip-type,gpu + </header> + <heuristics-table> + 0, g76 , 8, f32, best-performance, static, gemm-config-reshaped, [m,n,k,n] + </heuristics-table> + <heuristic, 0> + b ,0, var, m, ==, num, 10., 1, 2 + l ,1,gemm-config-reshaped,[4,2,4,2,8,1,0,1,0] + </heuristic> + )_"; + std::stringstream ss(mlgo_str); + MLGOHeuristics heuristics; + ARM_COMPUTE_EXPECT(!heuristics.reload_from_stream(ss), framework::LogLevel::ERRORS); +} +TEST_CASE(InvalidTreeRepeatedNodeIds, framework::DatasetMode::ALL) +{ + std::string mlgo_str = R"_( + <header> + gemm-version, [1,2,1] + ip-type,gpu + </header> + <heuristics-table> + 0, g76 , 8, f32, best-performance, static, gemm-config-reshaped, [m,n,k,n] + </heuristics-table> + <heuristic, 0> + b ,0, var, m, ==, num, 10., 1, 2 + l ,1,gemm-config-reshaped,[4,2,4,2,8,1,0,1,0] + l ,1,gemm-config-reshaped,[1,2,4,2,8,1,0,1,0] + l ,2,gemm-config-reshaped,[2,2,4,2,8,1,0,1,0] + </heuristic> + )_"; + std::stringstream ss(mlgo_str); + MLGOHeuristics heuristics; + ARM_COMPUTE_EXPECT(!heuristics.reload_from_stream(ss), framework::LogLevel::ERRORS); +} +TEST_CASE(InvalidTreeDisjointNodes, framework::DatasetMode::ALL) +{ + std::string mlgo_str = R"_( + <header> + gemm-version, [1,2,1] + ip-type,gpu + </header> + <heuristics-table> + 0, g76 , 8, f32, best-performance, static, gemm-config-reshaped, [m,n,k,n] + </heuristics-table> + <heuristic, 0> + b ,0, var, m, ==, num, 10., 1, 2 + l ,1,gemm-config-reshaped,[4,2,4,2,8,1,0,1,0] + l ,2,gemm-config-reshaped,[2,2,4,2,8,1,0,1,0] + + b ,4, var, n, ==, num, 10., 5, 6 + l ,5,gemm-config-reshaped,[4,2,4,2,8,1,0,1,0] + l ,6,gemm-config-reshaped,[2,2,4,2,8,1,0,1,0] + + l ,7,gemm-config-reshaped,[2,2,4,2,8,1,0,1,0] + </heuristic> + )_"; + std::stringstream ss(mlgo_str); + MLGOHeuristics heuristics; + ARM_COMPUTE_EXPECT(!heuristics.reload_from_stream(ss), framework::LogLevel::ERRORS); +} +TEST_CASE(InvalidTreeLoop, framework::DatasetMode::ALL) +{ + std::string mlgo_str = R"_( + <header> + gemm-version, [1,2,1] + ip-type,gpu + </header> + <heuristics-table> + 0, g76 , 8, f32, best-performance, static, gemm-config-reshaped, [m,n,k,n] + </heuristics-table> + <heuristic, 0> + b ,0, var, m, ==, num, 10., 0, 1 + l ,1,gemm-config-reshaped,[4,2,4,2,8,1,0,1,0] + </heuristic> + )_"; + std::stringstream ss(mlgo_str); + MLGOHeuristics heuristics; + ARM_COMPUTE_EXPECT(!heuristics.reload_from_stream(ss), framework::LogLevel::ERRORS); +} +TEST_CASE(InvalidTreeCycle, framework::DatasetMode::ALL) +{ + std::string mlgo_str = R"_( + <header> + gemm-version, [1,2,1] + ip-type,gpu + </header> + <heuristics-table> + 0, g76 , 8, f32, best-performance, static, gemm-config-reshaped, [m,n,k,n] + </heuristics-table> + <heuristic, 0> + b ,0, var, m, ==, num, 10., 1, 5 + b ,1, var, n, ==, num, 10., 2, 3 + l ,2,gemm-config-reshaped,[4,2,4,2,8,1,0,1,0] + b ,3, var, k, ==, num, 10., 0, 4 + l ,4,gemm-config-reshaped,[4,2,4,2,8,1,0,1,0] + l ,5,gemm-config-reshaped,[4,2,4,2,8,1,0,1,0] + </heuristic> + )_"; + std::stringstream ss(mlgo_str); + MLGOHeuristics heuristics; + ARM_COMPUTE_EXPECT(!heuristics.reload_from_stream(ss), framework::LogLevel::ERRORS); +} +TEST_CASE(InvalidTreeInvalidFeatures, framework::DatasetMode::ALL) +{ + std::string mlgo_str = R"_( + <header> + gemm-version, [1,2,1] + ip-type,gpu + </header> + <heuristics-table> + 0, g76 , 8, f32, best-performance, static, gemm-config-reshaped, [m,n,k,n] + </heuristics-table> + <heuristic, 0> + b ,0, var, magic_feature, ==, num, 10., 1, 2 + l ,1,gemm-config-reshaped,[4,2,4,2,8,1,0,1,0] + l ,2,gemm-config-reshaped,[4,2,4,2,8,1,0,1,0] + </heuristic> + )_"; + std::stringstream ss(mlgo_str); + MLGOHeuristics heuristics; + ARM_COMPUTE_EXPECT(!heuristics.reload_from_stream(ss), framework::LogLevel::ERRORS); +} +TEST_SUITE_END() // InvalidDotmlgoSemanticsShouldReturnInvalidStatus + +TEST_CASE(InvalidUsageOfHeuristicsShouldReturnInvalidStatus, framework::DatasetMode::ALL) +{ + std::string mlgo_str = R"_( + <header> + gemm-version, [1,2,1] + ip-type,gpu + </header> + <heuristics-table> + 0, g76 , 8, f32, best-performance, static, gemm-type, [m,n,k,n] + </heuristics-table> + <heuristic, 0> + b , 0, var, m, ==, num, 10., 1, 2 + l , 1, gemm-type, reshaped + b , 2, var, r_mn, >=, num, 2., 3, 6 + b , 3, var, n, >=, num, 200., 4, 5 + l , 4, gemm-type, reshaped-only-rhs + l , 5, gemm-type, reshaped + l , 6, gemm-type, reshaped-only-rhs + </heuristic> + )_"; + std::stringstream ss(mlgo_str); + MLGOHeuristics heuristics; + ARM_COMPUTE_EXPECT(heuristics.reload_from_stream(ss), framework::LogLevel::ERRORS); + + // Querying unavailable heuristic type should return invalid Status + ARM_COMPUTE_EXPECT(!heuristics.query_gemm_config_reshaped(Query{ "g76", DataType::F32, 1024, 1024, 100, 3 }).first, framework::LogLevel::ERRORS); + // Querying unavailable ip target should return invalid Status + ARM_COMPUTE_EXPECT(!heuristics.query_gemm_type(Query{ "g77", DataType::F32, 1024, 1024, 100, 3 }).first, framework::LogLevel::ERRORS); + // Querying unavailable data type should return invalid Status + ARM_COMPUTE_EXPECT(!heuristics.query_gemm_config_reshaped_only_rhs(Query{ "g76", DataType::QASYMM8, 1024, 1024, 100, 3 }).first, framework::LogLevel::ERRORS); +} +TEST_SUITE_END() // MLGOHeuristics +TEST_SUITE_END() // UNIT +TEST_SUITE_END() // CL +} // namespace validation +} // namespace test +} // namespace arm_compute diff --git a/tests/validation/CL/UNIT/MemoryManager.cpp b/tests/validation/CL/UNIT/MemoryManager.cpp index 8167026ec3..4db062ef81 100644 --- a/tests/validation/CL/UNIT/MemoryManager.cpp +++ b/tests/validation/CL/UNIT/MemoryManager.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2018-2020 ARM Limited. + * Copyright (c) 2018-2020 Arm Limited. * * SPDX-License-Identifier: MIT * diff --git a/tests/validation/CL/UNIT/Multithreaded.cpp b/tests/validation/CL/UNIT/Multithreaded.cpp new file mode 100644 index 0000000000..5c75df709d --- /dev/null +++ b/tests/validation/CL/UNIT/Multithreaded.cpp @@ -0,0 +1,113 @@ +/* + * Copyright (c) 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 "arm_compute/runtime/RuntimeContext.h" + +#include "tests/CL/CLAccessor.h" +#include "tests/framework/Macros.h" +#include "tests/framework/ParametersLibrary.h" +#include "tests/validation/Validation.h" +#include "arm_compute/runtime/CL/functions/CLActivationLayer.h" +#include "arm_compute/runtime/CL/functions/CLPixelWiseMultiplication.h" +#include "tests/validation/reference/ActivationLayer.h" +#include "tests/validation/reference/PixelWiseMultiplication.h" +#include <thread> + +namespace arm_compute +{ +namespace test +{ +namespace validation +{ +TEST_SUITE(CL) +TEST_SUITE(UNIT) +TEST_SUITE(RuntimeContext) +// This test tries scheduling work concurrently from two independent threads +TEST_CASE(MultipleThreadedScheduller, framework::DatasetMode::ALL) +{ + constexpr auto num_threads(16u); + std::array<CLActivationLayer, num_threads> func{}; + std::array<CLPixelWiseMultiplication, num_threads> pmul{}; + std::array<CLTensor, num_threads> s0{}; + std::array<CLTensor, num_threads> s1{}; + + std::array<CLTensor, num_threads> st{}; + std::array<CLTensor, num_threads> dt{}; + + const TensorShape tensor_shape(128u, 4u, 5u); + const ActivationLayerInfo ainfo(ActivationLayerInfo::ActivationFunction::LOGISTIC, 0.5f, 1.f); + std::array<std::thread, num_threads> threads; + auto ctx = parameters->get_ctx<CLTensor>(); + + for(auto i = 0u; i < num_threads; ++i) + { + s0[i] = create_tensor<CLTensor>(tensor_shape, DataType::F32, 1); + s1[i] = create_tensor<CLTensor>(tensor_shape, DataType::F32, 1); + st[i] = create_tensor<CLTensor>(tensor_shape, DataType::F32, 1); + dt[i] = create_tensor<CLTensor>(tensor_shape, DataType::F32, 1); + func[i] = CLActivationLayer(ctx); + pmul[i] = CLPixelWiseMultiplication(); + threads[i] = + std::thread([&,i] + { + auto &s = st[i]; + auto &t = dt[i]; + auto &p0 = s0[i]; + auto &p1 = s1[i]; + pmul[i].configure(&p0, &p1, &s, 1.f, ConvertPolicy::WRAP, RoundingPolicy::TO_NEAREST_UP); + func[i].configure(&s, &t, ainfo); + s.allocator()->allocate(); + t.allocator()->allocate(); + p0.allocator()->allocate(); + p1.allocator()->allocate(); + library->fill_tensor_uniform(CLAccessor(p0), 0, -1.f, 1.f); + library->fill_tensor_uniform(CLAccessor(p1), 0, -1.f, 1.f); + pmul[i].run(); + func[i].run(); + }); + } + + for(auto &t : threads) + { + t.join(); + } + + SimpleTensor<float> rs{ tensor_shape, DataType::F32, 1 }; + SimpleTensor<float> ra{ tensor_shape, DataType::F32, 1 }; + SimpleTensor<float> rb{ tensor_shape, DataType::F32, 1 }; + library->fill_tensor_uniform(ra, 0, -1.f, 1.f); + library->fill_tensor_uniform(rb, 0, -1.f, 1.f); + const auto mul = reference::pixel_wise_multiplication<float, float, float>(ra, rb, 1.f, ConvertPolicy::WRAP, RoundingPolicy::TO_NEAREST_UP, DataType::F32); + const auto golden = reference::activation_layer<float>(mul, ainfo); + for(auto &d : dt) + { + validate(CLAccessor(d), golden); + } +} + +TEST_SUITE_END() // MultipleThreadedScheduller +TEST_SUITE_END() // UNIT +TEST_SUITE_END() // CL +} // namespace validation +} // namespace test +} // namespace arm_compute diff --git a/tests/validation/CL/UNIT/TensorAllocator.cpp b/tests/validation/CL/UNIT/TensorAllocator.cpp index a3aabf9bc4..559f47e16c 100644 --- a/tests/validation/CL/UNIT/TensorAllocator.cpp +++ b/tests/validation/CL/UNIT/TensorAllocator.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2018-2019 ARM Limited. + * Copyright (c) 2018-2021 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -24,9 +24,14 @@ #include "arm_compute/runtime/CL/CLTensorAllocator.h" #include "arm_compute/core/utils/misc/MMappedFile.h" +#include "arm_compute/runtime/BlobLifetimeManager.h" +#include "arm_compute/runtime/CL/CLBufferAllocator.h" #include "arm_compute/runtime/CL/CLScheduler.h" #include "arm_compute/runtime/CL/functions/CLActivationLayer.h" +#include "arm_compute/runtime/CL/functions/CLGEMMConvolutionLayer.h" #include "arm_compute/runtime/MemoryGroup.h" +#include "arm_compute/runtime/MemoryManagerOnDemand.h" +#include "arm_compute/runtime/PoolManager.h" #include "tests/CL/CLAccessor.h" #include "tests/Globals.h" #include "tests/framework/Asserts.h" @@ -60,12 +65,108 @@ cl_mem import_malloc_memory_helper(void *ptr, size_t size) return buf; } + +class DummyAllocator final : public IAllocator +{ +public: + DummyAllocator() = default; + + void *allocate(size_t size, size_t alignment) override + { + ++_n_calls; + return _backend_allocator.allocate(size, alignment); + } + void free(void *ptr) override + { + return _backend_allocator.free(ptr); + } + std::unique_ptr<IMemoryRegion> make_region(size_t size, size_t alignment) override + { + // Needs to be implemented as is the one that is used internally by the CLTensorAllocator + ++_n_calls; + return _backend_allocator.make_region(size, alignment); + } + int get_n_calls() const + { + return _n_calls; + } + +private: + int _n_calls{}; + CLBufferAllocator _backend_allocator{}; +}; + +void run_conv2d(std::shared_ptr<IMemoryManager> mm, IAllocator &mm_allocator) +{ + // Create tensors + CLTensor src, weights, bias, dst; + src.allocator()->init(TensorInfo(TensorShape(16U, 32U, 32U, 2U), 1, DataType::F32, DataLayout::NHWC)); + weights.allocator()->init(TensorInfo(TensorShape(16U, 3U, 3U, 32U), 1, DataType::F32, DataLayout::NHWC)); + bias.allocator()->init(TensorInfo(TensorShape(32U), 1, DataType::F32, DataLayout::NHWC)); + dst.allocator()->init(TensorInfo(TensorShape(32U, 32U, 32U, 2U), 1, DataType::F32, DataLayout::NHWC)); + + // Create and configure function + CLGEMMConvolutionLayer conv(mm); + conv.configure(&src, &weights, &bias, &dst, PadStrideInfo(1U, 1U, 1U, 1U)); + + // Allocate tensors + src.allocator()->allocate(); + weights.allocator()->allocate(); + bias.allocator()->allocate(); + dst.allocator()->allocate(); + + // Finalize memory manager + if(mm != nullptr) + { + mm->populate(mm_allocator, 1 /* num_pools */); + ARM_COMPUTE_EXPECT(mm->lifetime_manager()->are_all_finalized(), framework::LogLevel::ERRORS); + ARM_COMPUTE_EXPECT(mm->pool_manager()->num_pools() == 1, framework::LogLevel::ERRORS); + } + + conv.run(); +} } // namespace TEST_SUITE(CL) TEST_SUITE(UNIT) TEST_SUITE(TensorAllocator) +/* Validate that an external global allocator can be used for all internal allocations */ +TEST_CASE(ExternalGlobalAllocator, framework::DatasetMode::ALL) +{ + DummyAllocator global_tensor_alloc; + CLTensorAllocator::set_global_allocator(&global_tensor_alloc); + + // Run a convolution + run_conv2d(nullptr /* mm */, global_tensor_alloc); + + // Check that allocator has been called multiple times > 4 + ARM_COMPUTE_EXPECT(global_tensor_alloc.get_n_calls() > 4, framework::LogLevel::ERRORS); + + // Nullify global allocator + CLTensorAllocator::set_global_allocator(nullptr); +} + +/* Validate that an external global allocator can be used for the pool manager */ +TEST_CASE(ExternalGlobalAllocatorMemoryPool, framework::DatasetMode::ALL) +{ + auto lifetime_mgr = std::make_shared<BlobLifetimeManager>(); + auto pool_mgr = std::make_shared<PoolManager>(); + auto mm = std::make_shared<MemoryManagerOnDemand>(lifetime_mgr, pool_mgr); + + DummyAllocator global_tensor_alloc; + CLTensorAllocator::set_global_allocator(&global_tensor_alloc); + + // Run a convolution + run_conv2d(mm, global_tensor_alloc); + + // Check that allocator has been called multiple times > 4 + ARM_COMPUTE_EXPECT(global_tensor_alloc.get_n_calls() > 4, framework::LogLevel::ERRORS); + + // Nullify global allocator + CLTensorAllocator::set_global_allocator(nullptr); +} + /** Validates import memory interface when importing cl buffer objects */ TEST_CASE(ImportMemoryBuffer, framework::DatasetMode::ALL) { @@ -79,31 +180,31 @@ TEST_CASE(ImportMemoryBuffer, framework::DatasetMode::ALL) // Negative case : Import nullptr CLTensor t1; t1.allocator()->init(info); - ARM_COMPUTE_EXPECT(!bool(t1.allocator()->import_memory(cl::Buffer())), framework::LogLevel::ERRORS); - ARM_COMPUTE_EXPECT(t1.info()->is_resizable(), framework::LogLevel::ERRORS); + ARM_COMPUTE_ASSERT(!bool(t1.allocator()->import_memory(cl::Buffer()))); + ARM_COMPUTE_ASSERT(t1.info()->is_resizable()); // Negative case : Import memory to a tensor that is memory managed CLTensor t2; MemoryGroup mg; t2.allocator()->set_associated_memory_group(&mg); - ARM_COMPUTE_EXPECT(!bool(t2.allocator()->import_memory(buf)), framework::LogLevel::ERRORS); - ARM_COMPUTE_EXPECT(t2.info()->is_resizable(), framework::LogLevel::ERRORS); + ARM_COMPUTE_ASSERT(!bool(t2.allocator()->import_memory(buf))); + ARM_COMPUTE_ASSERT(t2.info()->is_resizable()); // Negative case : Invalid buffer size CLTensor t3; const TensorInfo info_neg(TensorShape(32U, 16U, 3U), 1, DataType::F32); t3.allocator()->init(info_neg); - ARM_COMPUTE_EXPECT(!bool(t3.allocator()->import_memory(buf)), framework::LogLevel::ERRORS); - ARM_COMPUTE_EXPECT(t3.info()->is_resizable(), framework::LogLevel::ERRORS); + ARM_COMPUTE_ASSERT(!bool(t3.allocator()->import_memory(buf))); + ARM_COMPUTE_ASSERT(t3.info()->is_resizable()); // Positive case : Set raw pointer CLTensor t4; t4.allocator()->init(info); - ARM_COMPUTE_EXPECT(bool(t4.allocator()->import_memory(buf)), framework::LogLevel::ERRORS); - ARM_COMPUTE_EXPECT(!t4.info()->is_resizable(), framework::LogLevel::ERRORS); + ARM_COMPUTE_ASSERT(bool(t4.allocator()->import_memory(buf))); + ARM_COMPUTE_ASSERT(!t4.info()->is_resizable()); ARM_COMPUTE_EXPECT(t4.cl_buffer().get() == buf.get(), framework::LogLevel::ERRORS); t4.allocator()->free(); - ARM_COMPUTE_EXPECT(t4.info()->is_resizable(), framework::LogLevel::ERRORS); + ARM_COMPUTE_ASSERT(t4.info()->is_resizable()); ARM_COMPUTE_EXPECT(t4.cl_buffer().get() != buf.get(), framework::LogLevel::ERRORS); } @@ -135,14 +236,14 @@ TEST_CASE(ImportMemoryMalloc, framework::DatasetMode::ALL) const size_t total_size_in_bytes = tensor.info()->total_size(); const size_t alignment = CLKernelLibrary::get().get_device().getInfo<CL_DEVICE_GLOBAL_MEM_CACHELINE_SIZE>(); size_t space = total_size_in_bytes + alignment; - auto raw_data = support::cpp14::make_unique<uint8_t[]>(space); + auto raw_data = std::make_unique<uint8_t[]>(space); void *aligned_ptr = raw_data.get(); - support::cpp11::align(alignment, total_size_in_bytes, aligned_ptr, space); + std::align(alignment, total_size_in_bytes, aligned_ptr, space); cl::Buffer wrapped_buffer(import_malloc_memory_helper(aligned_ptr, total_size_in_bytes)); - ARM_COMPUTE_EXPECT(bool(tensor.allocator()->import_memory(wrapped_buffer)), framework::LogLevel::ERRORS); - ARM_COMPUTE_EXPECT(!tensor.info()->is_resizable(), framework::LogLevel::ERRORS); + ARM_COMPUTE_ASSERT(bool(tensor.allocator()->import_memory(wrapped_buffer))); + ARM_COMPUTE_ASSERT(!tensor.info()->is_resizable()); // Fill tensor std::uniform_real_distribution<float> distribution(-5.f, 5.f); @@ -205,12 +306,12 @@ TEST_CASE(ImportMemoryMappedFile, framework::DatasetMode::ALL) // Map file utils::mmap_io::MMappedFile mmapped_file("test_mmap_import.bin", 0 /** Whole file */, 0); - ARM_COMPUTE_EXPECT(mmapped_file.is_mapped(), framework::LogLevel::ERRORS); + ARM_COMPUTE_ASSERT(mmapped_file.is_mapped()); unsigned char *data = mmapped_file.data(); cl::Buffer wrapped_buffer(import_malloc_memory_helper(data, total_size_in_bytes)); - ARM_COMPUTE_EXPECT(bool(tensor.allocator()->import_memory(wrapped_buffer)), framework::LogLevel::ERRORS); - ARM_COMPUTE_EXPECT(!tensor.info()->is_resizable(), framework::LogLevel::ERRORS); + ARM_COMPUTE_ASSERT(bool(tensor.allocator()->import_memory(wrapped_buffer))); + ARM_COMPUTE_ASSERT(!tensor.info()->is_resizable()); // Fill tensor std::uniform_real_distribution<float> distribution(-5.f, 5.f); @@ -233,7 +334,7 @@ TEST_CASE(ImportMemoryMappedFile, framework::DatasetMode::ALL) // Release resources tensor.allocator()->free(); - ARM_COMPUTE_EXPECT(tensor.info()->is_resizable(), framework::LogLevel::ERRORS); + ARM_COMPUTE_ASSERT(tensor.info()->is_resizable()); } } #endif // !defined(BARE_METAL) diff --git a/tests/validation/CL/UNIT/Tuner.cpp b/tests/validation/CL/UNIT/Tuner.cpp deleted file mode 100644 index 6345b7cd11..0000000000 --- a/tests/validation/CL/UNIT/Tuner.cpp +++ /dev/null @@ -1,78 +0,0 @@ -/* - * Copyright (c) 2018-2020 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 "arm_compute/core/CL/kernels/CLDirectConvolutionLayerKernel.h" -#include "arm_compute/runtime/CL/CLScheduler.h" -#include "arm_compute/runtime/CL/CLTensor.h" -#include "arm_compute/runtime/CL/tuners/BifrostTuner.h" -#include "tests/Utils.h" -#include "tests/framework/Asserts.h" -#include "tests/framework/Macros.h" - -namespace arm_compute -{ -namespace test -{ -namespace validation -{ -TEST_SUITE(CL) -TEST_SUITE(UNIT) -TEST_SUITE(Tuner) - -/** Validates static tuning of Bifrost tuner */ -TEST_CASE(BifrostTunerSimple, framework::DatasetMode::ALL) -{ - // Create tuner - tuners::BifrostTuner tuner; - - // Create tensors - auto src = create_tensor<CLTensor>(TensorShape(13U, 13U, 16U), DataType::F32); - auto weights = create_tensor<CLTensor>(TensorShape(3U, 3U, 16U, 3U), DataType::F32); - auto bias = create_tensor<CLTensor>(TensorShape(3U), DataType::F32); - auto dst = create_tensor<CLTensor>(TensorShape(13U, 13U, 3U), DataType::F32); - - // Create kernel - cl::NDRange fake_lws(2000); - CLDirectConvolutionLayerKernel conv; - conv.set_target(GPUTarget::G72); - - // Configure - conv.configure(&src, &weights, &bias, &dst, PadStrideInfo(1, 1, 1, 1)); - - // Hard-wire lws to kernel and validate lws - conv.set_lws_hint(fake_lws); - ARM_COMPUTE_EXPECT(conv.lws_hint()[0] == 2000, framework::LogLevel::ERRORS); - - // Tune kernel and validate - tuner.tune_kernel_static(conv); - ARM_COMPUTE_EXPECT(conv.lws_hint()[0] != 2000, framework::LogLevel::ERRORS); - - // Clear tuner - CLScheduler::get().default_init(); -} -TEST_SUITE_END() -TEST_SUITE_END() -TEST_SUITE_END() -} // namespace validation -} // namespace test -} // namespace arm_compute diff --git a/tests/validation/CL/UNIT/WeightsRetention.cpp b/tests/validation/CL/UNIT/WeightsRetention.cpp index 9a6fddcc80..357c88af10 100644 --- a/tests/validation/CL/UNIT/WeightsRetention.cpp +++ b/tests/validation/CL/UNIT/WeightsRetention.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2018-2020 ARM Limited. + * Copyright (c) 2018-2021 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -22,6 +22,7 @@ * SOFTWARE. */ #include "arm_compute/runtime/CL/functions/CLFullyConnectedLayer.h" +#include "src/core/CL/kernels/CLFillBorderKernel.h" #include "tests/AssetsLibrary.h" #include "tests/CL/CLAccessor.h" #include "tests/Globals.h" |