diff options
Diffstat (limited to 'tests/validation/CL/GEMMMatrixMultiplyReshaped.cpp')
| -rw-r--r-- | tests/validation/CL/GEMMMatrixMultiplyReshaped.cpp | 768 |
1 files changed, 715 insertions, 53 deletions
diff --git a/tests/validation/CL/GEMMMatrixMultiplyReshaped.cpp b/tests/validation/CL/GEMMMatrixMultiplyReshaped.cpp index 833a9240bf..b06e4bf213 100644 --- a/tests/validation/CL/GEMMMatrixMultiplyReshaped.cpp +++ b/tests/validation/CL/GEMMMatrixMultiplyReshaped.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2018-2020 ARM Limited. + * Copyright (c) 2018-2023 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -21,14 +21,14 @@ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE * SOFTWARE. */ -#include "arm_compute/core/CL/kernels/CLGEMMMatrixMultiplyReshapedKernel.h" -#include "arm_compute/core/CL/kernels/CLGEMMReshapeLHSMatrixKernel.h" -#include "arm_compute/core/CL/kernels/CLGEMMReshapeRHSMatrixKernel.h" #include "arm_compute/core/KernelDescriptors.h" #include "arm_compute/core/Types.h" #include "arm_compute/core/utils/misc/ShapeCalculator.h" #include "arm_compute/runtime/CL/CLTensor.h" #include "arm_compute/runtime/CL/CLTensorAllocator.h" +#include "src/gpu/cl/kernels/ClGemmMatrixMultiplyReshapedKernel.h" +#include "src/gpu/cl/kernels/ClGemmReshapeLhsMatrixKernel.h" +#include "src/gpu/cl/kernels/ClGemmReshapeRhsMatrixKernel.h" #include "tests/CL/CLAccessor.h" #include "tests/CL/Helper.h" #include "tests/PaddingCalculator.h" @@ -46,15 +46,16 @@ namespace test namespace validation { using namespace arm_compute::misc::shape_calculator; +using namespace arm_compute::opencl::kernels; -// Create function for CLGEMMReshapeLHSMatrixKernel -using CLGEMMReshapeLHSMatrix = CLSynthetizeFunction<CLGEMMReshapeLHSMatrixKernel>; +// Create function for ClGemmReshapeLhsMatrixKernel +using CLGEMMReshapeLHSMatrix = CLSynthetizeOperator<ClGemmReshapeLhsMatrixKernel>; -// Create function for CLGEMMReshapeRHSMatrixKernel -using CLGEMMReshapeRHSMatrix = CLSynthetizeFunction<CLGEMMReshapeRHSMatrixKernel>; +// Create function for ClGemmReshapeRhsMatrixKernel +using CLGEMMReshapeRHSMatrix = CLSynthetizeOperator<ClGemmReshapeRhsMatrixKernel>; -// Create function for CLGEMMMatrixMultiplyReshapedKernel -using CLGEMMMatrixMultiplyReshaped = CLSynthetizeFunction<CLGEMMMatrixMultiplyReshapedKernel>; +// Create function for ClGemmMatrixMultiplyReshapedKernel +using CLGEMMMatrixMultiplyReshaped = CLSynthetizeOperator<ClGemmMatrixMultiplyReshapedKernel>; // Fixture for CLGEMMMatrixMultiplyReshaped template <typename T> @@ -109,6 +110,7 @@ const auto b_values = framework::dataset::make("batch_size", 2, 3); const auto act_values = framework::dataset::make("Activation", { ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU, 8.f, 2.f), + ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::ELU), }); /** Alpha values to test - Precommit */ @@ -139,19 +141,25 @@ const auto a_values_nightly = framework::dataset::make("alpha", {1.0f} ); const auto beta_values_nightly = framework::dataset::make("beta", {1.0f} ); /** M0 values to test - Nightly */ -const auto m0_values_nightly = framework::dataset::make("M0", { 2, 3, 4, 8 }); +const auto m0_values_nightly = framework::dataset::make("M0", { 8 }); /** N0 values to test - Nightly */ -const auto n0_values_nightly = framework::dataset::make("N0", { 2, 3, 4, 8 }); +const auto n0_values_nightly = framework::dataset::make("N0", { 8 }); /** K0 values to test - Nightly */ -const auto k0_values_nightly = framework::dataset::make("K0", { 2, 3, 4, 8 }); +const auto k0_values_nightly = framework::dataset::make("K0", { 4 }); + +/** N0 values to test with export to OpenCL image object - Nightly */ +const auto n0_export_to_cl_image_values_nightly = framework::dataset::make("N0", { 4, 8, 16 }); + +/** K0 values to test with export to OpenCL image object - Nightly */ +const auto k0_export_to_cl_image_values_nightly = framework::dataset::make("K0", { 4, 8, 16 }); /** V0 values to test - Nightly */ -const auto v0_values_nightly = framework::dataset::make("V0", 1, 4); +const auto v0_values_nightly = framework::dataset::make("V0", 1, 3); /** H0 values to test - Nightly */ -const auto h0_values_nightly = framework::dataset::make("H0", 1, 4); +const auto h0_values_nightly = framework::dataset::make("H0", 1, 3); /** Interleave values to test with LHS matrix */ const auto i_values_lhs = framework::dataset::make("interleave_lhs", { true, false }); @@ -164,6 +172,7 @@ const auto broadcast_bias_values = framework::dataset::make("broadcast_bias", { /** LHS transposed values */ const auto lhs_transpose_values = framework::dataset::make("lhs_transpose", { false, true } ); + } // namespace TEST_SUITE(CL) @@ -224,14 +233,14 @@ DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(zip(zip(zip(zi })), framework::dataset::make("RHSMInfo",{ - GEMMRHSMatrixInfo(4,4,1,true,true), - GEMMRHSMatrixInfo(4,4,1, true,true), - GEMMRHSMatrixInfo(4,4,1,true,true), - GEMMRHSMatrixInfo(2,2,1,true,false), - GEMMRHSMatrixInfo(2,2,1,true,false), - GEMMRHSMatrixInfo(4,4,1,true,true), - GEMMRHSMatrixInfo(4,4,1,true,true), - GEMMRHSMatrixInfo(4,4,2,true,false), + GEMMRHSMatrixInfo(4,4,1,true,true,false), + GEMMRHSMatrixInfo(4,4,1,true,true,false), + GEMMRHSMatrixInfo(4,4,1,true,true,false), + GEMMRHSMatrixInfo(2,2,1,true,false,false), + GEMMRHSMatrixInfo(2,2,1,true,false,false), + GEMMRHSMatrixInfo(4,4,1,true,true,false), + GEMMRHSMatrixInfo(4,4,1,true,true,false), + GEMMRHSMatrixInfo(4,4,2,true,false,false), })), @@ -244,11 +253,12 @@ DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(zip(zip(zip(zi false /**< reinterpret the input as 3D */, true /**< Flag used to broadcast the bias addition */, false /**< wider accumm */, + false /**< has pad y */, ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU, 1 /**< Multiplication factor for the width of the 1xW transposed block */, 1 /**< Multiplication factor for the height of the 4x4 interleaved block */, GEMMLHSMatrixInfo(4,4,1,false,true), - GEMMRHSMatrixInfo(4,4,1,true,true), + GEMMRHSMatrixInfo(4,4,1,true,true,false), 0 /**< Offset to be added to each element of the matrix A */, 0 /**< Offset to be added to each element of the matrix B */), @@ -258,11 +268,12 @@ DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(zip(zip(zip(zi false /**< reinterpret the input as 3D */, true /**< Flag used to broadcast the bias addition */, false /**< wider accumm */, + false /**< has pad y */, ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU, 1 /**< Multiplication factor for the width of the 1xW transposed block */, 1 /**< Multiplication factor for the height of the 4x4 interleaved block */, GEMMLHSMatrixInfo(4,4,1,false,true), - GEMMRHSMatrixInfo(4,4,1,true,true), + GEMMRHSMatrixInfo(4,4,1,true,true,false), 0 /**< Offset to be added to each element of the matrix A */, 0 /**< Offset to be added to each element of the matrix B */), GEMMKernelInfo(), @@ -275,11 +286,12 @@ DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(zip(zip(zip(zi false /**< reinterpret the input as 3D */, false /**< Flag used to broadcast the bias addition */, false /**< wider accumm */, + false /**< has pad y */, ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU, 1 /**< Multiplication factor for the width of the 1xW transposed block */, 1 /**< Multiplication factor for the height of the 4x4 interleaved block */, GEMMLHSMatrixInfo(4,4,1,false,true), - GEMMRHSMatrixInfo(4,4,1,true,true), + GEMMRHSMatrixInfo(4,4,1,true,true,false), 0 /**< Offset to be added to each element of the matrix A */, 0 /**< Offset to be added to each element of the matrix B */), @@ -290,11 +302,12 @@ DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(zip(zip(zip(zi false /**< reinterpret the input as 3D */, false /**< Flag used to broadcast the bias addition */, true /**< wider accumm */, + true /**< has pad y */, ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU, 1 /**< Multiplication factor for the width of the 1xW transposed block */, 1 /**< Multiplication factor for the height of the 4x4 interleaved block */, GEMMLHSMatrixInfo(4,4,1,false,true), - GEMMRHSMatrixInfo(4,4,1,true,true), + GEMMRHSMatrixInfo(4,4,1,true,true,false), 0 /**< Offset to be added to each element of the matrix A */, 0 /**< Offset to be added to each element of the matrix B */), @@ -304,18 +317,19 @@ DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(zip(zip(zip(zi false /**< reinterpret the input as 3D */, false /**< Flag used to broadcast the bias addition */, false /**< wider accumm */, + false /**< has pad y */, ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU, 1 /**< Multiplication factor for the width of the 1xW transposed block */, 1 /**< Multiplication factor for the height of the 4x4 interleaved block */, GEMMLHSMatrixInfo(4,4,1,false,true), - GEMMRHSMatrixInfo(4,4,2,true,false), + GEMMRHSMatrixInfo(4,4,2,true,false,false), 0 /**< Offset to be added to each element of the matrix A */, 0 /**< Offset to be added to each element of the matrix B */), })), framework::dataset::make("Expected", { true, true, false, false, false, true, true,true})), input0_info ,input1_info, input2_info, output_info, lhs_info, rhs_info, gemm_info, expected) { - ARM_COMPUTE_EXPECT(bool(CLGEMMMatrixMultiplyReshapedKernel::validate(&input0_info.clone()->set_is_resizable(true), + ARM_COMPUTE_EXPECT(bool(ClGemmMatrixMultiplyReshapedKernel::validate(&input0_info.clone()->set_is_resizable(true), &input1_info.clone()->set_is_resizable(true), &input2_info.clone()->set_is_resizable(true), &output_info.clone()->set_is_resizable(true),1.f,1.f, @@ -323,11 +337,12 @@ DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(zip(zip(zip(zi rhs_info, gemm_info)) == expected, framework::LogLevel::ERRORS); } + TEST_SUITE(Float) TEST_SUITE(FP32) FIXTURE_DATA_TEST_CASE(RunSmall, CLGEMMMatrixMultiplyReshapedFixture<float>, framework::DatasetMode::ALL, - combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine( + combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine( m_values, n_values), k_values), @@ -339,6 +354,7 @@ FIXTURE_DATA_TEST_CASE(RunSmall, CLGEMMMatrixMultiplyReshapedFixture<float>, fra h0_values_precommit), i_values_lhs), i_values_rhs), + framework::dataset::make("export_to_cl_image_rhs", false)), framework::dataset::make("DataType", DataType::F32)), a_values_precommit), beta_values_precommit), @@ -347,11 +363,19 @@ FIXTURE_DATA_TEST_CASE(RunSmall, CLGEMMMatrixMultiplyReshapedFixture<float>, fra act_values)) { // Validate output - validate(CLAccessor(_target), _reference, rel_tolerance_f32, 0.f, abs_tolerance_f32); + if(validate_result) + { + validate(CLAccessor(_target), _reference, rel_tolerance_f32, 0.f, abs_tolerance_f32); + } + else + { + ARM_COMPUTE_TEST_INFO("cl_khr_image2d_from_buffer not supported. TEST skipped"); + framework::ARM_COMPUTE_PRINT_INFO(); + } } FIXTURE_DATA_TEST_CASE(RunLarge, CLGEMMMatrixMultiplyReshapedFixture<float>, framework::DatasetMode::DISABLED, - combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine( + combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine( m_values, n_values), k_values), @@ -363,6 +387,7 @@ FIXTURE_DATA_TEST_CASE(RunLarge, CLGEMMMatrixMultiplyReshapedFixture<float>, fra h0_values_nightly), i_values_lhs), i_values_rhs), + framework::dataset::make("export_to_cl_image_rhs", false)), framework::dataset::make("DataType", DataType::F32)), a_values_nightly), beta_values_nightly), @@ -371,11 +396,19 @@ FIXTURE_DATA_TEST_CASE(RunLarge, CLGEMMMatrixMultiplyReshapedFixture<float>, fra act_values)) { // Validate output - validate(CLAccessor(_target), _reference, rel_tolerance_f32, 0.f, abs_tolerance_f32); + if(validate_result) + { + validate(CLAccessor(_target), _reference, rel_tolerance_f32, 0.f, abs_tolerance_f32); + } + else + { + ARM_COMPUTE_TEST_INFO("cl_khr_image2d_from_buffer not supported. TEST skipped"); + framework::ARM_COMPUTE_PRINT_INFO(); + } } FIXTURE_DATA_TEST_CASE(RunSmall3D, CLGEMMMatrixMultiplyReshaped3DFixture<float>, framework::DatasetMode::ALL, - combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine( + combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine( m_w_values, m_h_values), n_values), @@ -388,6 +421,7 @@ FIXTURE_DATA_TEST_CASE(RunSmall3D, CLGEMMMatrixMultiplyReshaped3DFixture<float>, h0_values_precommit), i_values_lhs), i_values_rhs), + framework::dataset::make("export_to_cl_image_rhs", false)), framework::dataset::make("DataType", DataType::F32)), a_values_precommit), beta_values_precommit), @@ -395,11 +429,19 @@ FIXTURE_DATA_TEST_CASE(RunSmall3D, CLGEMMMatrixMultiplyReshaped3DFixture<float>, act_values)) { // Validate output - validate(CLAccessor(_target), _reference, rel_tolerance_f32, 0.f, abs_tolerance_f32); + if(validate_result) + { + validate(CLAccessor(_target), _reference, rel_tolerance_f32, 0.f, abs_tolerance_f32); + } + else + { + ARM_COMPUTE_TEST_INFO("cl_khr_image2d_from_buffer not supported. TEST skipped"); + framework::ARM_COMPUTE_PRINT_INFO(); + } } FIXTURE_DATA_TEST_CASE(RunLarge3D, CLGEMMMatrixMultiplyReshaped3DFixture<float>, framework::DatasetMode::DISABLED, - combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine( + combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine( m_w_values, m_h_values), n_values), @@ -412,6 +454,7 @@ FIXTURE_DATA_TEST_CASE(RunLarge3D, CLGEMMMatrixMultiplyReshaped3DFixture<float>, h0_values_nightly), i_values_lhs), i_values_rhs), + framework::dataset::make("export_to_cl_image_rhs", false)), framework::dataset::make("DataType", DataType::F32)), a_values_nightly), beta_values_nightly), @@ -419,14 +462,291 @@ FIXTURE_DATA_TEST_CASE(RunLarge3D, CLGEMMMatrixMultiplyReshaped3DFixture<float>, act_values)) { // Validate output - validate(CLAccessor(_target), _reference, rel_tolerance_f32, 0.f, abs_tolerance_f32); + if(validate_result) + { + validate(CLAccessor(_target), _reference, rel_tolerance_f32, 0.f, abs_tolerance_f32); + } + else + { + ARM_COMPUTE_TEST_INFO("cl_khr_image2d_from_buffer not supported. TEST skipped"); + framework::ARM_COMPUTE_PRINT_INFO(); + } +} + +TEST_SUITE(ExportToCLImage) +DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(zip(zip(zip(zip( + framework::dataset::make("Input0Info", { TensorInfo(TensorShape(256U, 16U, 2U), 1, DataType::F32), // OK or incorrect if cl_khr_image2d_from_buffer not supported + TensorInfo(TensorShape(256U, 16U, 2U), 1, DataType::F32), // OK or incorrect if cl_khr_image2d_from_buffer not supported + TensorInfo(TensorShape(256U, 16U, 2U), 1, DataType::F32), // OK or incorrect if cl_khr_image2d_from_buffer not supported + TensorInfo(TensorShape(256U, 16U, 2U), 1, DataType::F32), // Incorrect k0 + TensorInfo(TensorShape(256U, 16U, 2U), 1, DataType::F32), // Incorrect n0 + + }), + framework::dataset::make("Input1Info",{ TensorInfo(TensorShape(256U, 16U, 2U), 1, DataType::F32), + TensorInfo(TensorShape(256U, 16U, 2U), 1, DataType::F32), + TensorInfo(TensorShape(512U, 8U, 2U), 1, DataType::F32), + TensorInfo(TensorShape(256U, 16U, 2U), 1, DataType::F32), + TensorInfo(TensorShape(128U, 32U, 2U), 1, DataType::F32), + + })), + framework::dataset::make("Input2Info", { TensorInfo(TensorShape(64U), 1, DataType::F32), + TensorInfo(TensorShape(64U), 1, DataType::F32), + TensorInfo(TensorShape(64U), 1, DataType::F32), + TensorInfo(TensorShape(64U), 1, DataType::F32), + TensorInfo(TensorShape(64U), 1, DataType::F32), + + })), + framework::dataset::make("OutputInfo",{ TensorInfo(TensorShape(64U, 64U, 2U), 1, DataType::F32), + TensorInfo(TensorShape(64U, 64U, 2U), 1, DataType::F32), + TensorInfo(TensorShape(64U, 64U, 2U), 1, DataType::F32), + TensorInfo(TensorShape(64U, 64U, 2U), 1, DataType::F32), + TensorInfo(TensorShape(64U, 64U, 2U), 1, DataType::F32), + TensorInfo(TensorShape(64U, 64U, 2U), 1, DataType::F32), + + })), + framework::dataset::make("LHSMInfo",{ + GEMMLHSMatrixInfo(4, 4, 1, false, true), + GEMMLHSMatrixInfo(4, 8, 1, false, true), + GEMMLHSMatrixInfo(4, 4, 1, false, true), + GEMMLHSMatrixInfo(4, 2, 1, false, false), + GEMMLHSMatrixInfo(4, 4, 1, false, false), + + })), + framework::dataset::make("RHSMInfo",{ + GEMMRHSMatrixInfo(4, 4, 1, true, true, true), + GEMMRHSMatrixInfo(4, 8, 1, true, true, true), + GEMMRHSMatrixInfo(8, 4, 1, true, true, true), + GEMMRHSMatrixInfo(4, 2, 1, true, false, true), + GEMMRHSMatrixInfo(2, 4, 1, true, false, true), + })), + framework::dataset::make("GEMMInfo",{GEMMKernelInfo( 64 /**<M Number of LHS rows*/, + 64 /**<N Number of RHS columns*/, + 64 /**<K Number of LHS columns or RHS rows */, 0 /**< Depth of the output tensor in case is reinterpreted as 3D */, + false /**< reinterpret the input as 3D */, + true /**< Flag used to broadcast the bias addition */, + false /**< wider accumm */, + false /**< has pad y */, + ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU, + 1 /**< Multiplication factor for the width of the 1xW transposed block */, + 1 /**< Multiplication factor for the height of the 4x4 interleaved block */, + GEMMLHSMatrixInfo(), + GEMMRHSMatrixInfo(), + 0 /**< Offset to be added to each element of the matrix A */, + 0 /**< Offset to be added to each element of the matrix B */), + GEMMKernelInfo( 64 /**<M Number of LHS rows*/, + 64 /**<N Number of RHS columns*/, + 64 /**<K Number of LHS columns or RHS rows */, 0 /**< Depth of the output tensor in case is reinterpreted as 3D */, + false /**< reinterpret the input as 3D */, + true /**< Flag used to broadcast the bias addition */, + false /**< wider accumm */, + false /**< has pad y */, + ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU, + 1 /**< Multiplication factor for the width of the 1xW transposed block */, + 1 /**< Multiplication factor for the height of the 4x4 interleaved block */, + GEMMLHSMatrixInfo(), + GEMMRHSMatrixInfo(), + 0 /**< Offset to be added to each element of the matrix A */, + 0 /**< Offset to be added to each element of the matrix B */), + GEMMKernelInfo( 64 /**<M Number of LHS rows*/, + 64 /**<N Number of RHS columns*/, + 64 /**<K Number of LHS columns or RHS rows */, 0 /**< Depth of the output tensor in case is reinterpreted as 3D */, + false /**< reinterpret the input as 3D */, + true /**< Flag used to broadcast the bias addition */, + false /**< wider accumm */, + false /**< has pad y */, + ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU, + 1 /**< Multiplication factor for the width of the 1xW transposed block */, + 1 /**< Multiplication factor for the height of the 4x4 interleaved block */, + GEMMLHSMatrixInfo(), + GEMMRHSMatrixInfo(), + 0 /**< Offset to be added to each element of the matrix A */, + 0 /**< Offset to be added to each element of the matrix B */), + + GEMMKernelInfo( 64 /**<M Number of LHS rows*/, + 64 /**<N Number of RHS columns*/, + 64 /**<K Number of LHS columns or RHS rows */, 0 /**< Depth of the output tensor in case is reinterpreted as 3D */, + false /**< reinterpret the input as 3D */, + true /**< Flag used to broadcast the bias addition */, + false /**< wider accumm */, + false /**< has pad y */, + ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU, + 1 /**< Multiplication factor for the width of the 1xW transposed block */, + 1 /**< Multiplication factor for the height of the 4x4 interleaved block */, + GEMMLHSMatrixInfo(), + GEMMRHSMatrixInfo(), + 0 /**< Offset to be added to each element of the matrix A */, + 0 /**< Offset to be added to each element of the matrix B */), + GEMMKernelInfo( 64 /**<M Number of LHS rows*/, + 64 /**<N Number of RHS columns*/, + 64 /**<K Number of LHS columns or RHS rows */, 0 /**< Depth of the output tensor in case is reinterpreted as 3D */, + false /**< reinterpret the input as 3D */, + true /**< Flag used to broadcast the bias addition */, + false /**< wider accumm */, + false /**< has pad y */, + ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU, + 1 /**< Multiplication factor for the width of the 1xW transposed block */, + 1 /**< Multiplication factor for the height of the 4x4 interleaved block */, + GEMMLHSMatrixInfo(), + GEMMRHSMatrixInfo(), + 0 /**< Offset to be added to each element of the matrix A */, + 0 /**< Offset to be added to each element of the matrix B */) + })), + framework::dataset::make("Expected", { true, + true, + true, + false, + true})), + input0_info ,input1_info, input2_info, output_info, lhs_info, rhs_info, gemm_info, expected) +{ + ARM_COMPUTE_EXPECT(bool(ClGemmMatrixMultiplyReshapedKernel::validate(&input0_info.clone()->set_is_resizable(true), + &input1_info.clone()->set_is_resizable(true), + &input2_info.clone()->set_is_resizable(true), + &output_info.clone()->set_is_resizable(true),1.f,1.f, + lhs_info, + rhs_info, + gemm_info)) == (expected && image2d_from_buffer_supported(CLKernelLibrary::get().get_device())), framework::LogLevel::ERRORS); +} + +FIXTURE_DATA_TEST_CASE(RunSmall, CLGEMMMatrixMultiplyReshapedFixture<float>, framework::DatasetMode::ALL, + combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine( + m_values, + n_values), + k_values), + b_values), + m0_values_precommit), + n0_values_precommit), + k0_values_precommit), + v0_values_precommit), + h0_values_precommit), + i_values_lhs), + i_values_rhs), + framework::dataset::make("export_to_cl_image_rhs", true)), + framework::dataset::make("DataType", DataType::F32)), + a_values_precommit), + beta_values_precommit), + broadcast_bias_values), + lhs_transpose_values), + act_values)) +{ + // Validate output only if validate() is successful + if(validate_result) + { + validate(CLAccessor(_target), _reference, rel_tolerance_f32, 0.f, abs_tolerance_f32); + } + else + { + ARM_COMPUTE_TEST_INFO("cl_khr_image2d_from_buffer not supported. TEST skipped"); + framework::ARM_COMPUTE_PRINT_INFO(); + } + +} + +FIXTURE_DATA_TEST_CASE(RunLarge, CLGEMMMatrixMultiplyReshapedFixture<float>, framework::DatasetMode::NIGHTLY, + combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine( + m_values, + n_values), + k_values), + b_values), + m0_values_nightly), + n0_export_to_cl_image_values_nightly), + k0_export_to_cl_image_values_nightly), + v0_values_nightly), + h0_values_nightly), + i_values_lhs), + i_values_rhs), + framework::dataset::make("export_to_cl_image_rhs", true)), + framework::dataset::make("DataType", DataType::F32)), + a_values_nightly), + beta_values_nightly), + broadcast_bias_values), + lhs_transpose_values), + act_values)) +{ + // Validate output only if validate() is successful + if(validate_result) + { + validate(CLAccessor(_target), _reference, rel_tolerance_f32, 0.f, abs_tolerance_f32); + } + else + { + ARM_COMPUTE_TEST_INFO("cl_khr_image2d_from_buffer not supported. TEST skipped"); + framework::ARM_COMPUTE_PRINT_INFO(); + } +} + +FIXTURE_DATA_TEST_CASE(RunSmall3D, CLGEMMMatrixMultiplyReshaped3DFixture<float>, framework::DatasetMode::ALL, + combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine( + m_w_values, + m_h_values), + n_values), + k_values), + b_values), + m0_values_precommit), + n0_values_precommit), + k0_values_precommit), + v0_values_precommit), + h0_values_precommit), + i_values_lhs), + i_values_rhs), + framework::dataset::make("export_to_cl_image_rhs", true)), + framework::dataset::make("DataType", DataType::F32)), + a_values_precommit), + beta_values_precommit), + lhs_transpose_values), + act_values)) +{ + // Validate output only if validate() is successful + if(validate_result) + { + validate(CLAccessor(_target), _reference, rel_tolerance_f32, 0.f, abs_tolerance_f32); + } + else + { + ARM_COMPUTE_TEST_INFO("cl_khr_image2d_from_buffer not supported. TEST skipped"); + framework::ARM_COMPUTE_PRINT_INFO(); + } +} + +FIXTURE_DATA_TEST_CASE(RunLarge3D, CLGEMMMatrixMultiplyReshaped3DFixture<float>, framework::DatasetMode::NIGHTLY, + combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine( + m_w_values, + m_h_values), + n_values), + k_values), + b_values), + m0_values_nightly), + n0_export_to_cl_image_values_nightly), + k0_export_to_cl_image_values_nightly), + v0_values_nightly), + h0_values_nightly), + i_values_lhs), + i_values_rhs), + framework::dataset::make("export_to_cl_image_rhs", true)), + framework::dataset::make("DataType", DataType::F32)), + a_values_nightly), + beta_values_nightly), + lhs_transpose_values), + act_values)) +{ + // Validate output only if validate() is successful + if(validate_result) + { + validate(CLAccessor(_target), _reference, rel_tolerance_f32, 0.f, abs_tolerance_f32); + } + else + { + ARM_COMPUTE_TEST_INFO("cl_khr_image2d_from_buffer not supported. TEST skipped"); + framework::ARM_COMPUTE_PRINT_INFO(); + } } + +TEST_SUITE_END() // ExportToCLImage TEST_SUITE_END() // FP32 TEST_SUITE(FP16) FIXTURE_DATA_TEST_CASE(RunSmall, CLGEMMMatrixMultiplyReshapedFixture<half>, framework::DatasetMode::ALL, - combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine( + combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine( m_values, n_values), k_values), @@ -438,6 +758,7 @@ FIXTURE_DATA_TEST_CASE(RunSmall, CLGEMMMatrixMultiplyReshapedFixture<half>, fram h0_values_precommit), i_values_lhs), i_values_rhs), + framework::dataset::make("export_to_cl_image_rhs", false)), framework::dataset::make("DataType", DataType::F16)), a_values_precommit), beta_values_precommit), @@ -446,11 +767,19 @@ FIXTURE_DATA_TEST_CASE(RunSmall, CLGEMMMatrixMultiplyReshapedFixture<half>, fram act_values)) { // Validate output - validate(CLAccessor(_target), _reference, rel_tolerance_f16, 0.f, abs_tolerance_f16); + if(validate_result) + { + validate(CLAccessor(_target), _reference, rel_tolerance_f16, 0.f, abs_tolerance_f16); + } + else + { + ARM_COMPUTE_TEST_INFO("cl_khr_image2d_from_buffer not supported. TEST skipped"); + framework::ARM_COMPUTE_PRINT_INFO(); + } } FIXTURE_DATA_TEST_CASE(RunLarge, CLGEMMMatrixMultiplyReshapedFixture<half>, framework::DatasetMode::DISABLED, - combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine( + combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine( m_values, n_values), k_values), @@ -462,6 +791,7 @@ FIXTURE_DATA_TEST_CASE(RunLarge, CLGEMMMatrixMultiplyReshapedFixture<half>, fram h0_values_nightly), i_values_lhs), i_values_rhs), + framework::dataset::make("export_to_cl_image_rhs", false)), framework::dataset::make("DataType", DataType::F16)), a_values_nightly), beta_values_nightly), @@ -470,11 +800,19 @@ FIXTURE_DATA_TEST_CASE(RunLarge, CLGEMMMatrixMultiplyReshapedFixture<half>, fram act_values)) { // Validate output - validate(CLAccessor(_target), _reference, rel_tolerance_f16, 0.f, abs_tolerance_f16); + if(validate_result) + { + validate(CLAccessor(_target), _reference, rel_tolerance_f16, 0.f, abs_tolerance_f16); + } + else + { + ARM_COMPUTE_TEST_INFO("cl_khr_image2d_from_buffer not supported. TEST skipped"); + framework::ARM_COMPUTE_PRINT_INFO(); + } } FIXTURE_DATA_TEST_CASE(RunSmall3D, CLGEMMMatrixMultiplyReshaped3DFixture<half>, framework::DatasetMode::ALL, - combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine( + combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine( m_w_values, m_h_values), n_values), @@ -487,6 +825,7 @@ FIXTURE_DATA_TEST_CASE(RunSmall3D, CLGEMMMatrixMultiplyReshaped3DFixture<half>, h0_values_precommit), i_values_lhs), i_values_rhs), + framework::dataset::make("export_to_cl_image_rhs", false)), framework::dataset::make("DataType", DataType::F16)), a_values_precommit), beta_values_precommit), @@ -494,11 +833,19 @@ FIXTURE_DATA_TEST_CASE(RunSmall3D, CLGEMMMatrixMultiplyReshaped3DFixture<half>, act_values)) { // Validate output - validate(CLAccessor(_target), _reference, rel_tolerance_f16, 0.f, abs_tolerance_f16); + if(validate_result) + { + validate(CLAccessor(_target), _reference, rel_tolerance_f16, 0.f, abs_tolerance_f16); + } + else + { + ARM_COMPUTE_TEST_INFO("cl_khr_image2d_from_buffer not supported. TEST skipped"); + framework::ARM_COMPUTE_PRINT_INFO(); + } } FIXTURE_DATA_TEST_CASE(RunLarge3D, CLGEMMMatrixMultiplyReshaped3DFixture<half>, framework::DatasetMode::DISABLED, - combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine( + combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine( m_w_values, m_h_values), n_values), @@ -511,6 +858,7 @@ FIXTURE_DATA_TEST_CASE(RunLarge3D, CLGEMMMatrixMultiplyReshaped3DFixture<half>, h0_values_nightly), i_values_lhs), i_values_rhs), + framework::dataset::make("export_to_cl_image_rhs", false)), framework::dataset::make("DataType", DataType::F16)), a_values_nightly), beta_values_nightly), @@ -518,14 +866,291 @@ FIXTURE_DATA_TEST_CASE(RunLarge3D, CLGEMMMatrixMultiplyReshaped3DFixture<half>, act_values)) { // Validate output - validate(CLAccessor(_target), _reference, rel_tolerance_f16, 0.f, abs_tolerance_f16); + if(validate_result) + { + validate(CLAccessor(_target), _reference, rel_tolerance_f16, 0.f, abs_tolerance_f16); + } + else + { + ARM_COMPUTE_TEST_INFO("cl_khr_image2d_from_buffer not supported. TEST skipped"); + framework::ARM_COMPUTE_PRINT_INFO(); + } +} + +TEST_SUITE(ExportToCLImage) +DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(zip(zip(zip(zip( + framework::dataset::make("Input0Info", { TensorInfo(TensorShape(256U, 16U, 2U), 1, DataType::F16), // OK or incorrect if cl_khr_image2d_from_buffer not supported + TensorInfo(TensorShape(256U, 16U, 2U), 1, DataType::F16), // OK or incorrect if cl_khr_image2d_from_buffer not supported + TensorInfo(TensorShape(256U, 16U, 2U), 1, DataType::F16), // OK or incorrect if cl_khr_image2d_from_buffer not supported + TensorInfo(TensorShape(256U, 16U, 2U), 1, DataType::F16), // Incorrect k0 + TensorInfo(TensorShape(256U, 16U, 2U), 1, DataType::F16), // Incorrect n0 + + }), + framework::dataset::make("Input1Info",{ TensorInfo(TensorShape(256U, 16U, 2U), 1, DataType::F16), + TensorInfo(TensorShape(256U, 16U, 2U), 1, DataType::F16), + TensorInfo(TensorShape(512U, 8U, 2U), 1, DataType::F16), + TensorInfo(TensorShape(256U, 16U, 2U), 1, DataType::F16), + TensorInfo(TensorShape(128U, 32U, 2U), 1, DataType::F16), + + })), + framework::dataset::make("Input2Info", { TensorInfo(TensorShape(64U), 1, DataType::F16), + TensorInfo(TensorShape(64U), 1, DataType::F16), + TensorInfo(TensorShape(64U), 1, DataType::F16), + TensorInfo(TensorShape(64U), 1, DataType::F16), + TensorInfo(TensorShape(64U), 1, DataType::F16), + + })), + framework::dataset::make("OutputInfo",{ TensorInfo(TensorShape(64U, 64U, 2U), 1, DataType::F16), + TensorInfo(TensorShape(64U, 64U, 2U), 1, DataType::F16), + TensorInfo(TensorShape(64U, 64U, 2U), 1, DataType::F16), + TensorInfo(TensorShape(64U, 64U, 2U), 1, DataType::F16), + TensorInfo(TensorShape(64U, 64U, 2U), 1, DataType::F16), + TensorInfo(TensorShape(64U, 64U, 2U), 1, DataType::F16), + + })), + framework::dataset::make("LHSMInfo",{ + GEMMLHSMatrixInfo(4, 4, 1, false, true), + GEMMLHSMatrixInfo(4, 8, 1, false, true), + GEMMLHSMatrixInfo(4, 4, 1, false, true), + GEMMLHSMatrixInfo(4, 2, 1, false, false), + GEMMLHSMatrixInfo(4, 4, 1, false, false), + + })), + framework::dataset::make("RHSMInfo",{ + GEMMRHSMatrixInfo(4, 4, 1, true, true, true), + GEMMRHSMatrixInfo(4, 8, 1, true, true, true), + GEMMRHSMatrixInfo(8, 4, 1, true, true, true), + GEMMRHSMatrixInfo(4, 2, 1, true, false, true), + GEMMRHSMatrixInfo(2, 4, 1, true, false, true), + })), + framework::dataset::make("GEMMInfo",{GEMMKernelInfo( 64 /**<M Number of LHS rows*/, + 64 /**<N Number of RHS columns*/, + 64 /**<K Number of LHS columns or RHS rows */, 0 /**< Depth of the output tensor in case is reinterpreted as 3D */, + false /**< reinterpret the input as 3D */, + true /**< Flag used to broadcast the bias addition */, + false /**< wider accumm */, + false /**< has pad y */, + ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU, + 1 /**< Multiplication factor for the width of the 1xW transposed block */, + 1 /**< Multiplication factor for the height of the 4x4 interleaved block */, + GEMMLHSMatrixInfo(), + GEMMRHSMatrixInfo(), + 0 /**< Offset to be added to each element of the matrix A */, + 0 /**< Offset to be added to each element of the matrix B */), + GEMMKernelInfo( 64 /**<M Number of LHS rows*/, + 64 /**<N Number of RHS columns*/, + 64 /**<K Number of LHS columns or RHS rows */, 0 /**< Depth of the output tensor in case is reinterpreted as 3D */, + false /**< reinterpret the input as 3D */, + true /**< Flag used to broadcast the bias addition */, + false /**< wider accumm */, + false /**< has pad y */, + ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU, + 1 /**< Multiplication factor for the width of the 1xW transposed block */, + 1 /**< Multiplication factor for the height of the 4x4 interleaved block */, + GEMMLHSMatrixInfo(), + GEMMRHSMatrixInfo(), + 0 /**< Offset to be added to each element of the matrix A */, + 0 /**< Offset to be added to each element of the matrix B */), + GEMMKernelInfo( 64 /**<M Number of LHS rows*/, + 64 /**<N Number of RHS columns*/, + 64 /**<K Number of LHS columns or RHS rows */, 0 /**< Depth of the output tensor in case is reinterpreted as 3D */, + false /**< reinterpret the input as 3D */, + true /**< Flag used to broadcast the bias addition */, + false /**< wider accumm */, + false /**< has pad y */, + ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU, + 1 /**< Multiplication factor for the width of the 1xW transposed block */, + 1 /**< Multiplication factor for the height of the 4x4 interleaved block */, + GEMMLHSMatrixInfo(), + GEMMRHSMatrixInfo(), + 0 /**< Offset to be added to each element of the matrix A */, + 0 /**< Offset to be added to each element of the matrix B */), + + GEMMKernelInfo( 64 /**<M Number of LHS rows*/, + 64 /**<N Number of RHS columns*/, + 64 /**<K Number of LHS columns or RHS rows */, 0 /**< Depth of the output tensor in case is reinterpreted as 3D */, + false /**< reinterpret the input as 3D */, + true /**< Flag used to broadcast the bias addition */, + false /**< wider accumm */, + false /**< has pad y */, + ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU, + 1 /**< Multiplication factor for the width of the 1xW transposed block */, + 1 /**< Multiplication factor for the height of the 4x4 interleaved block */, + GEMMLHSMatrixInfo(), + GEMMRHSMatrixInfo(), + 0 /**< Offset to be added to each element of the matrix A */, + 0 /**< Offset to be added to each element of the matrix B */), + GEMMKernelInfo( 64 /**<M Number of LHS rows*/, + 64 /**<N Number of RHS columns*/, + 64 /**<K Number of LHS columns or RHS rows */, 0 /**< Depth of the output tensor in case is reinterpreted as 3D */, + false /**< reinterpret the input as 3D */, + true /**< Flag used to broadcast the bias addition */, + false /**< wider accumm */, + false /**< has pad y */, + ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU, + 1 /**< Multiplication factor for the width of the 1xW transposed block */, + 1 /**< Multiplication factor for the height of the 4x4 interleaved block */, + GEMMLHSMatrixInfo(), + GEMMRHSMatrixInfo(), + 0 /**< Offset to be added to each element of the matrix A */, + 0 /**< Offset to be added to each element of the matrix B */) + })), + framework::dataset::make("Expected", { true, + true, + true, + false, + true})), + input0_info ,input1_info, input2_info, output_info, lhs_info, rhs_info, gemm_info, expected) +{ + ARM_COMPUTE_EXPECT(bool(ClGemmMatrixMultiplyReshapedKernel::validate(&input0_info.clone()->set_is_resizable(true), + &input1_info.clone()->set_is_resizable(true), + &input2_info.clone()->set_is_resizable(true), + &output_info.clone()->set_is_resizable(true),1.f,1.f, + lhs_info, + rhs_info, + gemm_info)) == (expected && image2d_from_buffer_supported(CLKernelLibrary::get().get_device())), framework::LogLevel::ERRORS); +} + +FIXTURE_DATA_TEST_CASE(RunSmall, CLGEMMMatrixMultiplyReshapedFixture<half>, framework::DatasetMode::ALL, + combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine( + m_values, + n_values), + k_values), + b_values), + m0_values_precommit), + n0_values_precommit), + k0_values_precommit), + v0_values_precommit), + h0_values_precommit), + i_values_lhs), + i_values_rhs), + framework::dataset::make("export_to_cl_image_rhs", true)), + framework::dataset::make("DataType", DataType::F16)), + a_values_precommit), + beta_values_precommit), + broadcast_bias_values), + lhs_transpose_values), + act_values)) +{ + // Validate output only if validate() is successful + if(validate_result) + { + validate(CLAccessor(_target), _reference, rel_tolerance_f16, 0.f, abs_tolerance_f16); + } + else + { + ARM_COMPUTE_TEST_INFO("cl_khr_image2d_from_buffer not supported. TEST skipped"); + framework::ARM_COMPUTE_PRINT_INFO(); + } + +} + +FIXTURE_DATA_TEST_CASE(RunLarge, CLGEMMMatrixMultiplyReshapedFixture<half>, framework::DatasetMode::NIGHTLY, + combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine( + m_values, + n_values), + k_values), + b_values), + m0_values_nightly), + n0_export_to_cl_image_values_nightly), + k0_export_to_cl_image_values_nightly), + v0_values_nightly), + h0_values_nightly), + i_values_lhs), + i_values_rhs), + framework::dataset::make("export_to_cl_image_rhs", true)), + framework::dataset::make("DataType", DataType::F16)), + a_values_nightly), + beta_values_nightly), + broadcast_bias_values), + lhs_transpose_values), + act_values)) +{ + // Validate output only if validate() is successful + if(validate_result) + { + validate(CLAccessor(_target), _reference, rel_tolerance_f16, 0.f, abs_tolerance_f16); + } + else + { + ARM_COMPUTE_TEST_INFO("cl_khr_image2d_from_buffer not supported. TEST skipped"); + framework::ARM_COMPUTE_PRINT_INFO(); + } } + +FIXTURE_DATA_TEST_CASE(RunSmall3D, CLGEMMMatrixMultiplyReshaped3DFixture<half>, framework::DatasetMode::ALL, + combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine( + m_w_values, + m_h_values), + n_values), + k_values), + b_values), + m0_values_precommit), + n0_values_precommit), + k0_values_precommit), + v0_values_precommit), + h0_values_precommit), + i_values_lhs), + i_values_rhs), + framework::dataset::make("export_to_cl_image_rhs", true)), + framework::dataset::make("DataType", DataType::F16)), + a_values_precommit), + beta_values_precommit), + lhs_transpose_values), + act_values)) +{ + // Validate output only if validate() is successful + if(validate_result) + { + validate(CLAccessor(_target), _reference, rel_tolerance_f16, 0.f, abs_tolerance_f16); + } + else + { + ARM_COMPUTE_TEST_INFO("cl_khr_image2d_from_buffer not supported. TEST skipped"); + framework::ARM_COMPUTE_PRINT_INFO(); + } +} + +FIXTURE_DATA_TEST_CASE(RunLarge3D, CLGEMMMatrixMultiplyReshaped3DFixture<half>, framework::DatasetMode::NIGHTLY, + combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine( + m_w_values, + m_h_values), + n_values), + k_values), + b_values), + m0_values_nightly), + n0_export_to_cl_image_values_nightly), + k0_export_to_cl_image_values_nightly), + v0_values_nightly), + h0_values_nightly), + i_values_lhs), + i_values_rhs), + framework::dataset::make("export_to_cl_image_rhs", true)), + framework::dataset::make("DataType", DataType::F16)), + a_values_nightly), + beta_values_nightly), + lhs_transpose_values), + act_values)) +{ + // Validate output only if validate() is successful + if(validate_result) + { + validate(CLAccessor(_target), _reference, rel_tolerance_f16, 0.f, abs_tolerance_f16); + } + else + { + ARM_COMPUTE_TEST_INFO("cl_khr_image2d_from_buffer not supported. TEST skipped"); + framework::ARM_COMPUTE_PRINT_INFO(); + } +} + +TEST_SUITE_END() // ExportToCLImage TEST_SUITE_END() // FP16 TEST_SUITE(MixedPrecision) FIXTURE_DATA_TEST_CASE(RunSmall, CLGEMMMatrixMultiplyReshapedMixedPrecisionFixture<half>, framework::DatasetMode::ALL, - combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine( + combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine( m_values, n_values), k_values), @@ -537,6 +1162,7 @@ FIXTURE_DATA_TEST_CASE(RunSmall, CLGEMMMatrixMultiplyReshapedMixedPrecisionFixtu h0_values_precommit), i_values_lhs), i_values_rhs), + framework::dataset::make("export_to_cl_image_rhs", false)), framework::dataset::make("DataType", DataType::F16)), a_values_precommit), beta_values_precommit), @@ -545,11 +1171,19 @@ FIXTURE_DATA_TEST_CASE(RunSmall, CLGEMMMatrixMultiplyReshapedMixedPrecisionFixtu act_values)) { // Validate output - validate(CLAccessor(_target), _reference, rel_tolerance_f16_mixed_precision, 0.f, abs_tolerance_f16_mixed_precision); + if(validate_result) + { + validate(CLAccessor(_target), _reference, rel_tolerance_f16_mixed_precision, 0.f, abs_tolerance_f16_mixed_precision); + } + else + { + ARM_COMPUTE_TEST_INFO("cl_khr_image2d_from_buffer not supported. TEST skipped"); + framework::ARM_COMPUTE_PRINT_INFO(); + } } FIXTURE_DATA_TEST_CASE(RunLarge, CLGEMMMatrixMultiplyReshapedMixedPrecisionFixture<half>, framework::DatasetMode::DISABLED, - combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine( + combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine( m_values, n_values), k_values), @@ -561,6 +1195,7 @@ FIXTURE_DATA_TEST_CASE(RunLarge, CLGEMMMatrixMultiplyReshapedMixedPrecisionFixtu h0_values_nightly), i_values_lhs), i_values_rhs), + framework::dataset::make("export_to_cl_image_rhs", false)), framework::dataset::make("DataType", DataType::F16)), a_values_nightly), beta_values_nightly), @@ -569,11 +1204,19 @@ FIXTURE_DATA_TEST_CASE(RunLarge, CLGEMMMatrixMultiplyReshapedMixedPrecisionFixtu act_values)) { // Validate output - validate(CLAccessor(_target), _reference, rel_tolerance_f16_mixed_precision, 0.f, abs_tolerance_f16_mixed_precision); + if(validate_result) + { + validate(CLAccessor(_target), _reference, rel_tolerance_f16_mixed_precision, 0.f, abs_tolerance_f16_mixed_precision); + } + else + { + ARM_COMPUTE_TEST_INFO("cl_khr_image2d_from_buffer not supported. TEST skipped"); + framework::ARM_COMPUTE_PRINT_INFO(); + } } FIXTURE_DATA_TEST_CASE(RunSmall3D, CLGEMMMatrixMultiplyReshaped3DMixedPrecisionFixture<half>, framework::DatasetMode::ALL, - combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine( + combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine( m_w_values, m_h_values), n_values), @@ -586,6 +1229,7 @@ FIXTURE_DATA_TEST_CASE(RunSmall3D, CLGEMMMatrixMultiplyReshaped3DMixedPrecisionF h0_values_precommit), i_values_lhs), i_values_rhs), + framework::dataset::make("export_to_cl_image_rhs", false)), framework::dataset::make("DataType", DataType::F16)), a_values_precommit), beta_values_precommit), @@ -593,11 +1237,19 @@ FIXTURE_DATA_TEST_CASE(RunSmall3D, CLGEMMMatrixMultiplyReshaped3DMixedPrecisionF act_values)) { // Validate output - validate(CLAccessor(_target), _reference, rel_tolerance_f16_mixed_precision, 0.f, abs_tolerance_f16_mixed_precision); + if(validate_result) + { + validate(CLAccessor(_target), _reference, rel_tolerance_f16_mixed_precision, 0.f, abs_tolerance_f16_mixed_precision); + } + else + { + ARM_COMPUTE_TEST_INFO("cl_khr_image2d_from_buffer not supported. TEST skipped"); + framework::ARM_COMPUTE_PRINT_INFO(); + } } FIXTURE_DATA_TEST_CASE(RunLarge3D, CLGEMMMatrixMultiplyReshaped3DMixedPrecisionFixture<half>, framework::DatasetMode::DISABLED, - combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine( + combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine( m_w_values, m_h_values), n_values), @@ -610,6 +1262,7 @@ FIXTURE_DATA_TEST_CASE(RunLarge3D, CLGEMMMatrixMultiplyReshaped3DMixedPrecisionF h0_values_nightly), i_values_lhs), i_values_rhs), + framework::dataset::make("export_to_cl_image_rhs", false)), framework::dataset::make("DataType", DataType::F16)), a_values_nightly), beta_values_nightly), @@ -617,8 +1270,17 @@ FIXTURE_DATA_TEST_CASE(RunLarge3D, CLGEMMMatrixMultiplyReshaped3DMixedPrecisionF act_values)) { // Validate output - validate(CLAccessor(_target), _reference, rel_tolerance_f16_mixed_precision, 0.f, abs_tolerance_f16_mixed_precision); + if(validate_result) + { + validate(CLAccessor(_target), _reference, rel_tolerance_f16_mixed_precision, 0.f, abs_tolerance_f16_mixed_precision); + } + else + { + ARM_COMPUTE_TEST_INFO("cl_khr_image2d_from_buffer not supported. TEST skipped"); + framework::ARM_COMPUTE_PRINT_INFO(); + } } + TEST_SUITE_END() // MixedPrecision TEST_SUITE_END() // Float TEST_SUITE_END() // GEMMMatrixMultiplyReshaped |