├── .clang-format
├── .clang-ignorelist
├── .clang-tidy
├── .gitattributes
├── .github
├── CODEOWNERS
├── ISSUE_TEMPLATE
│ ├── bug_report.md
│ ├── documentation.md
│ ├── feature_request.md
│ └── question.md
├── automation
│ ├── aarch64
│ │ ├── build.sh
│ │ ├── build_acl.sh
│ │ ├── ci.json
│ │ ├── common.sh
│ │ ├── skipped-tests.sh
│ │ └── test.sh
│ ├── clang-format.sh
│ ├── commit-msg-check.py
│ ├── performance
│ │ ├── bench_nightly_performance.sh
│ │ ├── bench_pr_performance.sh
│ │ ├── benchdnn_comparison.py
│ │ └── inputs
│ │ │ ├── conv
│ │ │ ├── conv_nightly
│ │ │ ├── eltwise
│ │ │ ├── eltwise_nightly
│ │ │ ├── matmul
│ │ │ ├── matmul_nightly
│ │ │ ├── reorder
│ │ │ └── reorder_nightly
│ └── x64
│ │ └── build_linters.sh
├── azure
│ ├── build.bat
│ ├── build.sh
│ ├── ci-x64.yml
│ ├── env
│ │ └── clang.sh
│ ├── test.bat
│ └── test.sh
├── codeql-config.yml
├── dependabot.yml
├── labels.yml
├── pull_request_template.md
└── workflows
│ ├── aarch64-acl.yml
│ ├── ci-aarch64.yml
│ ├── clang-tidy.yml
│ ├── codeql.yml
│ ├── labeler.yml
│ ├── nightly-aarch64.yml
│ ├── openssf-scorecard.yml
│ ├── performance-aarch64.yml
│ └── pr-linter.yml
├── .gitignore
├── CITATION.cff
├── CMakeLists.txt
├── CODE_OF_CONDUCT.md
├── CODING_STANDARDS.md
├── CONTRIBUTING.md
├── LICENSE
├── MAINTAINERS.md
├── README.binary.in
├── README.md
├── SECURITY.md
├── THIRD-PARTY-PROGRAMS
├── cmake
├── ACL.cmake
├── Doxygen.cmake
├── Doxyrest.cmake
├── FindACL.cmake
├── FindBLAS.cmake
├── FindDoxyrest.cmake
├── FindHIP.cmake
├── FindMIOpen.cmake
├── FindOpenCL.cmake
├── FindPI_CUDA.cmake
├── FindSphinx.cmake
├── FindcuBLAS.cmake
├── FindcuDNN.cmake
├── FindcublasLt.cmake
├── FindrocBLAS.cmake
├── OpenCL.cmake
├── OpenMP.cmake
├── SDL.cmake
├── SYCL.cmake
├── Sphinx.cmake
├── TBB.cmake
├── Threading.cmake
├── Threadpool.cmake
├── blas.cmake
├── build_types.cmake
├── config.cmake.in
├── configuring_primitive_list.cmake
├── coverage.cmake
├── dnnl_compat.cmake
├── doc.cmake
├── dpcpp_driver_check.cmake
├── gen_gpu_kernel.cmake
├── gen_gpu_kernel_list.cmake
├── host_compiler.cmake
├── host_compiler_id.cmake
├── host_compiler_id.cpp
├── lnx
│ └── TBBConfig.cmake
├── mac
│ └── TBBConfig.cmake
├── options.cmake
├── platform.cmake
├── run_with_env.bat.in
├── template.vcxproj.user
├── testing.cmake
├── utils.cmake
├── version.cmake
├── version.rc.in
└── win
│ └── TBBConfig.cmake
├── doc
├── Doxyfile.in
├── advanced
│ ├── design
│ │ ├── mem_fmt_blk.png
│ │ ├── mem_fmt_img1.png
│ │ ├── mem_fmt_img2.png
│ │ ├── mem_fmt_padded_blk.png
│ │ └── strides.png
│ ├── dpcpp_interoperability.md
│ ├── experimental.md
│ ├── int8_computations.md
│ ├── opencl_interoperability.md
│ ├── persistent_cache.md
│ ├── primitive_cache.md
│ ├── sparsity.md
│ ├── threadpool.md
│ ├── transition-to-dnnl.md
│ └── understanding_memory_formats.md
├── build
│ ├── build.md
│ ├── build_options.md
│ ├── link.md
│ └── system_requirements.md
├── doxyrest
│ ├── doxyrest-config.lua
│ └── frame
│ │ ├── cfamily
│ │ ├── class.rst.in
│ │ ├── compound.rst.in
│ │ ├── crefdb.py.in
│ │ ├── crefdb_enums.py.in
│ │ ├── crefdb_items.py.in
│ │ ├── crefdb_members.py.in
│ │ ├── details_aliases.rst.in
│ │ ├── details_construction.rst.in
│ │ ├── details_defines.rst.in
│ │ ├── details_events.rst.in
│ │ ├── details_functions.rst.in
│ │ ├── details_properties.rst.in
│ │ ├── details_typedefs.rst.in
│ │ ├── details_unnamed_enum_values.rst.in
│ │ ├── details_variables.rst.in
│ │ ├── enum.rst.in
│ │ ├── example.rst.in
│ │ ├── global.rst.in
│ │ ├── index.rst.in
│ │ ├── namespace.rst.in
│ │ ├── overview_aliases.rst.in
│ │ ├── overview_classes.rst.in
│ │ ├── overview_common.rst.in
│ │ ├── overview_compound.rst.in
│ │ ├── overview_construction.rst.in
│ │ ├── overview_defines.rst.in
│ │ ├── overview_enums.rst.in
│ │ ├── overview_events.rst.in
│ │ ├── overview_functions.rst.in
│ │ ├── overview_members.rst.in
│ │ ├── overview_namespaces.rst.in
│ │ ├── overview_properties.rst.in
│ │ ├── overview_typedefs.rst.in
│ │ ├── overview_variables.rst.in
│ │ ├── scope_class.rst.in
│ │ ├── scope_namespace.rst.in
│ │ └── utils.lua
│ │ └── common
│ │ ├── crefdb_external.py.in
│ │ ├── crefdb_groups.py.in
│ │ ├── crefdb_items.py.in
│ │ ├── details.rst.in
│ │ ├── doc.lua
│ │ ├── footnotes.rst.in
│ │ ├── group.rst.in
│ │ ├── index_main.rst.in
│ │ ├── item.lua
│ │ ├── page.rst.in
│ │ ├── string.lua
│ │ ├── table.lua
│ │ └── toc.lua
├── environment.yml
├── examples.md
├── graph
│ ├── constant_tensor_cache.md
│ ├── fusion_patterns
│ │ ├── binary_fusion_patterns.md
│ │ ├── convolution_fusion_patterns.md
│ │ ├── convtranspose_fusions_patterns.md
│ │ ├── gated_mlp.md
│ │ ├── gqa.md
│ │ ├── images
│ │ │ ├── binary_pattern.png
│ │ │ ├── compressed_sdpa_pattern.png
│ │ │ ├── conv_bwd_pattern.png
│ │ │ ├── conv_pattern.png
│ │ │ ├── convtranspose_pattern.png
│ │ │ ├── epilogue_subgraph_conv.png
│ │ │ ├── epilogue_subgraph_general_1.png
│ │ │ ├── epilogue_subgraph_general_2.png
│ │ │ ├── epilogue_subgraph_matmul.png
│ │ │ ├── f2f_conversion.png
│ │ │ ├── f2q_conversion_general.png
│ │ │ ├── f2q_conversion_quantized_conv.png
│ │ │ ├── f2q_conversion_quantized_matmul.png
│ │ │ ├── f2q_conversion_softmax.png
│ │ │ ├── fp-gated-mlp.png
│ │ │ ├── gated-mlp-swish.png
│ │ │ ├── gqa.png
│ │ │ ├── interpolate_pattern.png
│ │ │ ├── matmul_pattern.png
│ │ │ ├── norm_pattern.png
│ │ │ ├── pool_pattern.png
│ │ │ ├── q2f_conversion_quantized_conv_matmul.png
│ │ │ ├── q2f_conversion_quantized_convtranspose.png
│ │ │ ├── quantized_conv_pattern.png
│ │ │ ├── quantized_convtranspose_pattern.png
│ │ │ ├── quantized_matmul_pattern.png
│ │ │ ├── reduction_pattern.png
│ │ │ ├── sdpa-mask-1.png
│ │ │ ├── sdpa-mask-2.png
│ │ │ ├── sdpa-mask-3.png
│ │ │ ├── sdpa-mask-4.png
│ │ │ ├── sdpa-reorder.png
│ │ │ ├── sdpa.png
│ │ │ ├── softmax_pattern.png
│ │ │ └── unary_pattern.png
│ │ ├── interpolate_fusion_patterns.md
│ │ ├── matmul_fusion_patterns.md
│ │ ├── norm_fusion_patterns.md
│ │ ├── pool_fusion_patterns.md
│ │ ├── quantized_convolution_fusion_patterns.md
│ │ ├── quantized_convtranspose_fusion_patterns.md
│ │ ├── quantized_matmul_fusion_patterns.md
│ │ ├── reduction_fusion_patterns.md
│ │ ├── sdpa.md
│ │ ├── sdpa_with_compressed_kv.md
│ │ ├── softmax_fusion_patterns.md
│ │ └── unary_fusion_patterns.md
│ ├── graph_dump.md
│ ├── operations
│ │ ├── Abs.md
│ │ ├── AbsBackward.md
│ │ ├── Add.md
│ │ ├── AvgPool.md
│ │ ├── AvgPoolBackward.md
│ │ ├── BatchNormForwardTraining.md
│ │ ├── BatchNormInference.md
│ │ ├── BatchNormTrainingBackward.md
│ │ ├── BiasAdd.md
│ │ ├── BiasAddBackward.md
│ │ ├── Clamp.md
│ │ ├── ClampBackward.md
│ │ ├── Concat.md
│ │ ├── ConvTranspose.md
│ │ ├── ConvTransposeBackwardData.md
│ │ ├── ConvTransposeBackwardWeights.md
│ │ ├── Convolution.md
│ │ ├── ConvolutionBackwardData.md
│ │ ├── ConvolutionBackwardWeights.md
│ │ ├── Dequantize.md
│ │ ├── Divide.md
│ │ ├── DynamicDequantize.md
│ │ ├── DynamicQuantize.md
│ │ ├── Elu.md
│ │ ├── EluBackward.md
│ │ ├── End.md
│ │ ├── Exp.md
│ │ ├── GELU.md
│ │ ├── GELUBackward.md
│ │ ├── GenIndex.md
│ │ ├── GreaterEqual.md
│ │ ├── GroupNorm.md
│ │ ├── HardSigmoid.md
│ │ ├── HardSigmoidBackward.md
│ │ ├── HardSwish.md
│ │ ├── HardSwishBackward.md
│ │ ├── Interpolate.md
│ │ ├── InterpolateBackward.md
│ │ ├── LayerNorm.md
│ │ ├── LayerNormBackward.md
│ │ ├── LeakyReLU.md
│ │ ├── Log.md
│ │ ├── LogSoftmax.md
│ │ ├── LogSoftmaxBackward.md
│ │ ├── MatMul.md
│ │ ├── MaxPool.md
│ │ ├── MaxPoolBackward.md
│ │ ├── Maximum.md
│ │ ├── Minimum.md
│ │ ├── Mish.md
│ │ ├── MishBackward.md
│ │ ├── Multiply.md
│ │ ├── PReLU.md
│ │ ├── PReLUBackward.md
│ │ ├── Pow.md
│ │ ├── Quantize.md
│ │ ├── ReLU.md
│ │ ├── ReLUBackward.md
│ │ ├── Reciprocal.md
│ │ ├── ReduceL1.md
│ │ ├── ReduceL2.md
│ │ ├── ReduceMax.md
│ │ ├── ReduceMean.md
│ │ ├── ReduceMin.md
│ │ ├── ReduceProd.md
│ │ ├── ReduceSum.md
│ │ ├── Reorder.md
│ │ ├── Round.md
│ │ ├── Select.md
│ │ ├── Sigmoid.md
│ │ ├── SigmoidBackward.md
│ │ ├── SoftPlus.md
│ │ ├── SoftPlusBackward.md
│ │ ├── Softmax.md
│ │ ├── SoftmaxBackward.md
│ │ ├── Sqrt.md
│ │ ├── SqrtBackward.md
│ │ ├── Square.md
│ │ ├── SquaredDifference.md
│ │ ├── StaticReshape.md
│ │ ├── StaticTranspose.md
│ │ ├── Subtract.md
│ │ ├── Tanh.md
│ │ ├── TanhBackward.md
│ │ ├── TypeCast.md
│ │ └── Wildcard.md
│ ├── programming_model
│ │ ├── graph_basic_concepts.md
│ │ ├── images
│ │ │ ├── bf16_programming.jpg
│ │ │ ├── img_graph_programming_model.png
│ │ │ └── int8_programming.jpg
│ │ └── low_precision.md
│ └── rst
│ │ ├── graph_fusion_patterns.rst
│ │ ├── graph_programming_model.rst
│ │ ├── graph_supported_operations.rst
│ │ └── images
│ │ └── other_pattern.png
├── legal_information.md
├── mainpage.md
├── naming_conventions.md
├── performance_considerations
│ ├── benchdnn.md
│ ├── cpu_isa_hints.md
│ ├── dispatcher_control.md
│ ├── inspecting_jit.md
│ ├── perf_settings.md
│ ├── profilers.md
│ ├── verbose.md
│ ├── verbose_table.md
│ └── vtune.md
├── primitives
│ ├── batch_normalization.md
│ ├── binary.md
│ ├── concat.md
│ ├── convolution.md
│ ├── eltwise.md
│ ├── group_normalization.md
│ ├── images
│ │ └── unrolled_stack_rnn.jpg
│ ├── inner_product.md
│ ├── layer_normalization.md
│ ├── lrn.md
│ ├── matmul.md
│ ├── pooling.md
│ ├── prelu.md
│ ├── reduction.md
│ ├── reorder.md
│ ├── resampling.md
│ ├── rnn.md
│ ├── shuffle.md
│ ├── softmax.md
│ └── sum.md
├── programming_model
│ ├── api.md
│ ├── attributes.md
│ ├── attributes_accumulation_mode.md
│ ├── attributes_deterministic.md
│ ├── attributes_dropout.md
│ ├── attributes_fpmath_mode.md
│ ├── attributes_post_ops.md
│ ├── attributes_quantization.md
│ ├── attributes_rounding_mode.md
│ ├── attributes_scratchpad.md
│ ├── basic_concepts.md
│ ├── data_types.md
│ ├── images
│ │ ├── img_depthwise_fusion.jpg
│ │ ├── img_dnnl_object_snapshot.jpg
│ │ ├── img_dnnl_programming_flow.jpg
│ │ ├── img_overview_flow.jpg
│ │ └── img_programming_model.png
│ └── inference_and_training_aspects.md
├── rst
│ ├── advanced_topics.rst
│ ├── build_and_link.rst
│ ├── dev_guide_examples.rst
│ ├── graph_extension.rst
│ ├── index.rst
│ ├── interop_with_dpcpp_and_opencl.rst
│ ├── orphans.rst
│ ├── performance_profiling_and_inspection.rst
│ ├── programming_model.rst
│ ├── supported_primitives.rst
│ └── ukernels.rst
├── sphinx
│ ├── _static
│ │ ├── dnnl.js
│ │ ├── doxyrest_code_copy_button.js
│ │ ├── favicons.png
│ │ └── oneAPI-rgb-rev-100.png
│ ├── cleanup.py
│ └── conf.py
├── ukernel
│ ├── operations
│ │ ├── brgemm.md
│ │ └── transform.md
│ └── programming_model
│ │ └── ukernel_basic_concepts.md
└── usage_models
│ ├── images
│ ├── img_bf16_diagram.png
│ ├── img_diagram.png
│ ├── img_inference_scope.jpg
│ ├── img_multiscalar.png
│ ├── img_singlescalar.png
│ └── img_training_inference_scope.jpg
│ ├── inference.md
│ ├── inference_int8.md
│ ├── training.md
│ └── training_bf16.md
├── examples
├── CMakeLists.txt
├── CMakeLists.txt.in
├── bnorm_u8_via_binary_postops.cpp
├── cnn_inference_f32.c
├── cnn_inference_f32.cpp
├── cnn_inference_int8.cpp
├── cnn_training_bf16.cpp
├── cnn_training_f32.cpp
├── cpu_cnn_training_f32.c
├── cpu_matmul_coo.cpp
├── cpu_matmul_csr.cpp
├── cpu_matmul_weights_compression.cpp
├── cpu_rnn_inference_f32.cpp
├── cpu_rnn_inference_int8.cpp
├── cross_engine_reorder.c
├── cross_engine_reorder.cpp
├── example_utils.h
├── example_utils.hpp
├── getting_started.cpp
├── gpu_opencl_interop.cpp
├── graph
│ ├── cpu_getting_started.cpp
│ ├── cpu_inference_int8.cpp
│ ├── cpu_single_op_partition.cpp
│ ├── gated_mlp.cpp
│ ├── gated_mlp_int4.cpp
│ ├── gated_mlp_wei_combined.cpp
│ ├── gpu_opencl_getting_started.cpp
│ ├── gqa.cpp
│ ├── graph_example_utils.hpp
│ ├── mqa.cpp
│ ├── sdpa.cpp
│ ├── sdpa_stacked_qkv.cpp
│ ├── sycl_getting_started.cpp
│ └── sycl_single_op_partition.cpp
├── matmul_perf.cpp
├── memory_format_propagation.cpp
├── performance_profiling.cpp
├── primitives
│ ├── augru.cpp
│ ├── batch_normalization.cpp
│ ├── binary.cpp
│ ├── concat.cpp
│ ├── convolution.cpp
│ ├── deconvolution.cpp
│ ├── eltwise.cpp
│ ├── group_normalization.cpp
│ ├── inner_product.cpp
│ ├── layer_normalization.cpp
│ ├── lbr_gru.cpp
│ ├── lrn.cpp
│ ├── lstm.cpp
│ ├── matmul.cpp
│ ├── pooling.cpp
│ ├── prelu.cpp
│ ├── reduction.cpp
│ ├── reorder.cpp
│ ├── resampling.cpp
│ ├── shuffle.cpp
│ ├── softmax.cpp
│ ├── sum.cpp
│ └── vanilla_rnn.cpp
├── rnn_training_f32.cpp
├── sycl_interop_buffer.cpp
├── sycl_interop_usm.cpp
├── tutorials
│ └── matmul
│ │ ├── cpu_matmul_quantization.cpp
│ │ ├── cpu_sgemm_and_matmul.cpp
│ │ ├── inference_int8_matmul.cpp
│ │ └── weights_decompression_matmul.cpp
└── ukernels
│ └── cpu_brgemm.cpp
├── include
├── dnnl.h
├── dnnl.hpp
├── dnnl_config.h
├── dnnl_debug.h
├── dnnl_ocl.h
├── dnnl_ocl.hpp
├── dnnl_sycl.h
├── dnnl_sycl.hpp
├── dnnl_sycl_types.h
├── dnnl_threadpool.h
├── dnnl_threadpool.hpp
├── dnnl_threadpool_iface.hpp
├── dnnl_types.h
├── dnnl_version.h
└── oneapi
│ └── dnnl
│ ├── dnnl.h
│ ├── dnnl.hpp
│ ├── dnnl_common.h
│ ├── dnnl_common.hpp
│ ├── dnnl_common_types.h
│ ├── dnnl_config.h.in
│ ├── dnnl_debug.h
│ ├── dnnl_graph.h
│ ├── dnnl_graph.hpp
│ ├── dnnl_graph_ocl.h
│ ├── dnnl_graph_ocl.hpp
│ ├── dnnl_graph_sycl.h
│ ├── dnnl_graph_sycl.hpp
│ ├── dnnl_graph_types.h
│ ├── dnnl_ocl.h
│ ├── dnnl_ocl.hpp
│ ├── dnnl_ocl_types.h
│ ├── dnnl_sycl.h
│ ├── dnnl_sycl.hpp
│ ├── dnnl_sycl_types.h
│ ├── dnnl_threadpool.h
│ ├── dnnl_threadpool.hpp
│ ├── dnnl_threadpool_iface.hpp
│ ├── dnnl_types.h
│ ├── dnnl_ukernel.h
│ ├── dnnl_ukernel.hpp
│ ├── dnnl_ukernel_types.h
│ ├── dnnl_version.h.in
│ └── dnnl_version_hash.h.in
├── pyproject.toml
├── scripts
├── README.md
├── fix_header_guards.py
├── generate_dnnl_debug.py
├── generate_format_tags.py
├── synthdnn
│ ├── README.md
│ ├── matmul
│ │ ├── primitive.py
│ │ └── sampler.py
│ └── synthdnn.py
└── verbose_converter
│ ├── README.md
│ ├── src
│ ├── __init__.py
│ ├── benchdnn_generator.py
│ ├── breakdown_generator.py
│ ├── dnnl_parser.py
│ ├── ir.py
│ ├── parse.py
│ └── utils.py
│ ├── tests
│ ├── README.md
│ ├── benchdnn_test.py
│ ├── dataset_ci
│ └── dataset_simple
│ └── verbose_converter.py
├── src
├── CMakeLists.txt
├── common
│ ├── CMakeLists.txt
│ ├── batch_normalization.cpp
│ ├── batch_normalization_pd.hpp
│ ├── bfloat16.cpp
│ ├── bfloat16.hpp
│ ├── binary.cpp
│ ├── binary_pd.hpp
│ ├── bit_cast.hpp
│ ├── broadcast_strategy.cpp
│ ├── broadcast_strategy.hpp
│ ├── c_types_map.hpp
│ ├── cache_blob.hpp
│ ├── cache_blob_id.cpp
│ ├── cache_blob_id.hpp
│ ├── cache_hit_types.hpp
│ ├── cache_utils.hpp
│ ├── compiler_workarounds.hpp
│ ├── concat.cpp
│ ├── concat.hpp
│ ├── concat_pd.hpp
│ ├── convolution.cpp
│ ├── convolution_pd.cpp
│ ├── convolution_pd.hpp
│ ├── counting_barrier.hpp
│ ├── cpp_compat.hpp
│ ├── deconvolution.cpp
│ ├── deconvolution_pd.hpp
│ ├── dnnl_debug.cpp
│ ├── dnnl_debug_autogenerated.cpp
│ ├── dnnl_thread.hpp
│ ├── dnnl_thread_tbb_proxy.hpp
│ ├── dnnl_threadpool.cpp
│ ├── dnnl_traits.hpp
│ ├── eltwise.cpp
│ ├── eltwise_pd.hpp
│ ├── engine.cpp
│ ├── engine.hpp
│ ├── engine_id.hpp
│ ├── engine_impl.hpp
│ ├── experimental.cpp
│ ├── experimental.hpp
│ ├── float16.hpp
│ ├── float4.cpp
│ ├── float4.hpp
│ ├── float8.cpp
│ ├── float8.hpp
│ ├── fpmath_mode.cpp
│ ├── gemm.cpp
│ ├── gemm_pd.hpp
│ ├── gemm_types.hpp
│ ├── gemm_utils.hpp
│ ├── group_normalization.cpp
│ ├── group_normalization_pd.hpp
│ ├── impl_list_item.hpp
│ ├── impl_registration.hpp
│ ├── inner_product.cpp
│ ├── inner_product_pd.hpp
│ ├── int4.hpp
│ ├── internal_defs.hpp
│ ├── ittnotify.cpp
│ ├── ittnotify.hpp
│ ├── kernel_cache.cpp
│ ├── kernel_cache.hpp
│ ├── layer_normalization.cpp
│ ├── layer_normalization_pd.hpp
│ ├── logging.cpp
│ ├── logging.hpp
│ ├── lrn.cpp
│ ├── lrn_pd.hpp
│ ├── math_utils.hpp
│ ├── matmul.cpp
│ ├── matmul_pd.hpp
│ ├── memory.cpp
│ ├── memory.hpp
│ ├── memory_debug.cpp
│ ├── memory_debug.hpp
│ ├── memory_desc.cpp
│ ├── memory_desc.hpp
│ ├── memory_desc_wrapper.cpp
│ ├── memory_desc_wrapper.hpp
│ ├── memory_map_manager.hpp
│ ├── memory_storage.cpp
│ ├── memory_storage.hpp
│ ├── memory_tracking.cpp
│ ├── memory_tracking.hpp
│ ├── memory_zero_pad.cpp
│ ├── nstl.hpp
│ ├── opdesc.hpp
│ ├── optional.hpp
│ ├── pooling.cpp
│ ├── pooling_pd.hpp
│ ├── prelu.cpp
│ ├── prelu_pd.hpp
│ ├── primitive.cpp
│ ├── primitive.hpp
│ ├── primitive_attr.cpp
│ ├── primitive_attr.hpp
│ ├── primitive_attr_quant.cpp
│ ├── primitive_attr_quant.hpp
│ ├── primitive_cache.cpp
│ ├── primitive_cache.hpp
│ ├── primitive_desc.hpp
│ ├── primitive_desc_iface.cpp
│ ├── primitive_desc_iface.hpp
│ ├── primitive_desc_iterator.hpp
│ ├── primitive_exec_types.cpp
│ ├── primitive_exec_types.hpp
│ ├── primitive_hashing.cpp
│ ├── primitive_hashing.hpp
│ ├── primitive_iface.cpp
│ ├── primitive_iface.hpp
│ ├── primitive_serialization.cpp
│ ├── primitive_serialization.hpp
│ ├── profiler.hpp
│ ├── query.cpp
│ ├── reduction.cpp
│ ├── reduction_pd.hpp
│ ├── reorder.cpp
│ ├── reorder.hpp
│ ├── reorder_pd.hpp
│ ├── resampling.cpp
│ ├── resampling_pd.hpp
│ ├── resource.hpp
│ ├── rnn.cpp
│ ├── rnn.hpp
│ ├── rnn_pd.hpp
│ ├── rw_mutex.cpp
│ ├── rw_mutex.hpp
│ ├── scratchpad.cpp
│ ├── scratchpad.hpp
│ ├── scratchpad_debug.cpp
│ ├── scratchpad_debug.hpp
│ ├── sdpa_pd.hpp
│ ├── sdpa_test_iface.cpp
│ ├── sdpa_types.hpp
│ ├── sdpa_utils.hpp
│ ├── serialization.hpp
│ ├── shuffle.cpp
│ ├── shuffle_pd.hpp
│ ├── softmax.cpp
│ ├── softmax_pd.hpp
│ ├── stack_checker.hpp
│ ├── stream.cpp
│ ├── stream.hpp
│ ├── stream_impl.hpp
│ ├── stream_profiler.cpp
│ ├── stream_threadpool.cpp
│ ├── sum.cpp
│ ├── sum_pd.hpp
│ ├── tag_traits.hpp
│ ├── thread_local_storage.hpp
│ ├── type_helpers.hpp
│ ├── utils.cpp
│ ├── utils.hpp
│ ├── verbose.cpp
│ ├── verbose.hpp
│ ├── verbose_msg.hpp
│ └── z_magic.hpp
├── cpu
│ ├── CMakeLists.txt
│ ├── README.md
│ ├── aarch64
│ │ ├── CMakeLists.txt
│ │ ├── acl_batch_normalization.cpp
│ │ ├── acl_batch_normalization.hpp
│ │ ├── acl_benchmark_scheduler.cpp
│ │ ├── acl_benchmark_scheduler.hpp
│ │ ├── acl_binary.cpp
│ │ ├── acl_binary.hpp
│ │ ├── acl_convolution_utils.cpp
│ │ ├── acl_convolution_utils.hpp
│ │ ├── acl_deconvolution.cpp
│ │ ├── acl_deconvolution.hpp
│ │ ├── acl_depthwise_convolution.cpp
│ │ ├── acl_depthwise_convolution.hpp
│ │ ├── acl_eltwise.cpp
│ │ ├── acl_eltwise.hpp
│ │ ├── acl_gemm_convolution.cpp
│ │ ├── acl_gemm_convolution.hpp
│ │ ├── acl_indirect_gemm_convolution.cpp
│ │ ├── acl_indirect_gemm_convolution.hpp
│ │ ├── acl_inner_product.cpp
│ │ ├── acl_inner_product.hpp
│ │ ├── acl_layer_normalization.cpp
│ │ ├── acl_layer_normalization.hpp
│ │ ├── acl_pooling.cpp
│ │ ├── acl_pooling.hpp
│ │ ├── acl_post_ops.cpp
│ │ ├── acl_post_ops.hpp
│ │ ├── acl_prelu.cpp
│ │ ├── acl_prelu.hpp
│ │ ├── acl_reorder.cpp
│ │ ├── acl_reorder.hpp
│ │ ├── acl_softmax.cpp
│ │ ├── acl_softmax.hpp
│ │ ├── acl_thread.cpp
│ │ ├── acl_thread.hpp
│ │ ├── acl_threadpool_scheduler.cpp
│ │ ├── acl_threadpool_scheduler.hpp
│ │ ├── acl_utils.cpp
│ │ ├── acl_utils.hpp
│ │ ├── acl_winograd_convolution.cpp
│ │ ├── acl_winograd_convolution.hpp
│ │ ├── brgemm
│ │ │ ├── brgemm.cpp
│ │ │ ├── brgemm.hpp
│ │ │ ├── brgemm_containers.cpp
│ │ │ ├── brgemm_containers.hpp
│ │ │ ├── brgemm_types.hpp
│ │ │ ├── brgemm_utils.cpp
│ │ │ ├── brgemm_utils.hpp
│ │ │ ├── jit_brdgmm_kernel.cpp
│ │ │ ├── jit_brdgmm_kernel.hpp
│ │ │ └── jit_brgemm_kernel.cpp
│ │ ├── cpu_barrier.cpp
│ │ ├── cpu_barrier.hpp
│ │ ├── cpu_isa_traits.cpp
│ │ ├── cpu_isa_traits.hpp
│ │ ├── cpu_reducer.cpp
│ │ ├── cpu_reducer.hpp
│ │ ├── injectors
│ │ │ ├── injector_utils.cpp
│ │ │ ├── injector_utils.hpp
│ │ │ ├── jit_uni_binary_injector.cpp
│ │ │ ├── jit_uni_binary_injector.hpp
│ │ │ ├── jit_uni_eltwise_injector.cpp
│ │ │ ├── jit_uni_eltwise_injector.hpp
│ │ │ ├── jit_uni_postops_injector.cpp
│ │ │ └── jit_uni_postops_injector.hpp
│ │ ├── jit_brdgmm_dw_conv.cpp
│ │ ├── jit_brdgmm_dw_conv.hpp
│ │ ├── jit_brgemm_1x1_conv.cpp
│ │ ├── jit_brgemm_1x1_conv.hpp
│ │ ├── jit_brgemm_conv.cpp
│ │ ├── jit_brgemm_conv.hpp
│ │ ├── jit_brgemm_conv_bwd.cpp
│ │ ├── jit_brgemm_conv_bwd.hpp
│ │ ├── jit_brgemm_conv_comp_pad_kernel.cpp
│ │ ├── jit_brgemm_conv_comp_pad_kernel.hpp
│ │ ├── jit_brgemm_conv_trans_kernel.cpp
│ │ ├── jit_brgemm_conv_trans_kernel.hpp
│ │ ├── jit_brgemm_conv_utils.cpp
│ │ ├── jit_brgemm_conv_utils.hpp
│ │ ├── jit_brgemm_post_ops.hpp
│ │ ├── jit_brgemm_primitive_conf.hpp
│ │ ├── jit_brgemm_transpose_utils.cpp
│ │ ├── jit_brgemm_transpose_utils.hpp
│ │ ├── jit_generator.hpp
│ │ ├── jit_op_imm_check.hpp
│ │ ├── jit_primitive_conf.hpp
│ │ ├── jit_sve_1x1_conv_kernel.cpp
│ │ ├── jit_sve_1x1_conv_kernel.hpp
│ │ ├── jit_sve_1x1_convolution.cpp
│ │ ├── jit_sve_1x1_convolution.hpp
│ │ ├── jit_sve_512_core_x8s8s32x_deconvolution.cpp
│ │ ├── jit_sve_512_core_x8s8s32x_deconvolution.hpp
│ │ ├── jit_sve_512_x8s8s32x_conv_kernel.cpp
│ │ ├── jit_sve_512_x8s8s32x_conv_kernel.hpp
│ │ ├── jit_sve_512_x8s8s32x_convolution.cpp
│ │ ├── jit_sve_512_x8s8s32x_convolution.hpp
│ │ ├── jit_sve_conv_kernel.cpp
│ │ ├── jit_sve_conv_kernel.hpp
│ │ ├── jit_sve_convolution.cpp
│ │ ├── jit_sve_convolution.hpp
│ │ ├── jit_uni_1x1_conv_utils.hpp
│ │ ├── jit_uni_batch_normalization.cpp
│ │ ├── jit_uni_batch_normalization.hpp
│ │ ├── jit_uni_batch_normalization_s8.cpp
│ │ ├── jit_uni_batch_normalization_s8.hpp
│ │ ├── jit_uni_binary.cpp
│ │ ├── jit_uni_binary.hpp
│ │ ├── jit_uni_binary_kernel.cpp
│ │ ├── jit_uni_binary_kernel.hpp
│ │ ├── jit_uni_deconv_zp_pad_str_kernel.cpp
│ │ ├── jit_uni_deconv_zp_pad_str_kernel.hpp
│ │ ├── jit_uni_dw_conv_kernel_f32.cpp
│ │ ├── jit_uni_dw_conv_kernel_f32.hpp
│ │ ├── jit_uni_dw_conv_kernel_utils.hpp
│ │ ├── jit_uni_dw_convolution.cpp
│ │ ├── jit_uni_dw_convolution.hpp
│ │ ├── jit_uni_eltwise.cpp
│ │ ├── jit_uni_eltwise.hpp
│ │ ├── jit_uni_eltwise_int.cpp
│ │ ├── jit_uni_eltwise_int.hpp
│ │ ├── jit_uni_i8i8_pooling.cpp
│ │ ├── jit_uni_i8i8_pooling.hpp
│ │ ├── jit_uni_pool_kernel.cpp
│ │ ├── jit_uni_pool_kernel.hpp
│ │ ├── jit_uni_pooling.cpp
│ │ ├── jit_uni_pooling.hpp
│ │ ├── jit_uni_reorder.cpp
│ │ ├── jit_uni_reorder.hpp
│ │ ├── jit_uni_reorder_utils.cpp
│ │ ├── jit_uni_softmax.cpp
│ │ ├── jit_uni_softmax.hpp
│ │ ├── matmul
│ │ │ ├── acl_lowp_matmul.cpp
│ │ │ ├── acl_lowp_matmul.hpp
│ │ │ ├── acl_lowp_matmul_sq.cpp
│ │ │ ├── acl_lowp_matmul_sq.hpp
│ │ │ ├── acl_matmul.cpp
│ │ │ ├── acl_matmul.hpp
│ │ │ ├── acl_matmul_utils.cpp
│ │ │ ├── acl_matmul_utils.hpp
│ │ │ ├── brgemm_matmul.cpp
│ │ │ ├── brgemm_matmul.hpp
│ │ │ ├── brgemm_matmul_copy_utils.cpp
│ │ │ ├── brgemm_matmul_copy_utils.hpp
│ │ │ ├── brgemm_matmul_reorders.cpp
│ │ │ ├── brgemm_matmul_reorders.hpp
│ │ │ ├── brgemm_matmul_utils.cpp
│ │ │ ├── brgemm_matmul_utils.hpp
│ │ │ ├── jit_int8_kernel_types.hpp
│ │ │ ├── jit_int8_matmul.cpp
│ │ │ ├── jit_int8_matmul.hpp
│ │ │ ├── jit_int8_matmul_utils.cpp
│ │ │ └── jit_int8_matmul_utils.hpp
│ │ ├── shuffle
│ │ │ ├── jit_uni_shuffle.cpp
│ │ │ ├── jit_uni_shuffle.hpp
│ │ │ ├── jit_uni_shuffle_kernel.cpp
│ │ │ └── jit_uni_shuffle_kernel.hpp
│ │ └── utils
│ │ │ ├── jit_io_helper.cpp
│ │ │ └── jit_io_helper.hpp
│ ├── bfloat16.cpp
│ ├── binary_injector_utils.cpp
│ ├── binary_injector_utils.hpp
│ ├── cpu_batch_normalization_list.cpp
│ ├── cpu_batch_normalization_pd.hpp
│ ├── cpu_batch_normalization_utils.cpp
│ ├── cpu_batch_normalization_utils.hpp
│ ├── cpu_binary_list.cpp
│ ├── cpu_binary_pd.hpp
│ ├── cpu_concat.cpp
│ ├── cpu_concat_pd.hpp
│ ├── cpu_convolution_list.cpp
│ ├── cpu_convolution_pd.hpp
│ ├── cpu_deconvolution_list.cpp
│ ├── cpu_deconvolution_pd.hpp
│ ├── cpu_eltwise_list.cpp
│ ├── cpu_eltwise_pd.hpp
│ ├── cpu_engine.cpp
│ ├── cpu_engine.hpp
│ ├── cpu_group_normalization_list.cpp
│ ├── cpu_group_normalization_pd.hpp
│ ├── cpu_inner_product_list.cpp
│ ├── cpu_inner_product_pd.hpp
│ ├── cpu_layer_normalization_list.cpp
│ ├── cpu_layer_normalization_pd.hpp
│ ├── cpu_lrn_list.cpp
│ ├── cpu_lrn_pd.hpp
│ ├── cpu_memory_storage.hpp
│ ├── cpu_pooling_list.cpp
│ ├── cpu_pooling_pd.hpp
│ ├── cpu_prelu_list.cpp
│ ├── cpu_prelu_pd.hpp
│ ├── cpu_primitive.hpp
│ ├── cpu_reduction_list.cpp
│ ├── cpu_reduction_pd.hpp
│ ├── cpu_resampling_list.cpp
│ ├── cpu_resampling_pd.hpp
│ ├── cpu_rnn_list.cpp
│ ├── cpu_shuffle_list.cpp
│ ├── cpu_shuffle_pd.hpp
│ ├── cpu_softmax_list.cpp
│ ├── cpu_softmax_pd.hpp
│ ├── cpu_stream.hpp
│ ├── cpu_sum.cpp
│ ├── cpu_sum_pd.hpp
│ ├── dw_convolution_utils.hpp
│ ├── float16.cpp
│ ├── gemm
│ │ ├── bf16
│ │ │ ├── ref_gemm_bf16.cpp
│ │ │ └── ref_gemm_bf16.hpp
│ │ ├── f32
│ │ │ ├── gemm_utils_f32.cpp
│ │ │ ├── gemm_utils_f32.hpp
│ │ │ ├── ref_gemm_f32.cpp
│ │ │ └── ref_gemm_f32.hpp
│ │ ├── gemm.cpp
│ │ ├── gemm.hpp
│ │ ├── gemm_msan_unpoison.hpp
│ │ ├── gemm_pack.cpp
│ │ ├── gemm_pack.hpp
│ │ ├── os_blas.hpp
│ │ └── s8x8s32
│ │ │ ├── ref_gemm_s8x8s32.cpp
│ │ │ ├── ref_gemm_s8x8s32.hpp
│ │ │ ├── simple_gemm_s8s8s32.cpp
│ │ │ └── simple_gemm_s8s8s32.hpp
│ ├── gemm_convolution.cpp
│ ├── gemm_convolution.hpp
│ ├── gemm_convolution_utils.cpp
│ ├── gemm_convolution_utils.hpp
│ ├── gemm_inner_product.cpp
│ ├── gemm_inner_product.hpp
│ ├── gemm_inner_product_utils.cpp
│ ├── gemm_inner_product_utils.hpp
│ ├── gemm_x8s8s32x_conv_zp_src_pad_comp.cpp
│ ├── gemm_x8s8s32x_conv_zp_src_pad_comp.hpp
│ ├── gemm_x8s8s32x_convolution.cpp
│ ├── gemm_x8s8s32x_convolution.hpp
│ ├── gemm_x8s8s32x_convolution_utils.cpp
│ ├── gemm_x8s8s32x_convolution_utils.hpp
│ ├── gemm_x8s8s32x_inner_product.cpp
│ ├── gemm_x8s8s32x_inner_product.hpp
│ ├── jit_utils
│ │ ├── jit_utils.cpp
│ │ ├── jit_utils.hpp
│ │ └── linux_perf
│ │ │ ├── README.md
│ │ │ ├── linux_perf.cpp
│ │ │ └── linux_perf.hpp
│ ├── matmul
│ │ ├── cpu_matmul_list.cpp
│ │ ├── cpu_matmul_pd.hpp
│ │ ├── gemm_based_common.hpp
│ │ ├── gemm_bf16_matmul.cpp
│ │ ├── gemm_bf16_matmul.hpp
│ │ ├── gemm_f32_matmul.cpp
│ │ ├── gemm_f32_matmul.hpp
│ │ ├── gemm_x8s8s32x_matmul.cpp
│ │ ├── gemm_x8s8s32x_matmul.hpp
│ │ ├── matmul_utils.hpp
│ │ ├── ref_matmul.cpp
│ │ ├── ref_matmul.hpp
│ │ ├── ref_matmul_int8.cpp
│ │ ├── ref_matmul_int8.hpp
│ │ ├── ref_sparse_matmul.cpp
│ │ └── ref_sparse_matmul.hpp
│ ├── nchw_pooling.cpp
│ ├── nchw_pooling.hpp
│ ├── ncsp_batch_normalization.cpp
│ ├── ncsp_batch_normalization.hpp
│ ├── ncsp_group_normalization.cpp
│ ├── ncsp_group_normalization.hpp
│ ├── nhwc_pooling.cpp
│ ├── nhwc_pooling.hpp
│ ├── nspc_batch_normalization.cpp
│ ├── nspc_batch_normalization.hpp
│ ├── platform.cpp
│ ├── platform.hpp
│ ├── ppc64
│ │ ├── CMakeLists.txt
│ │ ├── gemm
│ │ │ ├── gemm_driver.cpp
│ │ │ ├── gemm_driver.hpp
│ │ │ ├── gemm_info.cpp
│ │ │ ├── gemm_info.hpp
│ │ │ ├── gemm_pack_storage.hpp
│ │ │ ├── gemm_partition.hpp
│ │ │ ├── gemm_threading.hpp
│ │ │ └── gemm_utils.hpp
│ │ ├── ppc64_gemm_driver.hpp
│ │ ├── ppc64_gemm_reorder.cpp
│ │ ├── ppc64_gemm_reorder.hpp
│ │ ├── ppc64_gemm_s8x8s32.cpp
│ │ └── ppc64_gemm_s8x8s32.hpp
│ ├── primitive_attr_postops.cpp
│ ├── primitive_attr_postops.hpp
│ ├── ref_batch_normalization.cpp
│ ├── ref_batch_normalization.hpp
│ ├── ref_binary.cpp
│ ├── ref_binary.hpp
│ ├── ref_concat.hpp
│ ├── ref_convolution.cpp
│ ├── ref_convolution.hpp
│ ├── ref_convolution_int8.cpp
│ ├── ref_convolution_int8.hpp
│ ├── ref_convolution_utils.hpp
│ ├── ref_deconvolution.cpp
│ ├── ref_deconvolution.hpp
│ ├── ref_eltwise.cpp
│ ├── ref_eltwise.hpp
│ ├── ref_fused_convolution.hpp
│ ├── ref_group_normalization.cpp
│ ├── ref_group_normalization.hpp
│ ├── ref_inner_product.cpp
│ ├── ref_inner_product.hpp
│ ├── ref_inner_product_int8.cpp
│ ├── ref_inner_product_int8.hpp
│ ├── ref_inner_product_utils.hpp
│ ├── ref_io_helper.hpp
│ ├── ref_layer_normalization.cpp
│ ├── ref_layer_normalization.hpp
│ ├── ref_lrn.cpp
│ ├── ref_lrn.hpp
│ ├── ref_pooling.cpp
│ ├── ref_pooling.hpp
│ ├── ref_prelu.cpp
│ ├── ref_prelu.hpp
│ ├── ref_reduction.cpp
│ ├── ref_reduction.hpp
│ ├── ref_resampling.cpp
│ ├── ref_resampling.hpp
│ ├── ref_shuffle.cpp
│ ├── ref_shuffle.hpp
│ ├── ref_softmax.cpp
│ ├── ref_softmax.hpp
│ ├── ref_sum.hpp
│ ├── reorder
│ │ ├── cpu_reorder.cpp
│ │ ├── cpu_reorder.hpp
│ │ ├── cpu_reorder_comp_bf16_s8.cpp
│ │ ├── cpu_reorder_comp_f32_s8.cpp
│ │ ├── cpu_reorder_comp_s8_s8.cpp
│ │ ├── cpu_reorder_pd.hpp
│ │ ├── cpu_reorder_regular_bf16.cpp
│ │ ├── cpu_reorder_regular_f16.cpp
│ │ ├── cpu_reorder_regular_f32_bf16.cpp
│ │ ├── cpu_reorder_regular_f32_f16.cpp
│ │ ├── cpu_reorder_regular_f32_f32.cpp
│ │ ├── cpu_reorder_regular_f32_fp8.cpp
│ │ ├── cpu_reorder_regular_f32_s32.cpp
│ │ ├── cpu_reorder_regular_f32_s8.cpp
│ │ ├── cpu_reorder_regular_f32_u8.cpp
│ │ ├── cpu_reorder_regular_fp4.cpp
│ │ ├── cpu_reorder_regular_fp8.cpp
│ │ ├── cpu_reorder_regular_s32.cpp
│ │ ├── cpu_reorder_regular_s4.cpp
│ │ ├── cpu_reorder_regular_s8.cpp
│ │ ├── cpu_reorder_regular_u4.cpp
│ │ ├── cpu_reorder_regular_u8.cpp
│ │ ├── simple_reorder.hpp
│ │ └── simple_sparse_reorder.hpp
│ ├── resampling_utils.hpp
│ ├── rnn
│ │ ├── brgemm_cell_common.cpp
│ │ ├── cell_common.cpp
│ │ ├── cell_gru.cpp
│ │ ├── cell_gru_lbr.cpp
│ │ ├── cpu_rnn_pd.hpp
│ │ ├── postgemm_dispatcher.hpp
│ │ ├── ref_postgemm_gru.cpp
│ │ ├── ref_postgemm_gru_lbr.cpp
│ │ ├── ref_postgemm_lstm.cpp
│ │ ├── ref_postgemm_lstm_projection.cpp
│ │ ├── ref_postgemm_rnn.cpp
│ │ ├── ref_rnn.cpp
│ │ ├── ref_rnn.hpp
│ │ ├── rnn_reorders.hpp
│ │ ├── rnn_utils.cpp
│ │ └── rnn_utils.hpp
│ ├── rv64
│ │ ├── CMakeLists.txt
│ │ ├── rvv_nchw_pooling.cpp
│ │ └── rvv_nchw_pooling.hpp
│ ├── s390x
│ │ ├── CMakeLists.txt
│ │ ├── gemm.h
│ │ ├── gemmu16.cpp
│ │ ├── helpers.h
│ │ ├── kernel_s16s16s32.hpp
│ │ └── pack.hpp
│ ├── scale_utils.cpp
│ ├── scale_utils.hpp
│ ├── simple_concat.cpp
│ ├── simple_concat.hpp
│ ├── simple_layer_normalization.cpp
│ ├── simple_layer_normalization.hpp
│ ├── simple_q10n.hpp
│ ├── simple_resampling.cpp
│ ├── simple_resampling.hpp
│ ├── simple_sum.cpp
│ ├── simple_sum.hpp
│ ├── sycl
│ │ ├── CMakeLists.txt
│ │ ├── engine.cpp
│ │ ├── engine.hpp
│ │ ├── stream.cpp
│ │ ├── stream.hpp
│ │ ├── stream_cpu_thunk.cpp
│ │ ├── stream_cpu_thunk.hpp
│ │ ├── stream_submit_cpu_primitive.cpp
│ │ └── stream_submit_cpu_primitive.hpp
│ ├── ukernel
│ │ ├── attr_params.cpp
│ │ ├── brgemm.cpp
│ │ ├── c_types_map.hpp
│ │ └── transform.cpp
│ ├── x64
│ │ ├── CMakeLists.txt
│ │ ├── amx_tile_configure.cpp
│ │ ├── amx_tile_configure.hpp
│ │ ├── brgemm
│ │ │ ├── brgemm.cpp
│ │ │ ├── brgemm.hpp
│ │ │ ├── brgemm_containers.cpp
│ │ │ ├── brgemm_containers.hpp
│ │ │ ├── brgemm_types.hpp
│ │ │ ├── brgemm_utils.cpp
│ │ │ ├── brgemm_utils.hpp
│ │ │ ├── jit_brdgmm_kernel.cpp
│ │ │ ├── jit_brdgmm_kernel.hpp
│ │ │ ├── jit_brgemm_amx_uker.cpp
│ │ │ └── jit_brgemm_kernel.cpp
│ │ ├── cpu_barrier.cpp
│ │ ├── cpu_barrier.hpp
│ │ ├── cpu_isa_traits.cpp
│ │ ├── cpu_isa_traits.hpp
│ │ ├── cpu_reducer.cpp
│ │ ├── cpu_reducer.hpp
│ │ ├── gemm
│ │ │ ├── amx
│ │ │ │ ├── jit_avx512_core_amx_copy_kern.cpp
│ │ │ │ ├── jit_avx512_core_amx_copy_kern.hpp
│ │ │ │ ├── jit_avx512_core_amx_gemm_kern.cpp
│ │ │ │ └── jit_avx512_core_amx_gemm_kern.hpp
│ │ │ ├── bf16
│ │ │ │ ├── common_s16.hpp
│ │ │ │ ├── jit_avx512_core_gemm_bf16bf16f32_kern.cpp
│ │ │ │ ├── jit_avx512_core_gemm_bf16bf16f32_kern.hpp
│ │ │ │ ├── jit_avx512_core_gemv_bf16bf16f32_kern.cpp
│ │ │ │ ├── jit_avx512_core_gemv_bf16bf16f32_kern.hpp
│ │ │ │ ├── jit_avx512_core_s16_24x8_copy_an_kern_autogen.cpp
│ │ │ │ ├── jit_avx512_core_s16_24x8_copy_at_kern_autogen.cpp
│ │ │ │ ├── jit_avx512_core_s16_24x8_copy_bn_kern_autogen.cpp
│ │ │ │ ├── jit_avx512_core_s16_24x8_copy_bt_kern_autogen.cpp
│ │ │ │ ├── jit_avx512_core_s16_48x8_copy_an_kern_autogen.cpp
│ │ │ │ ├── jit_avx512_core_s16_48x8_copy_at_kern_autogen.cpp
│ │ │ │ ├── jit_avx512_core_s16_48x8_copy_bn_kern_autogen.cpp
│ │ │ │ └── jit_avx512_core_s16_48x8_copy_bt_kern_autogen.cpp
│ │ │ ├── f32
│ │ │ │ ├── common_f32.hpp
│ │ │ │ ├── jit_avx2_f32_copy_an_kern_autogen.cpp
│ │ │ │ ├── jit_avx2_f32_copy_at_kern_autogen.cpp
│ │ │ │ ├── jit_avx2_f32_copy_bn_kern_autogen.cpp
│ │ │ │ ├── jit_avx2_f32_copy_bt_kern_autogen.cpp
│ │ │ │ ├── jit_avx2_kernel_sgemm_kern.cpp
│ │ │ │ ├── jit_avx2_kernel_sgemm_kern.hpp
│ │ │ │ ├── jit_avx512_common_gemm_f32.cpp
│ │ │ │ ├── jit_avx512_common_gemm_f32.hpp
│ │ │ │ ├── jit_avx512_core_f32_copy_an_kern_autogen.cpp
│ │ │ │ ├── jit_avx512_core_f32_copy_at_kern_autogen.hpp
│ │ │ │ ├── jit_avx512_core_f32_copy_at_kern_part1_autogen.cpp
│ │ │ │ ├── jit_avx512_core_f32_copy_at_kern_part2_autogen.cpp
│ │ │ │ ├── jit_avx512_core_f32_copy_bn_kern_autogen.cpp
│ │ │ │ ├── jit_avx512_core_f32_copy_bt_kern_autogen.cpp
│ │ │ │ ├── jit_avx512_core_gemm_smalln_tn_f32_kern.cpp
│ │ │ │ ├── jit_avx512_core_gemm_smalln_tn_f32_kern.hpp
│ │ │ │ ├── jit_avx_f32_copy_an_kern_autogen.cpp
│ │ │ │ ├── jit_avx_f32_copy_at_kern_autogen.cpp
│ │ │ │ ├── jit_avx_f32_copy_bn_kern_autogen.cpp
│ │ │ │ ├── jit_avx_f32_copy_bt_kern_autogen.cpp
│ │ │ │ ├── jit_avx_gemm_f32.cpp
│ │ │ │ ├── jit_avx_gemm_f32.hpp
│ │ │ │ ├── jit_avx_gemv_t_f32_kern.cpp
│ │ │ │ ├── jit_avx_gemv_t_f32_kern.hpp
│ │ │ │ ├── jit_avx_kernel_b0_sgemm_kern_autogen.hpp
│ │ │ │ ├── jit_avx_kernel_b0_sgemm_kern_part1_autogen.cpp
│ │ │ │ ├── jit_avx_kernel_b0_sgemm_kern_part2_autogen.cpp
│ │ │ │ ├── jit_avx_kernel_sgemm_kern_autogen.hpp
│ │ │ │ ├── jit_avx_kernel_sgemm_kern_part1_autogen.cpp
│ │ │ │ ├── jit_avx_kernel_sgemm_kern_part2_autogen.cpp
│ │ │ │ ├── jit_sse41_f32_copy_an_kern_autogen.cpp
│ │ │ │ ├── jit_sse41_f32_copy_at_kern_autogen.cpp
│ │ │ │ ├── jit_sse41_f32_copy_bn_kern_autogen.cpp
│ │ │ │ ├── jit_sse41_f32_copy_bt_kern_autogen.cpp
│ │ │ │ ├── jit_sse41_gemv_n_f32_kern.cpp
│ │ │ │ ├── jit_sse41_gemv_n_f32_kern.hpp
│ │ │ │ ├── jit_sse41_gemv_t_f32_kern.cpp
│ │ │ │ ├── jit_sse41_gemv_t_f32_kern.hpp
│ │ │ │ ├── jit_sse41_kernel_b0_sgemm_kern_autogen.cpp
│ │ │ │ └── jit_sse41_kernel_sgemm_kern_autogen.cpp
│ │ │ ├── gemm_driver.cpp
│ │ │ ├── gemm_driver.hpp
│ │ │ ├── gemm_info.cpp
│ │ │ ├── gemm_info.hpp
│ │ │ ├── gemm_pack.cpp
│ │ │ ├── gemm_pack.hpp
│ │ │ ├── gemm_pack_storage.hpp
│ │ │ ├── gemm_partition.hpp
│ │ │ ├── gemm_threading.hpp
│ │ │ ├── gemm_utils.hpp
│ │ │ ├── gemv_driver.cpp
│ │ │ ├── gemv_driver.hpp
│ │ │ └── s8x8s32
│ │ │ │ ├── common_u8.hpp
│ │ │ │ ├── jit_avx2_gemm_s8u8s32_kern.cpp
│ │ │ │ ├── jit_avx2_gemm_s8u8s32_kern.hpp
│ │ │ │ ├── jit_avx2_u8_copy_an_kern_autogen.cpp
│ │ │ │ ├── jit_avx2_u8_copy_at_kern_autogen.cpp
│ │ │ │ ├── jit_avx2_u8_copy_bn_kern_autogen.cpp
│ │ │ │ ├── jit_avx2_u8_copy_bt_kern_autogen.cpp
│ │ │ │ ├── jit_avx2_u8_copy_sum_an_kern_autogen.cpp
│ │ │ │ ├── jit_avx2_u8_copy_sum_at_kern_autogen.cpp
│ │ │ │ ├── jit_avx2_u8_copy_sum_bn_kern_autogen.cpp
│ │ │ │ ├── jit_avx2_u8_copy_sum_bt_kern_autogen.cpp
│ │ │ │ ├── jit_avx2_vnni_u8_copy_an_kern_autogen.cpp
│ │ │ │ ├── jit_avx2_vnni_u8_copy_at_kern_autogen.cpp
│ │ │ │ ├── jit_avx2_vnni_u8_copy_bn_kern_autogen.cpp
│ │ │ │ ├── jit_avx2_vnni_u8_copy_bt_kern_autogen.cpp
│ │ │ │ ├── jit_avx2_vnni_u8_copy_sum_an_kern_autogen.cpp
│ │ │ │ ├── jit_avx2_vnni_u8_copy_sum_at_kern_autogen.cpp
│ │ │ │ ├── jit_avx2_vnni_u8_copy_sum_bn_kern_autogen.cpp
│ │ │ │ ├── jit_avx2_vnni_u8_copy_sum_bt_kern_autogen.cpp
│ │ │ │ ├── jit_avx512_core_gemm_s8u8s32_kern.cpp
│ │ │ │ ├── jit_avx512_core_gemm_s8u8s32_kern.hpp
│ │ │ │ ├── jit_avx512_core_gemv_s8x8s32.cpp
│ │ │ │ ├── jit_avx512_core_gemv_s8x8s32.hpp
│ │ │ │ ├── jit_avx512_core_kernel_gemv_s8x8s32_kern.cpp
│ │ │ │ ├── jit_avx512_core_kernel_gemv_s8x8s32_kern.hpp
│ │ │ │ ├── jit_avx512_core_u8_copy_an_kern_autogen.cpp
│ │ │ │ ├── jit_avx512_core_u8_copy_at_kern_autogen.cpp
│ │ │ │ ├── jit_avx512_core_u8_copy_bn_kern_autogen.cpp
│ │ │ │ ├── jit_avx512_core_u8_copy_bt_kern_autogen.cpp
│ │ │ │ ├── jit_avx512_core_u8_copy_sum_an_kern_autogen.cpp
│ │ │ │ ├── jit_avx512_core_u8_copy_sum_at_kern_autogen.cpp
│ │ │ │ ├── jit_avx512_core_u8_copy_sum_bn_kern_autogen.cpp
│ │ │ │ ├── jit_avx512_core_u8_copy_sum_bt_kern_autogen.cpp
│ │ │ │ ├── jit_avx_kernel_b0_b_gemm_s8u8s32_kern_autogen.cpp
│ │ │ │ ├── jit_avx_kernel_b0_c_gemm_s8u8s32_kern_autogen.cpp
│ │ │ │ ├── jit_avx_kernel_b0_gemm_s8u8s32_kern_autogen.cpp
│ │ │ │ ├── jit_avx_kernel_b0_r_gemm_s8u8s32_kern_autogen.cpp
│ │ │ │ ├── jit_avx_kernel_b_gemm_s8u8s32_kern_autogen.cpp
│ │ │ │ ├── jit_avx_kernel_c_gemm_s8u8s32_kern_autogen.cpp
│ │ │ │ ├── jit_avx_kernel_gemm_s8u8s32_kern_autogen.cpp
│ │ │ │ ├── jit_avx_kernel_r_gemm_s8u8s32_kern_autogen.cpp
│ │ │ │ ├── jit_avx_u8_copy_an_kern_autogen.cpp
│ │ │ │ ├── jit_avx_u8_copy_at_kern_autogen.cpp
│ │ │ │ ├── jit_avx_u8_copy_bn_kern_autogen.cpp
│ │ │ │ ├── jit_avx_u8_copy_bt_kern_autogen.cpp
│ │ │ │ ├── jit_avx_u8_copy_sum_an_kern_autogen.cpp
│ │ │ │ ├── jit_avx_u8_copy_sum_at_kern_autogen.cpp
│ │ │ │ ├── jit_avx_u8_copy_sum_bn_kern_autogen.cpp
│ │ │ │ ├── jit_avx_u8_copy_sum_bt_kern_autogen.cpp
│ │ │ │ ├── jit_sse41_kernel_b0_b_gemm_s8u8s32_kern_autogen.cpp
│ │ │ │ ├── jit_sse41_kernel_b0_c_gemm_s8u8s32_kern_autogen.cpp
│ │ │ │ ├── jit_sse41_kernel_b0_gemm_s8u8s32_kern_autogen.cpp
│ │ │ │ ├── jit_sse41_kernel_b0_r_gemm_s8u8s32_kern_autogen.cpp
│ │ │ │ ├── jit_sse41_kernel_b_gemm_s8u8s32_kern_autogen.cpp
│ │ │ │ ├── jit_sse41_kernel_c_gemm_s8u8s32_kern_autogen.cpp
│ │ │ │ ├── jit_sse41_kernel_gemm_s8u8s32_kern_autogen.cpp
│ │ │ │ ├── jit_sse41_kernel_r_gemm_s8u8s32_kern_autogen.cpp
│ │ │ │ ├── jit_sse41_u8_copy_an_kern_autogen.cpp
│ │ │ │ ├── jit_sse41_u8_copy_at_kern_autogen.cpp
│ │ │ │ ├── jit_sse41_u8_copy_bn_kern_autogen.cpp
│ │ │ │ ├── jit_sse41_u8_copy_bt_kern_autogen.cpp
│ │ │ │ ├── jit_sse41_u8_copy_sum_an_kern_autogen.cpp
│ │ │ │ ├── jit_sse41_u8_copy_sum_at_kern_autogen.cpp
│ │ │ │ ├── jit_sse41_u8_copy_sum_bn_kern_autogen.cpp
│ │ │ │ └── jit_sse41_u8_copy_sum_bt_kern_autogen.cpp
│ │ ├── gemm_bf16_convolution.cpp
│ │ ├── gemm_bf16_convolution.hpp
│ │ ├── gemm_bf16_inner_product.cpp
│ │ ├── gemm_bf16_inner_product.hpp
│ │ ├── injectors
│ │ │ ├── injector_utils.cpp
│ │ │ ├── injector_utils.hpp
│ │ │ ├── jit_uni_binary_injector.cpp
│ │ │ ├── jit_uni_binary_injector.hpp
│ │ │ ├── jit_uni_eltwise_injector.cpp
│ │ │ ├── jit_uni_eltwise_injector.hpp
│ │ │ ├── jit_uni_postops_injector.cpp
│ │ │ └── jit_uni_postops_injector.hpp
│ │ ├── ip_convolution.cpp
│ │ ├── ip_convolution.hpp
│ │ ├── jit_avx2_1x1_conv_kernel_f32.cpp
│ │ ├── jit_avx2_1x1_conv_kernel_f32.hpp
│ │ ├── jit_avx2_1x1_convolution.cpp
│ │ ├── jit_avx2_1x1_convolution.hpp
│ │ ├── jit_avx2_conv_kernel_f32.cpp
│ │ ├── jit_avx2_conv_kernel_f32.hpp
│ │ ├── jit_avx2_convolution.cpp
│ │ ├── jit_avx2_convolution.hpp
│ │ ├── jit_avx512_common_1x1_conv_kernel.cpp
│ │ ├── jit_avx512_common_1x1_conv_kernel.hpp
│ │ ├── jit_avx512_common_1x1_convolution.cpp
│ │ ├── jit_avx512_common_1x1_convolution.hpp
│ │ ├── jit_avx512_common_conv_kernel.cpp
│ │ ├── jit_avx512_common_conv_kernel.hpp
│ │ ├── jit_avx512_common_convolution.cpp
│ │ ├── jit_avx512_common_convolution.hpp
│ │ ├── jit_avx512_core_amx_1x1_conv_kernel.cpp
│ │ ├── jit_avx512_core_amx_1x1_conv_kernel.hpp
│ │ ├── jit_avx512_core_amx_1x1_convolution.cpp
│ │ ├── jit_avx512_core_amx_1x1_convolution.hpp
│ │ ├── jit_avx512_core_amx_conv_kernel.cpp
│ │ ├── jit_avx512_core_amx_conv_kernel.hpp
│ │ ├── jit_avx512_core_amx_conv_utils.hpp
│ │ ├── jit_avx512_core_amx_convolution.cpp
│ │ ├── jit_avx512_core_amx_convolution.hpp
│ │ ├── jit_avx512_core_amx_deconvolution.cpp
│ │ ├── jit_avx512_core_amx_deconvolution.hpp
│ │ ├── jit_avx512_core_bf16_1x1_conv_kernel.cpp
│ │ ├── jit_avx512_core_bf16_1x1_conv_kernel.hpp
│ │ ├── jit_avx512_core_bf16_1x1_convolution.cpp
│ │ ├── jit_avx512_core_bf16_1x1_convolution.hpp
│ │ ├── jit_avx512_core_bf16_conv_kernel.cpp
│ │ ├── jit_avx512_core_bf16_conv_kernel.hpp
│ │ ├── jit_avx512_core_bf16_convolution.cpp
│ │ ├── jit_avx512_core_bf16_convolution.hpp
│ │ ├── jit_avx512_core_bf16_dw_conv_kernel.cpp
│ │ ├── jit_avx512_core_bf16_dw_conv_kernel.hpp
│ │ ├── jit_avx512_core_bf16cvt.hpp
│ │ ├── jit_avx512_core_f16_dw_conv_kernel.cpp
│ │ ├── jit_avx512_core_f16_dw_conv_kernel.hpp
│ │ ├── jit_avx512_core_fp16cvt.cpp
│ │ ├── jit_avx512_core_fp16cvt.hpp
│ │ ├── jit_avx512_core_fp8cvt.cpp
│ │ ├── jit_avx512_core_fp8cvt.hpp
│ │ ├── jit_avx512_core_resampling.cpp
│ │ ├── jit_avx512_core_resampling.hpp
│ │ ├── jit_avx512_core_scale_precompute.cpp
│ │ ├── jit_avx512_core_scale_precompute.hpp
│ │ ├── jit_avx512_core_x8s8s32x_1x1_conv_kernel.cpp
│ │ ├── jit_avx512_core_x8s8s32x_1x1_conv_kernel.hpp
│ │ ├── jit_avx512_core_x8s8s32x_1x1_convolution.cpp
│ │ ├── jit_avx512_core_x8s8s32x_1x1_convolution.hpp
│ │ ├── jit_avx512_core_x8s8s32x_1x1_deconvolution.hpp
│ │ ├── jit_avx512_core_x8s8s32x_conv_kernel.cpp
│ │ ├── jit_avx512_core_x8s8s32x_conv_kernel.hpp
│ │ ├── jit_avx512_core_x8s8s32x_convolution.cpp
│ │ ├── jit_avx512_core_x8s8s32x_convolution.hpp
│ │ ├── jit_avx512_core_x8s8s32x_deconvolution.cpp
│ │ ├── jit_avx512_core_x8s8s32x_deconvolution.hpp
│ │ ├── jit_avx512_sparse_decompress_kernel.cpp
│ │ ├── jit_avx512_sparse_decompress_kernel.hpp
│ │ ├── jit_brdgmm_dw_conv.cpp
│ │ ├── jit_brdgmm_dw_conv.hpp
│ │ ├── jit_brgemm_1x1_conv.cpp
│ │ ├── jit_brgemm_1x1_conv.hpp
│ │ ├── jit_brgemm_conv.cpp
│ │ ├── jit_brgemm_conv.hpp
│ │ ├── jit_brgemm_conv_bwd.cpp
│ │ ├── jit_brgemm_conv_bwd.hpp
│ │ ├── jit_brgemm_conv_bwd_copy_kernel.cpp
│ │ ├── jit_brgemm_conv_bwd_copy_kernel.hpp
│ │ ├── jit_brgemm_conv_bwd_strided.cpp
│ │ ├── jit_brgemm_conv_bwd_strided.hpp
│ │ ├── jit_brgemm_conv_bwd_trans_kernel.cpp
│ │ ├── jit_brgemm_conv_bwd_trans_kernel.hpp
│ │ ├── jit_brgemm_conv_bwd_utils.cpp
│ │ ├── jit_brgemm_conv_bwd_utils.hpp
│ │ ├── jit_brgemm_conv_bwd_w.cpp
│ │ ├── jit_brgemm_conv_bwd_w.hpp
│ │ ├── jit_brgemm_conv_comp_pad_kernel.cpp
│ │ ├── jit_brgemm_conv_comp_pad_kernel.hpp
│ │ ├── jit_brgemm_conv_trans_kernel.cpp
│ │ ├── jit_brgemm_conv_trans_kernel.hpp
│ │ ├── jit_brgemm_conv_utils.cpp
│ │ ├── jit_brgemm_conv_utils.hpp
│ │ ├── jit_brgemm_deconv.cpp
│ │ ├── jit_brgemm_deconv.hpp
│ │ ├── jit_brgemm_inner_product.cpp
│ │ ├── jit_brgemm_inner_product.hpp
│ │ ├── jit_brgemm_inner_product_utils.cpp
│ │ ├── jit_brgemm_inner_product_utils.hpp
│ │ ├── jit_brgemm_post_ops.cpp
│ │ ├── jit_brgemm_post_ops.hpp
│ │ ├── jit_brgemm_primitive_conf.cpp
│ │ ├── jit_brgemm_primitive_conf.hpp
│ │ ├── jit_brgemm_transpose_utils.cpp
│ │ ├── jit_brgemm_transpose_utils.hpp
│ │ ├── jit_gemm_inner_product_utils.cpp
│ │ ├── jit_gemm_inner_product_utils.hpp
│ │ ├── jit_gemm_x8s8s32x_conv_zp_src_pad_comp.cpp
│ │ ├── jit_gemm_x8s8s32x_conv_zp_src_pad_comp.hpp
│ │ ├── jit_gemm_x8s8s32x_convolution_utils.cpp
│ │ ├── jit_gemm_x8s8s32x_convolution_utils.hpp
│ │ ├── jit_generator.cpp
│ │ ├── jit_generator.hpp
│ │ ├── jit_primitive_conf.hpp
│ │ ├── jit_sse41_1x1_conv_kernel_f32.cpp
│ │ ├── jit_sse41_1x1_conv_kernel_f32.hpp
│ │ ├── jit_sse41_1x1_convolution.cpp
│ │ ├── jit_sse41_1x1_convolution.hpp
│ │ ├── jit_sse41_conv_kernel_f32.cpp
│ │ ├── jit_sse41_conv_kernel_f32.hpp
│ │ ├── jit_sse41_convolution.cpp
│ │ ├── jit_sse41_convolution.hpp
│ │ ├── jit_transpose_utils.cpp
│ │ ├── jit_transpose_utils.hpp
│ │ ├── jit_uni_1x1_conv_utils.hpp
│ │ ├── jit_uni_batch_normalization.cpp
│ │ ├── jit_uni_batch_normalization.hpp
│ │ ├── jit_uni_batch_normalization_s8.cpp
│ │ ├── jit_uni_batch_normalization_s8.hpp
│ │ ├── jit_uni_binary.cpp
│ │ ├── jit_uni_binary.hpp
│ │ ├── jit_uni_binary_kernel.cpp
│ │ ├── jit_uni_binary_kernel.hpp
│ │ ├── jit_uni_convert_xf16.cpp
│ │ ├── jit_uni_convert_xf16.hpp
│ │ ├── jit_uni_deconv_zp_pad_str_kernel.cpp
│ │ ├── jit_uni_deconv_zp_pad_str_kernel.hpp
│ │ ├── jit_uni_dw_conv_kernel_f32.cpp
│ │ ├── jit_uni_dw_conv_kernel_f32.hpp
│ │ ├── jit_uni_dw_conv_kernel_utils.cpp
│ │ ├── jit_uni_dw_conv_kernel_utils.hpp
│ │ ├── jit_uni_dw_convolution.cpp
│ │ ├── jit_uni_dw_convolution.hpp
│ │ ├── jit_uni_eltwise.cpp
│ │ ├── jit_uni_eltwise.hpp
│ │ ├── jit_uni_eltwise_int.cpp
│ │ ├── jit_uni_eltwise_int.hpp
│ │ ├── jit_uni_group_normalization.cpp
│ │ ├── jit_uni_group_normalization.hpp
│ │ ├── jit_uni_i8i8_pooling.cpp
│ │ ├── jit_uni_i8i8_pooling.hpp
│ │ ├── jit_uni_instance_normalization.cpp
│ │ ├── jit_uni_instance_normalization.hpp
│ │ ├── jit_uni_layer_normalization.cpp
│ │ ├── jit_uni_layer_normalization.hpp
│ │ ├── jit_uni_ncsp_convolution.cpp
│ │ ├── jit_uni_ncsp_convolution.hpp
│ │ ├── jit_uni_pool_kernel.cpp
│ │ ├── jit_uni_pool_kernel.hpp
│ │ ├── jit_uni_pooling.cpp
│ │ ├── jit_uni_pooling.hpp
│ │ ├── jit_uni_reduction.cpp
│ │ ├── jit_uni_reduction.hpp
│ │ ├── jit_uni_reduction_kernel.cpp
│ │ ├── jit_uni_reduction_kernel.hpp
│ │ ├── jit_uni_reorder.cpp
│ │ ├── jit_uni_reorder.hpp
│ │ ├── jit_uni_reorder_direct_copy.cpp
│ │ ├── jit_uni_reorder_direct_copy.hpp
│ │ ├── jit_uni_reorder_utils.cpp
│ │ ├── jit_uni_resampling.cpp
│ │ ├── jit_uni_resampling.hpp
│ │ ├── jit_uni_resampling_kernel.cpp
│ │ ├── jit_uni_resampling_kernel.hpp
│ │ ├── jit_uni_softmax.cpp
│ │ ├── jit_uni_softmax.hpp
│ │ ├── jit_uni_tbb_batch_normalization.cpp
│ │ ├── jit_uni_tbb_batch_normalization.hpp
│ │ ├── jit_uni_x8s8s32x_1x1_conv_kernel.cpp
│ │ ├── jit_uni_x8s8s32x_1x1_conv_kernel.hpp
│ │ ├── jit_uni_x8s8s32x_1x1_convolution.cpp
│ │ ├── jit_uni_x8s8s32x_1x1_convolution.hpp
│ │ ├── jit_uni_x8s8s32x_1x1_deconvolution.hpp
│ │ ├── jit_uni_x8s8s32x_conv_kernel.cpp
│ │ ├── jit_uni_x8s8s32x_conv_kernel.hpp
│ │ ├── jit_uni_x8s8s32x_convolution.cpp
│ │ ├── jit_uni_x8s8s32x_convolution.hpp
│ │ ├── jit_uni_x8s8s32x_deconvolution.cpp
│ │ ├── jit_uni_x8s8s32x_deconvolution.hpp
│ │ ├── jit_uni_xf16_sum.cpp
│ │ ├── jit_uni_xf16_sum.hpp
│ │ ├── lrn
│ │ │ ├── jit_avx512_common_lrn.cpp
│ │ │ ├── jit_avx512_common_lrn.hpp
│ │ │ ├── jit_avx512_common_lrn_bwd_base.cpp
│ │ │ ├── jit_avx512_common_lrn_bwd_base.hpp
│ │ │ ├── jit_avx512_common_lrn_bwd_blocked.cpp
│ │ │ ├── jit_avx512_common_lrn_bwd_blocked.hpp
│ │ │ ├── jit_avx512_common_lrn_bwd_nhwc.cpp
│ │ │ ├── jit_avx512_common_lrn_bwd_nhwc.hpp
│ │ │ ├── jit_avx512_common_lrn_fwd_base.cpp
│ │ │ ├── jit_avx512_common_lrn_fwd_base.hpp
│ │ │ ├── jit_avx512_common_lrn_fwd_blocked.cpp
│ │ │ ├── jit_avx512_common_lrn_fwd_blocked.hpp
│ │ │ ├── jit_avx512_common_lrn_fwd_nhwc.cpp
│ │ │ ├── jit_avx512_common_lrn_fwd_nhwc.hpp
│ │ │ ├── jit_avx512_common_lrn_utils.hpp
│ │ │ ├── jit_uni_lrn.cpp
│ │ │ ├── jit_uni_lrn.hpp
│ │ │ ├── jit_uni_lrn_kernel.cpp
│ │ │ ├── jit_uni_lrn_kernel.hpp
│ │ │ ├── lrn_avx512_blocked_executor.hpp
│ │ │ ├── lrn_avx512_nhwc_executor.hpp
│ │ │ ├── lrn_executor.hpp
│ │ │ └── lrn_executor_factory.hpp
│ │ ├── matmul
│ │ │ ├── amx_blocking_heuristics.cpp
│ │ │ ├── amx_blocking_heuristics.hpp
│ │ │ ├── brgemm_matmul.cpp
│ │ │ ├── brgemm_matmul.hpp
│ │ │ ├── brgemm_matmul_copy_utils.cpp
│ │ │ ├── brgemm_matmul_copy_utils.hpp
│ │ │ ├── brgemm_matmul_reorders.cpp
│ │ │ ├── brgemm_matmul_reorders.hpp
│ │ │ ├── brgemm_matmul_utils.cpp
│ │ │ ├── brgemm_matmul_utils.hpp
│ │ │ ├── jit_uni_sparse_matmul.cpp
│ │ │ └── jit_uni_sparse_matmul.hpp
│ │ ├── matmul_inner_product.cpp
│ │ ├── matmul_inner_product.hpp
│ │ ├── prelu
│ │ │ ├── jit_prelu_backward.cpp
│ │ │ ├── jit_prelu_backward.hpp
│ │ │ ├── jit_prelu_base_kernel.cpp
│ │ │ ├── jit_prelu_base_kernel.hpp
│ │ │ ├── jit_prelu_forward.cpp
│ │ │ ├── jit_prelu_forward.hpp
│ │ │ ├── jit_prelu_reduction_kernel.cpp
│ │ │ ├── jit_prelu_reduction_kernel.hpp
│ │ │ ├── jit_prelu_utils.cpp
│ │ │ ├── jit_prelu_utils.hpp
│ │ │ ├── jit_uni_prelu_backward_kernel.cpp
│ │ │ ├── jit_uni_prelu_backward_kernel.hpp
│ │ │ ├── jit_uni_prelu_forward_kernel.cpp
│ │ │ └── jit_uni_prelu_forward_kernel.hpp
│ │ ├── rnn
│ │ │ ├── brgemm_cell_common_bwd.cpp
│ │ │ ├── brgemm_cell_common_bwd.hpp
│ │ │ ├── brgemm_cell_common_fwd.cpp
│ │ │ ├── brgemm_cell_common_fwd.hpp
│ │ │ ├── brgemm_cell_common_reorders.cpp
│ │ │ ├── brgemm_cell_common_reorders.hpp
│ │ │ ├── brgemm_cell_common_utils.cpp
│ │ │ ├── brgemm_cell_common_utils.hpp
│ │ │ ├── jit_brgemm_transpose_single_row.cpp
│ │ │ ├── jit_brgemm_transpose_single_row.hpp
│ │ │ ├── jit_diff_weights_peephole.cpp
│ │ │ ├── jit_diff_weights_peephole.hpp
│ │ │ ├── jit_gates_reduction.cpp
│ │ │ ├── jit_gates_reduction.hpp
│ │ │ ├── jit_uni_gru_cell_postgemm_1_bwd.hpp
│ │ │ ├── jit_uni_gru_cell_postgemm_1_fwd.hpp
│ │ │ ├── jit_uni_gru_cell_postgemm_2_bwd.hpp
│ │ │ ├── jit_uni_gru_cell_postgemm_2_fwd.hpp
│ │ │ ├── jit_uni_gru_lbr_cell_postgemm_bwd.hpp
│ │ │ ├── jit_uni_gru_lbr_cell_postgemm_fwd.hpp
│ │ │ ├── jit_uni_lstm_cell_postgemm.hpp
│ │ │ ├── jit_uni_lstm_cell_postgemm_bwd.hpp
│ │ │ ├── jit_uni_lstm_cell_postgemm_fwd.hpp
│ │ │ ├── jit_uni_lstm_cell_projection_postgemm_fwd.hpp
│ │ │ ├── jit_uni_rnn_cell_postgemm_bwd.hpp
│ │ │ ├── jit_uni_rnn_cell_postgemm_fwd.hpp
│ │ │ ├── jit_uni_rnn_common_postgemm.hpp
│ │ │ ├── rnn_brgemm_utils.cpp
│ │ │ └── rnn_brgemm_utils.hpp
│ │ ├── shuffle
│ │ │ ├── jit_uni_shuffle.cpp
│ │ │ ├── jit_uni_shuffle.hpp
│ │ │ ├── jit_uni_shuffle_kernel.cpp
│ │ │ └── jit_uni_shuffle_kernel.hpp
│ │ ├── ukernel
│ │ │ ├── attr_params.cpp
│ │ │ ├── attr_params.hpp
│ │ │ ├── brgemm.cpp
│ │ │ ├── brgemm.hpp
│ │ │ ├── transform.cpp
│ │ │ └── transform.hpp
│ │ └── utils
│ │ │ ├── jit_io_helper.cpp
│ │ │ └── jit_io_helper.hpp
│ ├── zero_point_utils.cpp
│ └── zero_point_utils.hpp
├── gpu
│ ├── CMakeLists.txt
│ ├── README.md
│ ├── amd
│ │ ├── CMakeLists.txt
│ │ ├── README.md
│ │ ├── engine.cpp
│ │ ├── engine.hpp
│ │ ├── miopen_batch_normalization.cpp
│ │ ├── miopen_batch_normalization.hpp
│ │ ├── miopen_batch_normalization_executor.hpp
│ │ ├── miopen_batch_normalization_impl.hpp
│ │ ├── miopen_binary.cpp
│ │ ├── miopen_binary.hpp
│ │ ├── miopen_binary_impl.hpp
│ │ ├── miopen_conv_filter_adjustment_base.hpp
│ │ ├── miopen_convolution.cpp
│ │ ├── miopen_convolution.hpp
│ │ ├── miopen_convolution_impl.hpp
│ │ ├── miopen_convolution_pd.hpp
│ │ ├── miopen_deconvolution.cpp
│ │ ├── miopen_deconvolution.hpp
│ │ ├── miopen_deconvolution_impl.hpp
│ │ ├── miopen_eltwise.cpp
│ │ ├── miopen_eltwise.hpp
│ │ ├── miopen_eltwise_impl.hpp
│ │ ├── miopen_gemm_inner_product.hpp
│ │ ├── miopen_gemm_inner_product_impl.hpp
│ │ ├── miopen_inner_product.cpp
│ │ ├── miopen_inner_product.hpp
│ │ ├── miopen_inner_product_impl.hpp
│ │ ├── miopen_lrn.cpp
│ │ ├── miopen_lrn.hpp
│ │ ├── miopen_lrn_impl.hpp
│ │ ├── miopen_matmul.cpp
│ │ ├── miopen_matmul.hpp
│ │ ├── miopen_matmul_executor.hpp
│ │ ├── miopen_matmul_impl.hpp
│ │ ├── miopen_pooling.cpp
│ │ ├── miopen_pooling.hpp
│ │ ├── miopen_pooling_impl.hpp
│ │ ├── miopen_reduction.cpp
│ │ ├── miopen_reduction.hpp
│ │ ├── miopen_reduction_impl.hpp
│ │ ├── miopen_reorder.cpp
│ │ ├── miopen_reorder.hpp
│ │ ├── miopen_reorder_impl.hpp
│ │ ├── miopen_softmax.cpp
│ │ ├── miopen_softmax.hpp
│ │ ├── miopen_softmax_impl.hpp
│ │ ├── stream.cpp
│ │ ├── stream.hpp
│ │ ├── sycl_hip_compat.cpp
│ │ ├── sycl_hip_compat.hpp
│ │ ├── sycl_hip_scoped_context.cpp
│ │ ├── sycl_hip_scoped_context.hpp
│ │ ├── sycl_hip_utils.cpp
│ │ └── sycl_hip_utils.hpp
│ ├── generic
│ │ ├── CMakeLists.txt
│ │ ├── convolution_deconvolution.hpp
│ │ ├── cross_engine_reorder.cpp
│ │ ├── cross_engine_reorder.hpp
│ │ ├── direct_copy.hpp
│ │ ├── ref_concat.hpp
│ │ ├── ref_sum.hpp
│ │ └── sycl
│ │ │ ├── CMakeLists.txt
│ │ │ ├── README.md
│ │ │ ├── batch_normalizations_kernels.hpp
│ │ │ ├── binary_kernels.hpp
│ │ │ ├── convolution_kernels.hpp
│ │ │ ├── eltwise_kernels.hpp
│ │ │ ├── engine.cpp
│ │ │ ├── engine.hpp
│ │ │ ├── group_normalization_kernel.hpp
│ │ │ ├── layer_normalizations_kernels.hpp
│ │ │ ├── lrn_kernels.hpp
│ │ │ ├── matmul_kernels.hpp
│ │ │ ├── pooling_kernels.hpp
│ │ │ ├── prelu_kernels.hpp
│ │ │ ├── reduction_kernels.hpp
│ │ │ ├── ref_batch_normalization.cpp
│ │ │ ├── ref_batch_normalization.hpp
│ │ │ ├── ref_binary.cpp
│ │ │ ├── ref_binary.hpp
│ │ │ ├── ref_convolution.cpp
│ │ │ ├── ref_convolution.hpp
│ │ │ ├── ref_deconvolution.cpp
│ │ │ ├── ref_deconvolution.hpp
│ │ │ ├── ref_eltwise.cpp
│ │ │ ├── ref_eltwise.hpp
│ │ │ ├── ref_group_normalization.cpp
│ │ │ ├── ref_group_normalization.hpp
│ │ │ ├── ref_inner_product.cpp
│ │ │ ├── ref_inner_product.hpp
│ │ │ ├── ref_layer_normalizations.cpp
│ │ │ ├── ref_layer_normalizations.hpp
│ │ │ ├── ref_lrn.cpp
│ │ │ ├── ref_lrn.hpp
│ │ │ ├── ref_matmul.cpp
│ │ │ ├── ref_matmul.hpp
│ │ │ ├── ref_pooling.cpp
│ │ │ ├── ref_pooling.hpp
│ │ │ ├── ref_prelu.cpp
│ │ │ ├── ref_prelu.hpp
│ │ │ ├── ref_reduction.cpp
│ │ │ ├── ref_reduction.hpp
│ │ │ ├── ref_reorder.cpp
│ │ │ ├── ref_reorder.hpp
│ │ │ ├── ref_resampling.cpp
│ │ │ ├── ref_resampling.hpp
│ │ │ ├── ref_shuffle.cpp
│ │ │ ├── ref_shuffle.hpp
│ │ │ ├── ref_softmax.cpp
│ │ │ ├── ref_softmax.hpp
│ │ │ ├── ref_sum.cpp
│ │ │ ├── ref_sum.hpp
│ │ │ ├── ref_sum_many_inputs.cpp
│ │ │ ├── ref_sum_many_inputs.hpp
│ │ │ ├── reorder_kernels.hpp
│ │ │ ├── resampling_kernels.hpp
│ │ │ ├── resampling_utils.hpp
│ │ │ ├── rnn
│ │ │ ├── cell_common.cpp
│ │ │ ├── ref_rnn.cpp
│ │ │ ├── ref_rnn.hpp
│ │ │ ├── rnn_kernels.hpp
│ │ │ ├── rnn_utils.cpp
│ │ │ └── rnn_utils.hpp
│ │ │ ├── shuffle_kernels.hpp
│ │ │ ├── simple_reduction.cpp
│ │ │ ├── simple_reduction.hpp
│ │ │ ├── simple_reduction_kernels.hpp
│ │ │ ├── softmax_kernels.hpp
│ │ │ ├── stream.cpp
│ │ │ ├── stream.hpp
│ │ │ ├── sum_kernels.hpp
│ │ │ ├── sycl_gpu_kernel.cpp
│ │ │ ├── sycl_gpu_kernel.hpp
│ │ │ ├── sycl_gpu_primitive.hpp
│ │ │ ├── sycl_io_helper.hpp
│ │ │ ├── sycl_math_utils.hpp
│ │ │ ├── sycl_post_ops.hpp
│ │ │ ├── sycl_primitive_conf.hpp
│ │ │ ├── sycl_q10n.hpp
│ │ │ └── sycl_utils.hpp
│ ├── gpu_batch_normalization_list.cpp
│ ├── gpu_batch_normalization_pd.hpp
│ ├── gpu_binary_list.cpp
│ ├── gpu_binary_pd.hpp
│ ├── gpu_concat_list.cpp
│ ├── gpu_concat_pd.hpp
│ ├── gpu_convolution_list.cpp
│ ├── gpu_convolution_pd.hpp
│ ├── gpu_deconvolution_list.cpp
│ ├── gpu_deconvolution_pd.hpp
│ ├── gpu_eltwise_list.cpp
│ ├── gpu_eltwise_pd.hpp
│ ├── gpu_engine.hpp
│ ├── gpu_gemm_list.cpp
│ ├── gpu_gemm_pd.hpp
│ ├── gpu_group_normalization_list.cpp
│ ├── gpu_impl_list.cpp
│ ├── gpu_impl_list.hpp
│ ├── gpu_inner_product_list.cpp
│ ├── gpu_inner_product_pd.hpp
│ ├── gpu_layer_normalization_list.cpp
│ ├── gpu_layer_normalization_pd.hpp
│ ├── gpu_lrn_list.cpp
│ ├── gpu_lrn_pd.hpp
│ ├── gpu_matmul_list.cpp
│ ├── gpu_matmul_pd.hpp
│ ├── gpu_pooling_list.cpp
│ ├── gpu_pooling_pd.hpp
│ ├── gpu_prelu_list.cpp
│ ├── gpu_prelu_pd.hpp
│ ├── gpu_primitive.hpp
│ ├── gpu_reduction_list.cpp
│ ├── gpu_reduction_pd.hpp
│ ├── gpu_reorder_list.cpp
│ ├── gpu_reorder_pd.cpp
│ ├── gpu_reorder_pd.hpp
│ ├── gpu_resampling_list.cpp
│ ├── gpu_resampling_pd.hpp
│ ├── gpu_resource.hpp
│ ├── gpu_rnn_list.cpp
│ ├── gpu_rnn_pd.hpp
│ ├── gpu_sdpa_list.cpp
│ ├── gpu_shuffle_list.cpp
│ ├── gpu_shuffle_pd.hpp
│ ├── gpu_softmax_list.cpp
│ ├── gpu_softmax_pd.hpp
│ ├── gpu_stream.hpp
│ ├── gpu_sum_list.cpp
│ ├── gpu_sum_pd.hpp
│ ├── gpu_utils.hpp
│ ├── gpu_zero_pad_list.cpp
│ ├── gpu_zero_pad_pd.hpp
│ ├── gpu_zero_points_conv.cpp
│ ├── gpu_zero_points_conv.hpp
│ ├── intel
│ │ ├── CMakeLists.txt
│ │ ├── block_structure.cpp
│ │ ├── block_structure.hpp
│ │ ├── compute
│ │ │ ├── CMakeLists.txt
│ │ │ ├── block_manipulation.cpp
│ │ │ ├── block_manipulation.hpp
│ │ │ ├── compute_engine.cpp
│ │ │ ├── compute_engine.hpp
│ │ │ ├── compute_stream.cpp
│ │ │ ├── compute_stream.hpp
│ │ │ ├── data_type_converter.hpp
│ │ │ ├── device_info.cpp
│ │ │ ├── device_info.hpp
│ │ │ ├── dispatch.cpp
│ │ │ ├── dispatch.hpp
│ │ │ ├── dispatch_reusable.cpp
│ │ │ ├── dispatch_reusable.hpp
│ │ │ ├── kernel.hpp
│ │ │ ├── kernel_arg_list.hpp
│ │ │ ├── kernel_ctx.cpp
│ │ │ ├── kernel_ctx.hpp
│ │ │ ├── utils.hpp
│ │ │ ├── zero_pool.cpp
│ │ │ └── zero_pool.hpp
│ │ ├── config.hpp
│ │ ├── gemm
│ │ │ ├── gpu_gemm.hpp
│ │ │ └── gpu_gemm_exec_types.hpp
│ │ ├── gpu_post_ops.hpp
│ │ ├── gpu_primitive.hpp
│ │ ├── gpu_primitive_attr.hpp
│ │ ├── jit
│ │ │ ├── CMakeLists.txt
│ │ │ ├── README.md
│ │ │ ├── binary_format.cpp
│ │ │ ├── binary_format.hpp
│ │ │ ├── codegen
│ │ │ │ ├── bank_conflict_allocation.cpp
│ │ │ │ ├── bank_conflict_allocation.hpp
│ │ │ │ ├── codegen.cpp
│ │ │ │ ├── codegen.hpp
│ │ │ │ ├── kernel.hpp
│ │ │ │ ├── ngen_helpers.hpp
│ │ │ │ ├── operand.cpp
│ │ │ │ ├── operand.hpp
│ │ │ │ ├── reduce.hpp
│ │ │ │ ├── reg_buf.hpp
│ │ │ │ ├── register_allocator.hpp
│ │ │ │ ├── register_scope.hpp
│ │ │ │ ├── reorder.hpp
│ │ │ │ └── send.hpp
│ │ │ ├── config
│ │ │ │ ├── gemmstone_config.cpp
│ │ │ │ ├── gemmstone_config.hpp
│ │ │ │ └── ngen_config.hpp
│ │ │ ├── conv
│ │ │ │ ├── README.md
│ │ │ │ ├── config.cpp
│ │ │ │ ├── config.hpp
│ │ │ │ ├── conv_kernel.hpp
│ │ │ │ ├── gen_convolution.cpp
│ │ │ │ ├── gen_convolution.hpp
│ │ │ │ ├── grf_usage.cpp
│ │ │ │ ├── grf_usage.hpp
│ │ │ │ ├── ir_builder.cpp
│ │ │ │ ├── ir_builder.hpp
│ │ │ │ ├── key.cpp
│ │ │ │ ├── key.hpp
│ │ │ │ ├── lookup_table.cpp
│ │ │ │ ├── lookup_table.hpp
│ │ │ │ ├── lookup_table_data.cpp
│ │ │ │ ├── message_patterns.hpp
│ │ │ │ ├── model.hpp
│ │ │ │ ├── model_bridge.cpp
│ │ │ │ ├── model_bridge.hpp
│ │ │ │ ├── model_data.hpp
│ │ │ │ ├── model_xehpc_common_data.cpp
│ │ │ │ ├── model_xehpc_dw_data.cpp
│ │ │ │ ├── model_xehpg_common_data.cpp
│ │ │ │ ├── model_xehpg_dw_data.cpp
│ │ │ │ ├── normalization.cpp
│ │ │ │ ├── normalization.hpp
│ │ │ │ ├── pipeline.cpp
│ │ │ │ ├── pipeline.hpp
│ │ │ │ ├── plan.cpp
│ │ │ │ ├── plan.hpp
│ │ │ │ ├── plan_utils.hpp
│ │ │ │ ├── problem.cpp
│ │ │ │ ├── problem.hpp
│ │ │ │ ├── tiler.cpp
│ │ │ │ ├── tiler.hpp
│ │ │ │ ├── zero_out.cpp
│ │ │ │ ├── zero_out.hpp
│ │ │ │ ├── zp_plan.cpp
│ │ │ │ └── zp_plan.hpp
│ │ │ ├── eltwise_injector.cpp
│ │ │ ├── eltwise_injector.hpp
│ │ │ ├── emulated_generator.cpp
│ │ │ ├── emulated_generator.hpp
│ │ │ ├── gemm
│ │ │ │ ├── .clang-tidy
│ │ │ │ ├── CMakeLists.txt
│ │ │ │ ├── gemm_walk_orders.hpp
│ │ │ │ ├── gen_gemm.cpp
│ │ │ │ ├── gen_gemm.hpp
│ │ │ │ ├── gen_gemm_kernel.cpp
│ │ │ │ ├── gen_gemm_kernel.hpp
│ │ │ │ ├── gen_gemm_kernel_db.cpp
│ │ │ │ ├── gen_gemm_kernel_db.hpp
│ │ │ │ ├── generator
│ │ │ │ │ ├── .clang-format
│ │ │ │ │ ├── generator.cpp
│ │ │ │ │ ├── microkernel_provider.cpp
│ │ │ │ │ ├── pieces
│ │ │ │ │ │ ├── address_setup.cxx
│ │ │ │ │ │ ├── alloc_utils.cpp
│ │ │ │ │ │ ├── alloc_utils.hpp
│ │ │ │ │ │ ├── allocators.cpp
│ │ │ │ │ │ ├── allocators.hpp
│ │ │ │ │ │ ├── asm_helpers.cxx
│ │ │ │ │ │ ├── atomic_fusions.cxx
│ │ │ │ │ │ ├── atomic_fusions.hpp
│ │ │ │ │ │ ├── c_update.cxx
│ │ │ │ │ │ ├── common.cxx
│ │ │ │ │ │ ├── compute_utils.hpp
│ │ │ │ │ │ ├── cooperative_split.cpp
│ │ │ │ │ │ ├── cooperative_split.hpp
│ │ │ │ │ │ ├── copy.cxx
│ │ │ │ │ │ ├── copy_plan.cpp
│ │ │ │ │ │ ├── copy_plan.hpp
│ │ │ │ │ │ ├── driver_info.cxx
│ │ │ │ │ │ ├── emulation.cxx
│ │ │ │ │ │ ├── gemm.cxx
│ │ │ │ │ │ ├── gemm_microkernel.cxx
│ │ │ │ │ │ ├── gemm_setup.cxx
│ │ │ │ │ │ ├── grf_multirange.hpp
│ │ │ │ │ │ ├── hw_template_instantiations.cxx
│ │ │ │ │ │ ├── hw_utils.hpp
│ │ │ │ │ │ ├── invert.hpp
│ │ │ │ │ │ ├── k_loop.cxx
│ │ │ │ │ │ ├── k_loop_setup.cxx
│ │ │ │ │ │ ├── kernel_queries.cpp
│ │ │ │ │ │ ├── kernel_queries.hpp
│ │ │ │ │ │ ├── l3_prefetch.cxx
│ │ │ │ │ │ ├── layout_setup.cxx
│ │ │ │ │ │ ├── layout_utils.cpp
│ │ │ │ │ │ ├── layout_utils.hpp
│ │ │ │ │ │ ├── loop_sequencer.cpp
│ │ │ │ │ │ ├── loop_sequencer.hpp
│ │ │ │ │ │ ├── map.hpp
│ │ │ │ │ │ ├── masks.cxx
│ │ │ │ │ │ ├── math_helpers.cxx
│ │ │ │ │ │ ├── matrix_access.cxx
│ │ │ │ │ │ ├── matrix_multiply.cxx
│ │ │ │ │ │ ├── monolithic_k_loop_dpasw.cxx
│ │ │ │ │ │ ├── ngen_object_helpers.cpp
│ │ │ │ │ │ ├── ngen_object_helpers.hpp
│ │ │ │ │ │ ├── post_ops.cxx
│ │ │ │ │ │ ├── problem_utils.cpp
│ │ │ │ │ │ ├── quantization.cpp
│ │ │ │ │ │ ├── quantization.cxx
│ │ │ │ │ │ ├── quantization.hpp
│ │ │ │ │ │ ├── register_allocation.cxx
│ │ │ │ │ │ ├── register_layout.cpp
│ │ │ │ │ │ ├── register_layout.hpp
│ │ │ │ │ │ ├── remask.cpp
│ │ │ │ │ │ ├── remask.cxx
│ │ │ │ │ │ ├── remask.hpp
│ │ │ │ │ │ ├── row_column_sums.cxx
│ │ │ │ │ │ ├── state.cpp
│ │ │ │ │ │ ├── state.hpp
│ │ │ │ │ │ ├── state_debug.cpp
│ │ │ │ │ │ ├── state_utils.cpp
│ │ │ │ │ │ ├── state_utils.cxx
│ │ │ │ │ │ ├── state_utils.hpp
│ │ │ │ │ │ ├── stream_k.cxx
│ │ │ │ │ │ ├── tlb_warmup.cxx
│ │ │ │ │ │ ├── token_alloc_utils.cpp
│ │ │ │ │ │ ├── token_alloc_utils.hpp
│ │ │ │ │ │ └── walk_orders.cxx
│ │ │ │ │ ├── strategy.cpp
│ │ │ │ │ └── strategy_parser.cpp
│ │ │ │ ├── include
│ │ │ │ │ ├── .clang-format
│ │ │ │ │ ├── gemmstone
│ │ │ │ │ │ ├── config.hpp
│ │ │ │ │ │ ├── driver_info.hpp
│ │ │ │ │ │ ├── generator.hpp
│ │ │ │ │ │ ├── kernel_catalog.hpp
│ │ │ │ │ │ ├── kernel_evaluator.hpp
│ │ │ │ │ │ ├── kernel_selector.hpp
│ │ │ │ │ │ ├── microkernel_provider.hpp
│ │ │ │ │ │ ├── problem.hpp
│ │ │ │ │ │ ├── strategy.hpp
│ │ │ │ │ │ ├── strategy_parser.hpp
│ │ │ │ │ │ └── type.hpp
│ │ │ │ │ └── internal
│ │ │ │ │ │ ├── generator_inline.hxx
│ │ │ │ │ │ ├── namespace_end.hxx
│ │ │ │ │ │ ├── namespace_start.hxx
│ │ │ │ │ │ ├── ngen_includes.hpp
│ │ │ │ │ │ └── utils.hpp
│ │ │ │ ├── jit_gemm_pd.cpp
│ │ │ │ ├── jit_gemm_pd.hpp
│ │ │ │ ├── selector
│ │ │ │ │ ├── .clang-format
│ │ │ │ │ ├── db
│ │ │ │ │ │ ├── kernel.db
│ │ │ │ │ │ ├── ukernel_lmr.db
│ │ │ │ │ │ ├── ukernel_mlr.db
│ │ │ │ │ │ └── ukernel_mmr.db
│ │ │ │ │ ├── kernel_evaluator.cpp
│ │ │ │ │ └── kernel_selector.cpp
│ │ │ │ ├── xe_hp_systolic_gemm.cpp
│ │ │ │ └── xe_hp_systolic_gemm.hpp
│ │ │ ├── generator.cpp
│ │ │ ├── generator.hpp
│ │ │ ├── generator_base.hpp
│ │ │ ├── ir
│ │ │ │ ├── README.md
│ │ │ │ ├── block_2d_utils.hpp
│ │ │ │ ├── blocking.cpp
│ │ │ │ ├── blocking.hpp
│ │ │ │ ├── config.hpp
│ │ │ │ ├── core.cpp
│ │ │ │ ├── core.hpp
│ │ │ │ ├── eltwise.hpp
│ │ │ │ ├── epilogue.cpp
│ │ │ │ ├── epilogue.hpp
│ │ │ │ ├── fma.cpp
│ │ │ │ ├── fma.hpp
│ │ │ │ ├── gemm_schedule.cpp
│ │ │ │ ├── gemm_schedule.hpp
│ │ │ │ ├── grf_permutation.hpp
│ │ │ │ ├── hw.hpp
│ │ │ │ ├── ir.cpp
│ │ │ │ ├── ir.hpp
│ │ │ │ ├── ir_builder.cpp
│ │ │ │ ├── ir_builder.hpp
│ │ │ │ ├── kernel_desc.hpp
│ │ │ │ ├── kernel_info.hpp
│ │ │ │ ├── linear_expr.cpp
│ │ │ │ ├── linear_expr.hpp
│ │ │ │ ├── message.cpp
│ │ │ │ ├── message.hpp
│ │ │ │ ├── message_patterns.hpp
│ │ │ │ ├── post_ops.cpp
│ │ │ │ ├── post_ops.hpp
│ │ │ │ ├── primitive_plan.cpp
│ │ │ │ ├── primitive_plan.hpp
│ │ │ │ ├── problem.cpp
│ │ │ │ ├── problem.hpp
│ │ │ │ ├── reduce.cpp
│ │ │ │ ├── reduce.hpp
│ │ │ │ ├── reorder.hpp
│ │ │ │ ├── send_plan.cpp
│ │ │ │ ├── send_plan.hpp
│ │ │ │ ├── slm_reduce_builder.cpp
│ │ │ │ ├── slm_reduce_builder.hpp
│ │ │ │ ├── tensor.cpp
│ │ │ │ ├── tensor.hpp
│ │ │ │ ├── tensor_config.cpp
│ │ │ │ ├── tensor_config.hpp
│ │ │ │ └── walk_order.hpp
│ │ │ ├── pass
│ │ │ │ ├── alloc.cpp
│ │ │ │ ├── alloc.hpp
│ │ │ │ ├── bank_conflict.cpp
│ │ │ │ ├── bank_conflict.hpp
│ │ │ │ ├── barrier.cpp
│ │ │ │ ├── barrier.hpp
│ │ │ │ ├── cse.cpp
│ │ │ │ ├── cse.hpp
│ │ │ │ ├── dp4a.cpp
│ │ │ │ ├── dp4a.hpp
│ │ │ │ ├── dpas.cpp
│ │ │ │ ├── dpas.hpp
│ │ │ │ ├── dpasw.cpp
│ │ │ │ ├── dpasw.hpp
│ │ │ │ ├── expr_scalarizer.hpp
│ │ │ │ ├── hoist.cpp
│ │ │ │ ├── hoist.hpp
│ │ │ │ ├── overflow.cpp
│ │ │ │ ├── overflow.hpp
│ │ │ │ ├── pass.cpp
│ │ │ │ ├── pass.hpp
│ │ │ │ ├── peephole.cpp
│ │ │ │ ├── peephole.hpp
│ │ │ │ ├── send.cpp
│ │ │ │ ├── send.hpp
│ │ │ │ ├── shuffle_splitter.cpp
│ │ │ │ ├── shuffle_splitter.hpp
│ │ │ │ ├── simplify.cpp
│ │ │ │ ├── simplify.hpp
│ │ │ │ ├── slm.cpp
│ │ │ │ ├── slm.hpp
│ │ │ │ ├── strength_reduce.cpp
│ │ │ │ ├── strength_reduce.hpp
│ │ │ │ ├── unroll.cpp
│ │ │ │ └── unroll.hpp
│ │ │ ├── pooling
│ │ │ │ ├── config.hpp
│ │ │ │ ├── gen_pooling.cpp
│ │ │ │ ├── gen_pooling.hpp
│ │ │ │ ├── ir_builder.cpp
│ │ │ │ ├── ir_builder.hpp
│ │ │ │ └── pooling_kernel.hpp
│ │ │ ├── post_op_injector.cpp
│ │ │ ├── post_op_injector.hpp
│ │ │ ├── reduction.cpp
│ │ │ ├── reduction.hpp
│ │ │ ├── reduction_generator.hpp
│ │ │ ├── reduction_injector.cpp
│ │ │ ├── reduction_injector.hpp
│ │ │ ├── reorder
│ │ │ │ ├── config.cpp
│ │ │ │ ├── config.hpp
│ │ │ │ ├── gen_reorder.cpp
│ │ │ │ ├── gen_reorder.hpp
│ │ │ │ ├── ir_builder.cpp
│ │ │ │ ├── ir_builder.hpp
│ │ │ │ ├── normalization.cpp
│ │ │ │ ├── normalization.hpp
│ │ │ │ ├── reorder_kernel.hpp
│ │ │ │ ├── tiler.cpp
│ │ │ │ └── tiler.hpp
│ │ │ ├── utils
│ │ │ │ ├── iterator.hpp
│ │ │ │ ├── ngen_type_bridge.hpp
│ │ │ │ ├── range.hpp
│ │ │ │ ├── trace.cpp
│ │ │ │ ├── trace.hpp
│ │ │ │ ├── utils.cpp
│ │ │ │ └── utils.hpp
│ │ │ └── v2
│ │ │ │ ├── conv
│ │ │ │ ├── README.md
│ │ │ │ ├── bench_data.cpp
│ │ │ │ ├── bench_data.hpp
│ │ │ │ ├── bridge.hpp
│ │ │ │ ├── builder.cpp
│ │ │ │ ├── builder.hpp
│ │ │ │ ├── debug.cpp
│ │ │ │ ├── debug.hpp
│ │ │ │ ├── gen_convolution.cpp
│ │ │ │ ├── gen_convolution.hpp
│ │ │ │ ├── kernel.hpp
│ │ │ │ ├── kernel_desc.cpp
│ │ │ │ ├── kernel_desc.hpp
│ │ │ │ ├── kernel_desc_2d_reqs.cpp
│ │ │ │ ├── model.cpp
│ │ │ │ ├── model.hpp
│ │ │ │ ├── plan.cpp
│ │ │ │ ├── plan.hpp
│ │ │ │ ├── plan_registry.cpp
│ │ │ │ ├── plan_registry.hpp
│ │ │ │ ├── plan_registry_data.cpp
│ │ │ │ ├── planner
│ │ │ │ │ ├── CMakeLists.txt
│ │ │ │ │ ├── bench.cpp
│ │ │ │ │ ├── bench.hpp
│ │ │ │ │ ├── model_fit.cpp
│ │ │ │ │ ├── model_fit.hpp
│ │ │ │ │ ├── planner.cpp
│ │ │ │ │ ├── planner.hpp
│ │ │ │ │ ├── planner_main.cpp
│ │ │ │ │ ├── search.cpp
│ │ │ │ │ └── search.hpp
│ │ │ │ ├── problem.cpp
│ │ │ │ ├── problem.hpp
│ │ │ │ ├── tensor_utils.cpp
│ │ │ │ └── tensor_utils.hpp
│ │ │ │ └── ir
│ │ │ │ ├── bridge.hpp
│ │ │ │ ├── builder.cpp
│ │ │ │ ├── builder.hpp
│ │ │ │ ├── plan.hpp
│ │ │ │ ├── plan_utils.hpp
│ │ │ │ ├── reqs.cpp
│ │ │ │ ├── reqs.hpp
│ │ │ │ ├── send.cpp
│ │ │ │ ├── send.hpp
│ │ │ │ ├── tensor.cpp
│ │ │ │ └── tensor.hpp
│ │ ├── kernel_cache.cpp
│ │ ├── kernel_cache.hpp
│ │ ├── logging.hpp
│ │ ├── microkernels
│ │ │ ├── .clang-tidy
│ │ │ ├── CMakeLists.txt
│ │ │ ├── elf.hpp
│ │ │ ├── entrance_agent.cpp
│ │ │ ├── entrance_agent.hpp
│ │ │ ├── fuser.cpp
│ │ │ ├── fuser.hpp
│ │ │ ├── internal_utilities.hpp
│ │ │ ├── package.hpp
│ │ │ ├── protocol.cpp
│ │ │ ├── protocol.hpp
│ │ │ ├── shim.cpp
│ │ │ └── shim.hpp
│ │ ├── ocl
│ │ │ ├── CMakeLists.txt
│ │ │ ├── binary_common.h
│ │ │ ├── bnorm
│ │ │ │ ├── lookup_table.cpp
│ │ │ │ ├── lookup_table.hpp
│ │ │ │ ├── model.cpp
│ │ │ │ ├── model.hpp
│ │ │ │ ├── nhwc_batch_normalization.cpp
│ │ │ │ ├── nhwc_batch_normalization.hpp
│ │ │ │ ├── nhwc_reusable.cl
│ │ │ │ ├── nhwc_reusable.cpp
│ │ │ │ ├── nhwc_reusable.h
│ │ │ │ ├── nhwc_reusable.hpp
│ │ │ │ ├── ref_batch_normalization.cpp
│ │ │ │ ├── ref_batch_normalization.hpp
│ │ │ │ ├── ref_bnorm.cl
│ │ │ │ ├── reusable_bnorm.cl
│ │ │ │ ├── reusable_bnorm.cpp
│ │ │ │ ├── reusable_bnorm.hpp
│ │ │ │ ├── simple_bnorm.cl
│ │ │ │ ├── simple_bnorm.cpp
│ │ │ │ ├── simple_bnorm.hpp
│ │ │ │ ├── utils.cpp
│ │ │ │ ├── utils.hpp
│ │ │ │ ├── xe_batch_normalization.cpp
│ │ │ │ ├── xe_batch_normalization.hpp
│ │ │ │ ├── xe_bnorm.h
│ │ │ │ ├── xe_bnorm_bwd.cl
│ │ │ │ ├── xe_bnorm_fwd.cl
│ │ │ │ ├── xe_bnorm_nhwc_bwd.cl
│ │ │ │ ├── xe_bnorm_nhwc_fwd.cl
│ │ │ │ └── xe_bnorm_reduce.h
│ │ │ ├── concat_common.h
│ │ │ ├── concat_utils.hpp
│ │ │ ├── convolution_deconvolution.hpp
│ │ │ ├── convolution_inner_product.cpp
│ │ │ ├── convolution_inner_product.hpp
│ │ │ ├── custom_reorder.cl
│ │ │ ├── custom_reorder.cpp
│ │ │ ├── custom_reorder.hpp
│ │ │ ├── deconv_backward_bias.cl
│ │ │ ├── device_info.cpp
│ │ │ ├── device_info.hpp
│ │ │ ├── dispatch.h
│ │ │ ├── engine.cpp
│ │ │ ├── engine.hpp
│ │ │ ├── gemm
│ │ │ │ ├── conv_gemm.hpp
│ │ │ │ ├── gemm_with_post_ops.cl
│ │ │ │ ├── gemm_with_post_ops.cpp
│ │ │ │ ├── gemm_with_post_ops.hpp
│ │ │ │ ├── ocl_gemm_attrs.h
│ │ │ │ ├── ref_gemm.cl
│ │ │ │ ├── ref_gemm.cpp
│ │ │ │ ├── ref_gemm.hpp
│ │ │ │ ├── xe_hp_systolic_gemm_copy.cl
│ │ │ │ ├── xe_hpc_systolic_gemm_copy.cl
│ │ │ │ └── xe_systolic_gemm_copy_kernel.hpp
│ │ │ ├── gemm_inner_product.cpp
│ │ │ ├── gemm_inner_product.hpp
│ │ │ ├── gemm_matmul.cpp
│ │ │ ├── gemm_matmul.hpp
│ │ │ ├── generic_reorder.cl
│ │ │ ├── generic_reorder.cpp
│ │ │ ├── generic_reorder.hpp
│ │ │ ├── graph
│ │ │ │ └── gen_index.cl
│ │ │ ├── hw_info.cpp
│ │ │ ├── hw_info.hpp
│ │ │ ├── kernel.cpp
│ │ │ ├── kernel.hpp
│ │ │ ├── layer_norm_common.h
│ │ │ ├── lnorm_utils.hpp
│ │ │ ├── many_inputs_sum.cl
│ │ │ ├── many_inputs_sum.cpp
│ │ │ ├── many_inputs_sum.hpp
│ │ │ ├── mdapi_utils.cpp
│ │ │ ├── mdapi_utils.hpp
│ │ │ ├── micro_sdpa.cl
│ │ │ ├── micro_sdpa.cpp
│ │ │ ├── micro_sdpa.hpp
│ │ │ ├── micro_sdpa_configs.cpp
│ │ │ ├── micro_sdpa_configs.hpp
│ │ │ ├── multi_concat.hpp
│ │ │ ├── multi_po_reorder_binary.hpp
│ │ │ ├── multi_po_reorder_sum.hpp
│ │ │ ├── ocl_conversion.h
│ │ │ ├── ocl_custom_types.h
│ │ │ ├── ocl_eltwise.h
│ │ │ ├── ocl_generic_vector_ops.h
│ │ │ ├── ocl_io.h
│ │ │ ├── ocl_kernel_list.cpp.in
│ │ │ ├── ocl_math_utils.h
│ │ │ ├── ocl_overrides.md
│ │ │ ├── ocl_philox.h
│ │ │ ├── ocl_post_ops.h
│ │ │ ├── ocl_scales.h
│ │ │ ├── ocl_types.h
│ │ │ ├── ocl_types_specific.h
│ │ │ ├── ocl_utils.h
│ │ │ ├── offsets.h
│ │ │ ├── reduction
│ │ │ │ ├── atomic_reduction.cl
│ │ │ │ ├── atomic_reduction.cpp
│ │ │ │ ├── atomic_reduction.hpp
│ │ │ │ ├── combined_reduction.cl
│ │ │ │ ├── combined_reduction.cpp
│ │ │ │ ├── combined_reduction.hpp
│ │ │ │ ├── ocl_reduction.h
│ │ │ │ ├── ref_reduction.cl
│ │ │ │ ├── ref_reduction.cpp
│ │ │ │ ├── ref_reduction.hpp
│ │ │ │ ├── reusable_ref_reduction.cl
│ │ │ │ ├── reusable_ref_reduction.cpp
│ │ │ │ ├── reusable_ref_reduction.hpp
│ │ │ │ ├── utils.cpp
│ │ │ │ └── utils.hpp
│ │ │ ├── ref_convolution.cl
│ │ │ ├── ref_convolution.cpp
│ │ │ ├── ref_convolution.hpp
│ │ │ ├── ref_eltwise.cl
│ │ │ ├── ref_eltwise.cpp
│ │ │ ├── ref_eltwise.hpp
│ │ │ ├── ref_group_normalization.cl
│ │ │ ├── ref_group_normalization.cpp
│ │ │ ├── ref_group_normalization.hpp
│ │ │ ├── ref_inner_product.cl
│ │ │ ├── ref_inner_product.cpp
│ │ │ ├── ref_inner_product.hpp
│ │ │ ├── ref_layer_normalization.cl
│ │ │ ├── ref_layer_normalization.cpp
│ │ │ ├── ref_layer_normalization.hpp
│ │ │ ├── ref_lrn.cl
│ │ │ ├── ref_lrn.cpp
│ │ │ ├── ref_lrn.hpp
│ │ │ ├── ref_matmul.cl
│ │ │ ├── ref_matmul.cpp
│ │ │ ├── ref_matmul.hpp
│ │ │ ├── ref_pooling.cl
│ │ │ ├── ref_pooling.cpp
│ │ │ ├── ref_pooling.hpp
│ │ │ ├── ref_prelu.cl
│ │ │ ├── ref_prelu.cpp
│ │ │ ├── ref_prelu.hpp
│ │ │ ├── ref_reorder.cl
│ │ │ ├── ref_reorder.cpp
│ │ │ ├── ref_reorder.hpp
│ │ │ ├── ref_resampling.cl
│ │ │ ├── ref_resampling.cpp
│ │ │ ├── ref_resampling.hpp
│ │ │ ├── ref_sdpa.cl
│ │ │ ├── ref_sdpa.cpp
│ │ │ ├── ref_sdpa.hpp
│ │ │ ├── ref_shuffle.cl
│ │ │ ├── ref_shuffle.cpp
│ │ │ ├── ref_shuffle.hpp
│ │ │ ├── ref_sparse_matmul.cl
│ │ │ ├── ref_sparse_matmul.cpp
│ │ │ ├── ref_sparse_matmul.hpp
│ │ │ ├── reorder_common.h
│ │ │ ├── reusable_lnorm.cl
│ │ │ ├── reusable_lnorm.cpp
│ │ │ ├── reusable_lnorm.hpp
│ │ │ ├── reusable_simple_concat.cl
│ │ │ ├── reusable_simple_concat.cpp
│ │ │ ├── reusable_simple_concat.hpp
│ │ │ ├── reusable_softmax.cl
│ │ │ ├── reusable_softmax.cpp
│ │ │ ├── reusable_softmax.hpp
│ │ │ ├── reusable_vectorized_lnorm.cl
│ │ │ ├── reusable_vectorized_lnorm.cpp
│ │ │ ├── reusable_vectorized_lnorm.hpp
│ │ │ ├── rnn
│ │ │ │ ├── cell_common.cpp
│ │ │ │ ├── cell_compute.h
│ │ │ │ ├── cell_gru.cpp
│ │ │ │ ├── cell_gru_lbr.cpp
│ │ │ │ ├── cell_kind_utility.h
│ │ │ │ ├── common.h
│ │ │ │ ├── grid.cl
│ │ │ │ ├── grid.cpp
│ │ │ │ ├── grid.hpp
│ │ │ │ ├── reorders.cpp
│ │ │ │ ├── reorders.hpp
│ │ │ │ ├── rnn_reorder.cl
│ │ │ │ ├── simple_cell_fusion.cpp
│ │ │ │ ├── simple_cell_fusion.hpp
│ │ │ │ ├── simple_postgemm.cpp
│ │ │ │ ├── utils.cpp
│ │ │ │ └── utils.hpp
│ │ │ ├── sdpa_utils.h
│ │ │ ├── shuffle_by_reorder.hpp
│ │ │ ├── simple_binary.cl
│ │ │ ├── simple_binary.cpp
│ │ │ ├── simple_binary.hpp
│ │ │ ├── simple_layer_normalization.cl
│ │ │ ├── simple_layer_normalization.cpp
│ │ │ ├── simple_layer_normalization.hpp
│ │ │ ├── simple_softmax.cl
│ │ │ ├── simple_softmax.cpp
│ │ │ ├── simple_softmax.h
│ │ │ ├── simple_softmax.hpp
│ │ │ ├── simple_sum.cl
│ │ │ ├── simple_sum.cpp
│ │ │ ├── simple_sum.hpp
│ │ │ ├── simple_zero_pad.cl
│ │ │ ├── simple_zero_pad.cpp
│ │ │ ├── simple_zero_pad.hpp
│ │ │ ├── stream.cpp
│ │ │ ├── stream.hpp
│ │ │ ├── subbyte_pack.cl
│ │ │ ├── tile_ops.h
│ │ │ ├── types_interop.h
│ │ │ ├── types_interop.hpp
│ │ │ ├── usm_utils.cpp
│ │ │ ├── usm_utils.hpp
│ │ │ ├── utils.cpp
│ │ │ ├── utils.hpp
│ │ │ ├── vectorized_lnorm.cl
│ │ │ ├── vectorized_lnorm.cpp
│ │ │ ├── vectorized_lnorm.hpp
│ │ │ ├── vectorized_lnorm_fused.cl
│ │ │ ├── vectorized_resampling.cl
│ │ │ ├── vectorized_resampling.cpp
│ │ │ ├── vectorized_resampling.hpp
│ │ │ ├── xe_binary.cl
│ │ │ ├── xe_binary.cpp
│ │ │ ├── xe_binary.hpp
│ │ │ ├── xe_concat.cl
│ │ │ ├── xe_concat.cpp
│ │ │ ├── xe_concat.hpp
│ │ │ ├── xe_eltwise.cl
│ │ │ ├── xe_eltwise.cpp
│ │ │ ├── xe_eltwise.hpp
│ │ │ ├── xe_global_pooling.cl
│ │ │ ├── xe_global_pooling.cpp
│ │ │ ├── xe_global_pooling.hpp
│ │ │ ├── xe_pooling.cl
│ │ │ ├── xe_pooling.cpp
│ │ │ ├── xe_pooling.hpp
│ │ │ ├── xe_softmax.cl
│ │ │ ├── xe_softmax.cpp
│ │ │ ├── xe_softmax.hpp
│ │ │ ├── xe_sum.cl
│ │ │ ├── xe_sum.cpp
│ │ │ ├── xe_sum.hpp
│ │ │ ├── xe_wino_conv_fwd_data_2x3.cl
│ │ │ ├── xe_wino_conv_fwd_data_fused.cl
│ │ │ ├── xe_wino_convolution.cpp
│ │ │ └── xe_wino_convolution.hpp
│ │ ├── primitive_conf.cpp
│ │ ├── primitive_conf.hpp
│ │ ├── sycl
│ │ │ ├── CMakeLists.txt
│ │ │ ├── compat.cpp
│ │ │ ├── compat.hpp
│ │ │ ├── device_info.cpp
│ │ │ ├── device_info.hpp
│ │ │ ├── engine.cpp
│ │ │ ├── engine.hpp
│ │ │ ├── interop_kernel.cpp
│ │ │ ├── interop_kernel.hpp
│ │ │ ├── l0
│ │ │ │ ├── utils.cpp
│ │ │ │ └── utils.hpp
│ │ │ ├── stream.cpp
│ │ │ ├── stream.hpp
│ │ │ ├── utils.cpp
│ │ │ └── utils.hpp
│ │ ├── utils.cpp
│ │ └── utils.hpp
│ └── nvidia
│ │ ├── CMakeLists.txt
│ │ ├── README.md
│ │ ├── cudnn_batch_normalization.cpp
│ │ ├── cudnn_batch_normalization.hpp
│ │ ├── cudnn_batch_normalization_executor.hpp
│ │ ├── cudnn_batch_normalization_impl.hpp
│ │ ├── cudnn_binary.cpp
│ │ ├── cudnn_binary.hpp
│ │ ├── cudnn_binary_impl.hpp
│ │ ├── cudnn_conv_filter_adjustment_base.hpp
│ │ ├── cudnn_conv_inner_product.hpp
│ │ ├── cudnn_conv_inner_product_impl.hpp
│ │ ├── cudnn_convolution.cpp
│ │ ├── cudnn_convolution.hpp
│ │ ├── cudnn_convolution_impl.hpp
│ │ ├── cudnn_convolution_pd.hpp
│ │ ├── cudnn_deconvolution.cpp
│ │ ├── cudnn_deconvolution.hpp
│ │ ├── cudnn_deconvolution_impl.hpp
│ │ ├── cudnn_eltwise.cpp
│ │ ├── cudnn_eltwise.hpp
│ │ ├── cudnn_eltwise_impl.hpp
│ │ ├── cudnn_gemm_inner_product.hpp
│ │ ├── cudnn_gemm_inner_product_impl.hpp
│ │ ├── cudnn_inner_product.cpp
│ │ ├── cudnn_inner_product.hpp
│ │ ├── cudnn_inner_product_impl.hpp
│ │ ├── cudnn_lrn.cpp
│ │ ├── cudnn_lrn.hpp
│ │ ├── cudnn_lrn_impl.hpp
│ │ ├── cudnn_matmul.cpp
│ │ ├── cudnn_matmul.hpp
│ │ ├── cudnn_matmul_base_impl.hpp
│ │ ├── cudnn_matmul_executor.hpp
│ │ ├── cudnn_matmul_impl.hpp
│ │ ├── cudnn_matmul_lt.hpp
│ │ ├── cudnn_matmul_lt_impl.hpp
│ │ ├── cudnn_pooling.cpp
│ │ ├── cudnn_pooling.hpp
│ │ ├── cudnn_pooling_impl.hpp
│ │ ├── cudnn_reduction.cpp
│ │ ├── cudnn_reduction.hpp
│ │ ├── cudnn_reduction_impl.hpp
│ │ ├── cudnn_reorder.cpp
│ │ ├── cudnn_reorder.hpp
│ │ ├── cudnn_reorder_impl.hpp
│ │ ├── cudnn_reorder_lt.cpp
│ │ ├── cudnn_reorder_lt.hpp
│ │ ├── cudnn_reorder_lt_impl.hpp
│ │ ├── cudnn_softmax.cpp
│ │ ├── cudnn_softmax.hpp
│ │ ├── cudnn_softmax_impl.hpp
│ │ ├── cudnn_sum.hpp
│ │ ├── engine.cpp
│ │ ├── engine.hpp
│ │ ├── stream.cpp
│ │ ├── stream.hpp
│ │ ├── sycl_cuda_compat.cpp
│ │ ├── sycl_cuda_compat.hpp
│ │ ├── sycl_cuda_scoped_context.cpp
│ │ ├── sycl_cuda_scoped_context.hpp
│ │ ├── sycl_cuda_stream_utils.hpp
│ │ ├── sycl_cuda_utils.cpp
│ │ └── sycl_cuda_utils.hpp
├── graph
│ ├── CMakeLists.txt
│ ├── backend
│ │ ├── CMakeLists.txt
│ │ ├── dnnl
│ │ │ ├── CMakeLists.txt
│ │ │ ├── common.cpp
│ │ │ ├── common.hpp
│ │ │ ├── dnnl_backend.cpp
│ │ │ ├── dnnl_backend.hpp
│ │ │ ├── dnnl_constant_tensor_cache.hpp
│ │ │ ├── dnnl_op_def.hpp
│ │ │ ├── dnnl_opset.hpp
│ │ │ ├── dnnl_partition_impl.cpp
│ │ │ ├── dnnl_partition_impl.hpp
│ │ │ ├── dnnl_shape_infer.cpp
│ │ │ ├── dnnl_shape_infer.hpp
│ │ │ ├── fusion_info.cpp
│ │ │ ├── fusion_info.hpp
│ │ │ ├── internal_attrs.hpp
│ │ │ ├── internal_ops.hpp
│ │ │ ├── kernels
│ │ │ │ ├── batch_norm.cpp
│ │ │ │ ├── batch_norm.hpp
│ │ │ │ ├── binary.cpp
│ │ │ │ ├── binary.hpp
│ │ │ │ ├── concat.cpp
│ │ │ │ ├── concat.hpp
│ │ │ │ ├── conv.cpp
│ │ │ │ ├── conv.hpp
│ │ │ │ ├── conv_base.cpp
│ │ │ │ ├── conv_base.hpp
│ │ │ │ ├── conv_transpose.cpp
│ │ │ │ ├── conv_transpose.hpp
│ │ │ │ ├── dummy.cpp
│ │ │ │ ├── dummy.hpp
│ │ │ │ ├── eltwise.cpp
│ │ │ │ ├── eltwise.hpp
│ │ │ │ ├── gen_index.cpp
│ │ │ │ ├── gen_index.hpp
│ │ │ │ ├── group_norm.cpp
│ │ │ │ ├── group_norm.hpp
│ │ │ │ ├── kernel_base.cpp
│ │ │ │ ├── kernel_base.hpp
│ │ │ │ ├── kernels.hpp
│ │ │ │ ├── large_partition.cpp
│ │ │ │ ├── large_partition.hpp
│ │ │ │ ├── layer_norm.cpp
│ │ │ │ ├── layer_norm.hpp
│ │ │ │ ├── log_softmax.cpp
│ │ │ │ ├── log_softmax.hpp
│ │ │ │ ├── matmul.cpp
│ │ │ │ ├── matmul.hpp
│ │ │ │ ├── mqa.hpp
│ │ │ │ ├── mqa_decomp.cpp
│ │ │ │ ├── mqa_decomp.hpp
│ │ │ │ ├── mqa_decomp_config.cpp
│ │ │ │ ├── mqa_decomp_config.hpp
│ │ │ │ ├── pool.cpp
│ │ │ │ ├── pool.hpp
│ │ │ │ ├── prelu.cpp
│ │ │ │ ├── prelu.hpp
│ │ │ │ ├── quantize.cpp
│ │ │ │ ├── quantize.hpp
│ │ │ │ ├── reduction.cpp
│ │ │ │ ├── reduction.hpp
│ │ │ │ ├── reorder.cpp
│ │ │ │ ├── reorder.hpp
│ │ │ │ ├── resampling.cpp
│ │ │ │ ├── resampling.hpp
│ │ │ │ ├── sdp.hpp
│ │ │ │ ├── sdp_decomp.cpp
│ │ │ │ ├── sdp_decomp.hpp
│ │ │ │ ├── sdp_decomp_config.cpp
│ │ │ │ ├── sdp_decomp_config.hpp
│ │ │ │ ├── sdp_primitive.cpp
│ │ │ │ ├── sdp_primitive.hpp
│ │ │ │ ├── sdp_primitive_config.cpp
│ │ │ │ ├── sdp_primitive_config.hpp
│ │ │ │ ├── sdp_primitive_v1.cpp
│ │ │ │ ├── sdp_primitive_v1.hpp
│ │ │ │ ├── shuffle.cpp
│ │ │ │ ├── shuffle.hpp
│ │ │ │ ├── softmax.cpp
│ │ │ │ ├── softmax.hpp
│ │ │ │ ├── sum.cpp
│ │ │ │ └── sum.hpp
│ │ │ ├── layout_id_mgr.cpp
│ │ │ ├── layout_id_mgr.hpp
│ │ │ ├── layout_propagator.cpp
│ │ │ ├── layout_propagator.hpp
│ │ │ ├── op_executable.cpp
│ │ │ ├── op_executable.hpp
│ │ │ ├── passes
│ │ │ │ ├── compile_ops.cpp
│ │ │ │ ├── compile_ops.hpp
│ │ │ │ ├── constant_propagation.cpp
│ │ │ │ ├── constant_propagation.hpp
│ │ │ │ ├── insert_ops.cpp
│ │ │ │ ├── insert_ops.hpp
│ │ │ │ ├── layout_propagation.cpp
│ │ │ │ ├── layout_propagation.hpp
│ │ │ │ ├── lower.cpp
│ │ │ │ ├── lower.hpp
│ │ │ │ ├── memory_planning.cpp
│ │ │ │ ├── memory_planning.hpp
│ │ │ │ ├── transform.cpp
│ │ │ │ ├── transform.hpp
│ │ │ │ ├── utils.cpp
│ │ │ │ └── utils.hpp
│ │ │ ├── patterns
│ │ │ │ ├── binary_fusion.cpp
│ │ │ │ ├── bn_fusion.cpp
│ │ │ │ ├── concat_fusion.cpp
│ │ │ │ ├── conv_block_fusion.cpp
│ │ │ │ ├── conv_post_ops.cpp
│ │ │ │ ├── convtranspose_fusion.cpp
│ │ │ │ ├── data_type_check_pass.hpp
│ │ │ │ ├── eltwise_fusion.cpp
│ │ │ │ ├── fusions.hpp
│ │ │ │ ├── groupnorm_fusion.cpp
│ │ │ │ ├── interpolate_fusion.cpp
│ │ │ │ ├── layernorm_fusion.cpp
│ │ │ │ ├── matmul_post_ops.cpp
│ │ │ │ ├── mlp.cpp
│ │ │ │ ├── pattern_matcher_pass.hpp
│ │ │ │ ├── pool_post_ops.cpp
│ │ │ │ ├── quantize_fusion.cpp
│ │ │ │ ├── reduction_fusion.cpp
│ │ │ │ ├── reorder_fusion.cpp
│ │ │ │ ├── sdp.cpp
│ │ │ │ ├── shuffle_fusion.cpp
│ │ │ │ ├── single_op_pattern.cpp
│ │ │ │ ├── softmax_post_ops.cpp
│ │ │ │ ├── sum_fusion.cpp
│ │ │ │ └── utils.hpp
│ │ │ ├── platform.cpp
│ │ │ ├── platform.hpp
│ │ │ ├── scratchpad.hpp
│ │ │ ├── subgraph.cpp
│ │ │ ├── subgraph.hpp
│ │ │ ├── thread_local_cache.hpp
│ │ │ └── utils.hpp
│ │ └── fake
│ │ │ ├── CMakeLists.txt
│ │ │ ├── fake_backend.cpp
│ │ │ ├── fake_backend.hpp
│ │ │ ├── fake_partition_impl.hpp
│ │ │ ├── pattern_utils.hpp
│ │ │ ├── single_op_pass.hpp
│ │ │ └── transformation_pass.hpp
│ ├── interface
│ │ ├── CMakeLists.txt
│ │ ├── allocator.cpp
│ │ ├── allocator.hpp
│ │ ├── backend.cpp
│ │ ├── backend.hpp
│ │ ├── c_types_map.hpp
│ │ ├── constant_tensor_cache.cpp
│ │ ├── constant_tensor_cache.hpp
│ │ ├── graph.cpp
│ │ ├── graph.hpp
│ │ ├── graph_attr.hpp
│ │ ├── logical_tensor.cpp
│ │ ├── logical_tensor.hpp
│ │ ├── op.cpp
│ │ ├── op.hpp
│ │ ├── op_def.hpp
│ │ ├── op_def_constraint.cpp
│ │ ├── op_def_constraint.hpp
│ │ ├── op_schema.cpp
│ │ ├── op_schema.hpp
│ │ ├── opset.hpp
│ │ ├── partition.cpp
│ │ ├── partition.hpp
│ │ ├── partition_cache.cpp
│ │ ├── partition_cache.hpp
│ │ ├── partition_hashing.cpp
│ │ ├── partition_hashing.hpp
│ │ ├── partition_impl.cpp
│ │ ├── partition_impl.hpp
│ │ ├── shape_infer.cpp
│ │ ├── shape_infer.hpp
│ │ ├── tensor.cpp
│ │ ├── tensor.hpp
│ │ ├── value.cpp
│ │ └── value.hpp
│ └── utils
│ │ ├── CMakeLists.txt
│ │ ├── alloc.cpp
│ │ ├── alloc.hpp
│ │ ├── any.hpp
│ │ ├── attribute_value.hpp
│ │ ├── debug.cpp
│ │ ├── debug.hpp
│ │ ├── id.cpp
│ │ ├── id.hpp
│ │ ├── json.hpp
│ │ ├── ocl_check.hpp
│ │ ├── ocl_usm_utils.cpp
│ │ ├── ocl_usm_utils.hpp
│ │ ├── pm
│ │ ├── dag_check_pass.hpp
│ │ ├── nested_matcher.cpp
│ │ ├── nested_matcher.hpp
│ │ ├── op_depth_check_pass.hpp
│ │ ├── pass_base.cpp
│ │ ├── pass_base.hpp
│ │ ├── pass_manager.cpp
│ │ ├── pass_manager.hpp
│ │ ├── pbuilder.cpp
│ │ └── pbuilder.hpp
│ │ ├── utils.cpp
│ │ ├── utils.hpp
│ │ ├── verbose.cpp
│ │ └── verbose.hpp
└── xpu
│ ├── CMakeLists.txt
│ ├── README.md
│ ├── context.hpp
│ ├── ocl
│ ├── CMakeLists.txt
│ ├── buffer_memory_storage.cpp
│ ├── buffer_memory_storage.hpp
│ ├── c_types_map.hpp
│ ├── capi
│ │ ├── engine.cpp
│ │ ├── memory.cpp
│ │ ├── primitive.cpp
│ │ └── stream.cpp
│ ├── context.hpp
│ ├── engine_factory.hpp
│ ├── engine_id.hpp
│ ├── engine_impl.cpp
│ ├── engine_impl.hpp
│ ├── memory_storage.hpp
│ ├── memory_storage_base.hpp
│ ├── stream_impl.cpp
│ ├── stream_impl.hpp
│ ├── stream_profiler.cpp
│ ├── stream_profiler.hpp
│ ├── usm_memory_storage.cpp
│ ├── usm_memory_storage.hpp
│ ├── usm_utils.cpp
│ ├── usm_utils.hpp
│ ├── utils.cpp
│ ├── utils.hpp
│ └── verbose.hpp
│ ├── stream_profiler.hpp
│ ├── sycl
│ ├── CMakeLists.txt
│ ├── buffer_memory_storage.cpp
│ ├── buffer_memory_storage.hpp
│ ├── c_types_map.hpp
│ ├── capi
│ │ ├── capi_engine.cpp
│ │ ├── capi_memory.cpp
│ │ ├── capi_primitive.cpp
│ │ └── capi_stream.cpp
│ ├── compat.cpp
│ ├── compat.hpp
│ ├── context.hpp
│ ├── engine_factory.cpp
│ ├── engine_factory.hpp
│ ├── engine_id.hpp
│ ├── engine_impl.cpp
│ ├── engine_impl.hpp
│ ├── memory_storage.hpp
│ ├── memory_storage_base.cpp
│ ├── memory_storage_base.hpp
│ ├── memory_storage_helper.hpp
│ ├── stream_impl.cpp
│ ├── stream_impl.hpp
│ ├── stream_profiler.cpp
│ ├── stream_profiler.hpp
│ ├── types.hpp
│ ├── usm_memory_storage.cpp
│ ├── usm_memory_storage.hpp
│ ├── utils.cpp
│ ├── utils.hpp
│ └── verbose.hpp
│ ├── utils.cpp
│ └── utils.hpp
├── tests
├── CMakeLists.txt
├── api.c
├── benchdnn
│ ├── CMakeLists.txt
│ ├── README.md
│ ├── benchdnn.cpp
│ ├── binary
│ │ ├── bench_binary.cpp
│ │ ├── binary.cpp
│ │ ├── binary.hpp
│ │ ├── binary_aux.cpp
│ │ └── ref_binary.cpp
│ ├── bnorm
│ │ ├── bench_bnorm.cpp
│ │ ├── bnorm.cpp
│ │ ├── bnorm.hpp
│ │ ├── bnorm_aux.cpp
│ │ └── ref_bnorm.cpp
│ ├── brgemm
│ │ ├── bench_brgemm.cpp
│ │ ├── brgemm.cpp
│ │ ├── brgemm.hpp
│ │ ├── brgemm_aux.cpp
│ │ ├── cfg.cpp
│ │ └── ref_brgemm.cpp
│ ├── common.cpp
│ ├── common.hpp
│ ├── concat
│ │ ├── bench_concat.cpp
│ │ ├── concat.cpp
│ │ ├── concat.hpp
│ │ ├── concat_aux.cpp
│ │ └── ref_concat.cpp
│ ├── conv
│ │ ├── bench_conv.cpp
│ │ ├── cfg.cpp
│ │ ├── conv.cpp
│ │ ├── conv.hpp
│ │ ├── conv_aux.cpp
│ │ ├── conv_dw_fusion.cpp
│ │ ├── conv_dw_fusion.hpp
│ │ ├── ref_conv.cpp
│ │ ├── ref_conv.hpp
│ │ └── ref_wino.cpp
│ ├── deconv
│ │ ├── bench_deconv.cpp
│ │ ├── cfg.cpp
│ │ ├── deconv.cpp
│ │ ├── deconv.hpp
│ │ ├── deconv_aux.cpp
│ │ ├── ref_deconv.cpp
│ │ ├── ref_deconv.hpp
│ │ └── ref_wino.cpp
│ ├── dnn_types.cpp
│ ├── dnn_types.hpp
│ ├── dnnl_common.cpp
│ ├── dnnl_common.hpp
│ ├── dnnl_debug.hpp
│ ├── dnnl_debug_autogenerated.cpp
│ ├── dnnl_memory.cpp
│ ├── dnnl_memory.hpp
│ ├── doc
│ │ ├── benchdnn_general_info.md
│ │ ├── benchdnn_input_files_naming_convention.md
│ │ ├── driver_binary.md
│ │ ├── driver_bnorm.md
│ │ ├── driver_brgemm.md
│ │ ├── driver_concat.md
│ │ ├── driver_conv.md
│ │ ├── driver_eltwise.md
│ │ ├── driver_gnorm.md
│ │ ├── driver_graph.md
│ │ ├── driver_ip.md
│ │ ├── driver_lnorm.md
│ │ ├── driver_lrn.md
│ │ ├── driver_matmul.md
│ │ ├── driver_pool.md
│ │ ├── driver_prelu.md
│ │ ├── driver_reduction.md
│ │ ├── driver_reorder.md
│ │ ├── driver_resampling.md
│ │ ├── driver_rnn.md
│ │ ├── driver_shuffle.md
│ │ ├── driver_softmax.md
│ │ ├── driver_sum.md
│ │ ├── driver_zeropad.md
│ │ ├── knob_cold_cache.md
│ │ ├── knob_impl_filter.md
│ │ ├── knob_strides.md
│ │ ├── knob_summary.md
│ │ ├── knob_use_fast_ref.md
│ │ ├── knobs_attr.md
│ │ ├── knobs_common.md
│ │ ├── knobs_desc.md
│ │ ├── knobs_dir.md
│ │ ├── knobs_dt.md
│ │ ├── knobs_encoding.md
│ │ ├── knobs_perf_report.md
│ │ ├── knobs_tag.md
│ │ └── knobs_verbose.md
│ ├── eltwise
│ │ ├── bench_eltwise.cpp
│ │ ├── eltwise.cpp
│ │ ├── eltwise.hpp
│ │ ├── eltwise_aux.cpp
│ │ └── ref_eltwise.cpp
│ ├── gnorm
│ │ ├── bench_gnorm.cpp
│ │ ├── gnorm.cpp
│ │ ├── gnorm.hpp
│ │ ├── gnorm_aux.cpp
│ │ └── ref_gnorm.cpp
│ ├── graph
│ │ ├── allocator.cpp
│ │ ├── allocator.hpp
│ │ ├── bench_graph.cpp
│ │ ├── custom_driver.cpp
│ │ ├── custom_driver.hpp
│ │ ├── deserialize.cpp
│ │ ├── deserialize.hpp
│ │ ├── flex_rewrite.cpp
│ │ ├── flex_rewrite.hpp
│ │ ├── graph.cpp
│ │ ├── graph.hpp
│ │ ├── graph_memory.cpp
│ │ ├── graph_memory.hpp
│ │ ├── input_displacer.cpp
│ │ ├── input_displacer.hpp
│ │ ├── memory_pool.hpp
│ │ ├── parser.cpp
│ │ ├── parser.hpp
│ │ ├── ref_partition.cpp
│ │ ├── ref_partition.hpp
│ │ ├── ref_primitive.cpp
│ │ ├── ref_primitive.hpp
│ │ ├── setting_handler.cpp
│ │ ├── setting_handler.hpp
│ │ ├── utils.cpp
│ │ └── utils.hpp
│ ├── inputs
│ │ ├── binary
│ │ │ ├── harness_binary_bf16
│ │ │ ├── harness_binary_different_dt
│ │ │ ├── harness_binary_f16
│ │ │ ├── harness_binary_f32
│ │ │ ├── harness_binary_i8
│ │ │ ├── harness_binary_regression
│ │ │ ├── option_set_all
│ │ │ ├── option_set_fwks_ext_gpu
│ │ │ ├── option_set_fwks_key_gpu
│ │ │ ├── option_set_minimal
│ │ │ ├── option_set_src0_bcast
│ │ │ ├── perf_binary_gpu
│ │ │ ├── shapes_ci
│ │ │ ├── shapes_perf_1st_conv
│ │ │ ├── shapes_perf_scaleshift
│ │ │ ├── test_binary_all
│ │ │ ├── test_binary_bfloat16
│ │ │ ├── test_binary_ci
│ │ │ ├── test_binary_different_dt_ci
│ │ │ ├── test_binary_float16
│ │ │ ├── test_binary_gpu
│ │ │ └── test_binary_smoke
│ │ ├── bnorm
│ │ │ ├── option_set_fwks_ext_gpu
│ │ │ ├── option_set_fwks_key_gpu
│ │ │ ├── perf_bnorm_gpu
│ │ │ ├── set_nd
│ │ │ ├── shapes_1d
│ │ │ ├── shapes_2d
│ │ │ ├── shapes_3d
│ │ │ ├── shapes_ci
│ │ │ ├── shapes_densenet_121
│ │ │ ├── shapes_googlenet_v2
│ │ │ ├── shapes_googlenet_v3
│ │ │ ├── shapes_large
│ │ │ ├── shapes_regressions
│ │ │ ├── shapes_resnet_50
│ │ │ ├── shapes_topologies_small
│ │ │ ├── test_bnorm_all_blocked
│ │ │ ├── test_bnorm_all_plain
│ │ │ ├── test_bnorm_bfloat16_blocked
│ │ │ ├── test_bnorm_bfloat16_plain
│ │ │ ├── test_bnorm_ci
│ │ │ ├── test_bnorm_float16_plain
│ │ │ ├── test_bnorm_gpu
│ │ │ ├── test_bnorm_regressions
│ │ │ ├── test_bnorm_regressions_large
│ │ │ └── test_bnorm_smoke
│ │ ├── brgemm
│ │ │ ├── harness_brgemm_f32
│ │ │ ├── harness_brgemm_f8
│ │ │ ├── harness_brgemm_fpmath
│ │ │ ├── option_set_bf16
│ │ │ ├── option_set_f32
│ │ │ ├── option_set_int8
│ │ │ ├── shapes_2d_big_k_bf16
│ │ │ ├── shapes_2d_big_k_f32
│ │ │ ├── shapes_2d_big_k_int8
│ │ │ ├── shapes_2d_big_k_tail_n_bf16
│ │ │ ├── shapes_2d_big_k_tail_n_f32
│ │ │ ├── shapes_2d_big_k_tail_n_int8
│ │ │ ├── shapes_2d_no_tail_bf16
│ │ │ ├── shapes_2d_no_tail_f32
│ │ │ ├── shapes_2d_no_tail_int8
│ │ │ ├── shapes_2d_tail_k_bf16
│ │ │ ├── shapes_2d_tail_k_f32
│ │ │ ├── shapes_2d_tail_k_int8
│ │ │ ├── shapes_2d_tail_k_tail_n_bf16
│ │ │ ├── shapes_2d_tail_k_tail_n_f32
│ │ │ ├── shapes_2d_tail_k_tail_n_int8
│ │ │ ├── shapes_2d_tail_n_bf16
│ │ │ ├── shapes_2d_tail_n_f32
│ │ │ ├── shapes_2d_tail_n_int8
│ │ │ ├── test_brgemm_all
│ │ │ ├── test_brgemm_bf16
│ │ │ ├── test_brgemm_ci
│ │ │ ├── test_brgemm_f16
│ │ │ ├── test_brgemm_f32
│ │ │ ├── test_brgemm_f8
│ │ │ ├── test_brgemm_int8
│ │ │ ├── test_brgemm_regression
│ │ │ └── test_brgemm_smoke
│ │ ├── concat
│ │ │ ├── option_set_fwks_ext_gpu
│ │ │ ├── option_set_fwks_key_gpu
│ │ │ ├── option_set_xe_gpu
│ │ │ ├── test_concat_all
│ │ │ ├── test_concat_bfloat16
│ │ │ ├── test_concat_ci
│ │ │ ├── test_concat_float16
│ │ │ ├── test_concat_gpu
│ │ │ ├── test_concat_large_gpu
│ │ │ └── test_concat_smoke
│ │ ├── conv
│ │ │ ├── harness_conv_arbitrary_dst
│ │ │ ├── harness_conv_attrs_gpu
│ │ │ ├── harness_conv_attrs_int8
│ │ │ ├── harness_conv_attrs_int8_asymmetric
│ │ │ ├── harness_conv_auto
│ │ │ ├── harness_conv_deepbench
│ │ │ ├── harness_conv_depthwise_int8
│ │ │ ├── harness_conv_dilated_3d
│ │ │ ├── harness_conv_dilated_int8
│ │ │ ├── harness_conv_dw_bfloat16
│ │ │ ├── harness_conv_dw_bfloat16_nxc
│ │ │ ├── harness_conv_dw_float16_nxc
│ │ │ ├── harness_conv_dw_fp8_nxc
│ │ │ ├── harness_conv_f32
│ │ │ ├── harness_conv_f32_plain
│ │ │ ├── harness_conv_fused_depthwise
│ │ │ ├── harness_conv_int8
│ │ │ ├── harness_conv_output_striding
│ │ │ ├── harness_conv_regression_general
│ │ │ ├── harness_conv_saturation_int8
│ │ │ ├── harness_conv_smoke_ref
│ │ │ ├── harness_conv_tags
│ │ │ ├── harness_conv_zero_points
│ │ │ ├── option_gpu_ci
│ │ │ ├── option_set_all_eltwise_postops
│ │ │ ├── option_set_combined_postops
│ │ │ ├── option_set_fwks_ext_gpu
│ │ │ ├── option_set_fwks_key_gpu
│ │ │ ├── option_set_v2
│ │ │ ├── perf_conv_bdw_1sock
│ │ │ ├── perf_conv_clx_1sock
│ │ │ ├── perf_conv_skx_1sock
│ │ │ ├── perf_conv_xe
│ │ │ ├── perf_conv_xe_hp
│ │ │ ├── perf_conv_xe_lp
│ │ │ ├── set_all_topologies
│ │ │ ├── set_conv_3d
│ │ │ ├── set_conv_all
│ │ │ ├── set_conv_dw
│ │ │ ├── set_dilated-conv
│ │ │ ├── set_dilated-conv_1st
│ │ │ ├── set_dilated-conv_3d
│ │ │ ├── set_fastrcnn
│ │ │ ├── set_gpu
│ │ │ ├── set_maskrcnn
│ │ │ ├── set_perf_cpu_all_mb
│ │ │ ├── set_perf_cpu_inference_only
│ │ │ ├── set_perf_cpu_large_mb
│ │ │ ├── set_perf_cpu_small_mb
│ │ │ ├── set_perf_gpu_all_mb
│ │ │ ├── set_perf_gpu_large_mb
│ │ │ ├── set_perf_gpu_small_mb
│ │ │ ├── set_topologies_inference_only
│ │ │ ├── shapes_1d
│ │ │ ├── shapes_1d_wavenet
│ │ │ ├── shapes_1x1
│ │ │ ├── shapes_3d
│ │ │ ├── shapes_3d_1st_strided_padding
│ │ │ ├── shapes_3d_1x1_strided_no-padding
│ │ │ ├── shapes_3d_1x1_strided_padding
│ │ │ ├── shapes_3d_1x1_unit-stride_no-padding
│ │ │ ├── shapes_3d_1x1_unit-stride_padding
│ │ │ ├── shapes_3d_2d_strided_padding
│ │ │ ├── shapes_3d_gpu
│ │ │ ├── shapes_3d_i3d
│ │ │ ├── shapes_3d_resnext101
│ │ │ ├── shapes_3d_strided_no-padding
│ │ │ ├── shapes_3d_strided_padding
│ │ │ ├── shapes_3d_unet
│ │ │ ├── shapes_3d_unit-stride_no-padding
│ │ │ ├── shapes_3d_unit-stride_padding
│ │ │ ├── shapes_4bit
│ │ │ ├── shapes_a3c
│ │ │ ├── shapes_alexnet
│ │ │ ├── shapes_auto
│ │ │ ├── shapes_basic
│ │ │ ├── shapes_ci_gpu
│ │ │ ├── shapes_cosmictagger
│ │ │ ├── shapes_deepbench_inference_device
│ │ │ ├── shapes_deepbench_inference_server
│ │ │ ├── shapes_deepbench_training
│ │ │ ├── shapes_densnet
│ │ │ ├── shapes_dilated
│ │ │ ├── shapes_dilated_1d_1st_strided_padding
│ │ │ ├── shapes_dilated_1d_strided_no-padding
│ │ │ ├── shapes_dilated_1d_strided_padding
│ │ │ ├── shapes_dilated_1d_unit-stride_no-padding
│ │ │ ├── shapes_dilated_1d_unit-stride_padding
│ │ │ ├── shapes_dilated_2d_1st_strided_padding
│ │ │ ├── shapes_dilated_2d_strided_no-padding
│ │ │ ├── shapes_dilated_2d_strided_padding
│ │ │ ├── shapes_dilated_2d_unit-stride_no-padding
│ │ │ ├── shapes_dilated_2d_unit-stride_padding
│ │ │ ├── shapes_dilated_3d_strided_no-padding
│ │ │ ├── shapes_dilated_3d_strided_padding
│ │ │ ├── shapes_dilated_3d_unit-stride_no-padding
│ │ │ ├── shapes_dilated_3d_unit-stride_padding
│ │ │ ├── shapes_dilated_rfcn
│ │ │ ├── shapes_dw_1d_stride_no-padding
│ │ │ ├── shapes_dw_1d_unit-stride_no-padding
│ │ │ ├── shapes_dw_1d_unit-stride_padding
│ │ │ ├── shapes_dw_2d_1d_strided_padding
│ │ │ ├── shapes_dw_2d_strided_no-padding
│ │ │ ├── shapes_dw_2d_strided_padding
│ │ │ ├── shapes_dw_2d_unit-stride_no-padding
│ │ │ ├── shapes_dw_2d_unit-stride_padding
│ │ │ ├── shapes_dw_3d_strided_no-padding
│ │ │ ├── shapes_dw_3d_strided_padding
│ │ │ ├── shapes_dw_3d_unit-stride_no-padding
│ │ │ ├── shapes_dw_3d_unit-stride_padding
│ │ │ ├── shapes_dw_minibatch_2d-spatial
│ │ │ ├── shapes_dw_minibatch_channel_2d-spatial
│ │ │ ├── shapes_efficientdet
│ │ │ ├── shapes_fastrcnn_p1
│ │ │ ├── shapes_fastrcnn_p2
│ │ │ ├── shapes_fastrcnn_p3
│ │ │ ├── shapes_ffn
│ │ │ ├── shapes_fused_large_src
│ │ │ ├── shapes_fused_mobilenet_stride_1
│ │ │ ├── shapes_fused_mobilenet_stride_2
│ │ │ ├── shapes_gemm
│ │ │ ├── shapes_googlenet_v1
│ │ │ ├── shapes_googlenet_v2
│ │ │ ├── shapes_googlenet_v3
│ │ │ ├── shapes_large_conv
│ │ │ ├── shapes_large_padding
│ │ │ ├── shapes_maskrcnn_p1
│ │ │ ├── shapes_maskrcnn_p2
│ │ │ ├── shapes_mem_strided
│ │ │ ├── shapes_mobilenet
│ │ │ ├── shapes_mobilenet_dw
│ │ │ ├── shapes_movinet_dw
│ │ │ ├── shapes_pointnet
│ │ │ ├── shapes_regression_1x1
│ │ │ ├── shapes_regression_dw
│ │ │ ├── shapes_regression_gemm
│ │ │ ├── shapes_regression_padding
│ │ │ ├── shapes_regression_small_spatial
│ │ │ ├── shapes_resnet_50
│ │ │ ├── shapes_resnet_50_sparse
│ │ │ ├── shapes_resnet_50_v1_5
│ │ │ ├── shapes_resnext_101
│ │ │ ├── shapes_segnet
│ │ │ ├── shapes_src-transpose_padding
│ │ │ ├── shapes_ssd_300_voc0712
│ │ │ ├── shapes_ssd_mobilenet
│ │ │ ├── shapes_ssd_resnet34_inference
│ │ │ ├── shapes_ssd_resnet34_training
│ │ │ ├── shapes_tails
│ │ │ ├── shapes_tails_gpu
│ │ │ ├── shapes_unet
│ │ │ ├── shapes_vgg_11
│ │ │ ├── shapes_vgg_19
│ │ │ ├── shapes_x3d_dw
│ │ │ ├── shapes_xception
│ │ │ ├── shapes_yolov2
│ │ │ ├── test_conv_3d
│ │ │ ├── test_conv_3d_f32_plain
│ │ │ ├── test_conv_all
│ │ │ ├── test_conv_all_topologies
│ │ │ ├── test_conv_all_topologies_f32_plain
│ │ │ ├── test_conv_attrs
│ │ │ ├── test_conv_attrs_f32_plain
│ │ │ ├── test_conv_bfloat16
│ │ │ ├── test_conv_bfloat16_nxc
│ │ │ ├── test_conv_bfloat16_ymm
│ │ │ ├── test_conv_ci
│ │ │ ├── test_conv_depthwise
│ │ │ ├── test_conv_dilated
│ │ │ ├── test_conv_dilated_f32_plain
│ │ │ ├── test_conv_dt
│ │ │ ├── test_conv_dt_plain
│ │ │ ├── test_conv_float16_nxc
│ │ │ ├── test_conv_fp4
│ │ │ ├── test_conv_fp8_nxc
│ │ │ ├── test_conv_function
│ │ │ ├── test_conv_gemm_bfloat16
│ │ │ ├── test_conv_gemm_bfloat16_nxc
│ │ │ ├── test_conv_gemm_dt
│ │ │ ├── test_conv_gemm_dt_nxc
│ │ │ ├── test_conv_gemm_int8
│ │ │ ├── test_conv_gpu
│ │ │ ├── test_conv_gpu_ci
│ │ │ ├── test_conv_int8
│ │ │ ├── test_conv_large_gpu
│ │ │ ├── test_conv_regression
│ │ │ ├── test_conv_regression_gpu
│ │ │ ├── test_conv_smoke
│ │ │ ├── test_conv_wino_f32
│ │ │ └── test_conv_wino_gpu
│ │ ├── deconv
│ │ │ ├── harness_deconv_attrs_int8
│ │ │ ├── harness_deconv_attrs_int8_asymmetric
│ │ │ ├── harness_deconv_regression_general_f32
│ │ │ ├── harness_deconv_regression_general_int8
│ │ │ ├── option_set_fwks_ext_gpu
│ │ │ ├── option_set_fwks_key_gpu
│ │ │ ├── set_all
│ │ │ ├── shapes_1d
│ │ │ ├── shapes_1x1
│ │ │ ├── shapes_2d
│ │ │ ├── shapes_3d
│ │ │ ├── shapes_ci
│ │ │ ├── shapes_dilated
│ │ │ ├── test_deconv_all
│ │ │ ├── test_deconv_all_f32_nxc
│ │ │ ├── test_deconv_bfloat16
│ │ │ ├── test_deconv_bfloat16_nxc
│ │ │ ├── test_deconv_bfloat16_ymm
│ │ │ ├── test_deconv_ci
│ │ │ ├── test_deconv_float16_nxc
│ │ │ ├── test_deconv_fp8_nxc
│ │ │ ├── test_deconv_gpu
│ │ │ ├── test_deconv_int8
│ │ │ └── test_deconv_smoke
│ │ ├── eltwise
│ │ │ ├── harness_eltwise_large_buffer
│ │ │ ├── harness_eltwise_regression
│ │ │ ├── harness_eltwise_saturation
│ │ │ ├── option_set_all_algs
│ │ │ ├── option_set_all_algs_ci
│ │ │ ├── option_set_all_algs_int8
│ │ │ ├── option_set_all_algs_int8_ci
│ │ │ ├── option_set_fwks_ext_gpu
│ │ │ ├── option_set_fwks_key_gpu
│ │ │ ├── shapes_ci
│ │ │ ├── shapes_eltwise
│ │ │ ├── shapes_large_buffer
│ │ │ ├── test_eltwise_all
│ │ │ ├── test_eltwise_bfloat16
│ │ │ ├── test_eltwise_ci
│ │ │ ├── test_eltwise_float16
│ │ │ ├── test_eltwise_float8
│ │ │ ├── test_eltwise_gpu
│ │ │ └── test_eltwise_smoke
│ │ ├── gnorm
│ │ │ ├── shapes_all
│ │ │ ├── shapes_ci
│ │ │ ├── shapes_sd
│ │ │ ├── test_gnorm_all
│ │ │ └── test_gnorm_ci
│ │ ├── graph
│ │ │ ├── complex_fusion
│ │ │ │ ├── harness_mha_all
│ │ │ │ ├── harness_mha_ci
│ │ │ │ ├── harness_mlp_all
│ │ │ │ ├── harness_mlp_ci
│ │ │ │ ├── mha
│ │ │ │ │ ├── GQA-fp16-v2.json
│ │ │ │ │ ├── GQA-fp16.json
│ │ │ │ │ ├── JAX-MHA-inf-fp32.json
│ │ │ │ │ ├── JAX-MQA-inf-fp32.json
│ │ │ │ │ ├── MHA-GPT-inf-fp32-bs1.json
│ │ │ │ │ ├── MHA-GPT-inf-int8-bs1.json
│ │ │ │ │ ├── MHA-bert_large-inf-fp32-bs1.json
│ │ │ │ │ ├── MHA-bert_large-inf-int8-bs1.json
│ │ │ │ │ ├── MHA-distill_bert-inf-fp32-bs1.json
│ │ │ │ │ ├── MHA-distill_bert-inf-int8-bs1.json
│ │ │ │ │ ├── MHA-stable_diffusion-inf-fp32-bs1.json
│ │ │ │ │ ├── codegemma-bf16-f32.json
│ │ │ │ │ ├── gemma2-bf16-f32.json
│ │ │ │ │ ├── gqa-plain-bottom-right-implicit-causal-mask-f16-f32.json
│ │ │ │ │ ├── gqa-plain-implicit-causal-mask-fp32-bs1.json
│ │ │ │ │ ├── sdpa-compressed-k-int8-gs32.json
│ │ │ │ │ ├── sdpa-compressed-kv-implicit-causal-mask-int8-gs128.json
│ │ │ │ │ ├── sdpa-compressed-kv-int4-gs32.json
│ │ │ │ │ ├── sdpa-compressed-v-int8-gs32.json
│ │ │ │ │ ├── sdpa-plain-bottom-right-implicit-causal-mask-f16-f32.json
│ │ │ │ │ ├── sdpa-plain-implicit-causal-mask-fp32-bs1.json
│ │ │ │ │ ├── sdpa-plain-simplified-f16-f32.json
│ │ │ │ │ ├── sdpa-plain-simplified-f16.json
│ │ │ │ │ ├── sdpa-plain-wo-mask-f16.json
│ │ │ │ │ ├── sdpa-plain-wo-scale-f16-bs1.json
│ │ │ │ │ └── sdpa-plain-wo-scale-int8-bs1.json
│ │ │ │ └── mlp
│ │ │ │ │ ├── gated-mlp-f32.json
│ │ │ │ │ └── gated-mlp-int4.json
│ │ │ ├── op
│ │ │ │ ├── bf16
│ │ │ │ │ ├── dynamicdq_s4.json
│ │ │ │ │ ├── dynamicdq_u4.json
│ │ │ │ │ └── typecast.json
│ │ │ │ ├── f16
│ │ │ │ │ ├── dynamicdq_s4.json
│ │ │ │ │ ├── dynamicdq_u4.json
│ │ │ │ │ └── typecast.json
│ │ │ │ ├── f32
│ │ │ │ │ ├── abs.json
│ │ │ │ │ ├── abs_bwd.json
│ │ │ │ │ ├── add.json
│ │ │ │ │ ├── avgpool.json
│ │ │ │ │ ├── avgpool_bwd.json
│ │ │ │ │ ├── biasadd.json
│ │ │ │ │ ├── biasadd_bwd.json
│ │ │ │ │ ├── bnorm.json
│ │ │ │ │ ├── bnorm_bwd.json
│ │ │ │ │ ├── bnorm_fwd_d.json
│ │ │ │ │ ├── clamp.json
│ │ │ │ │ ├── clamp_bwd.json
│ │ │ │ │ ├── concat.json
│ │ │ │ │ ├── concat_2.json
│ │ │ │ │ ├── concat_3.json
│ │ │ │ │ ├── conv_2d.json
│ │ │ │ │ ├── conv_3d.json
│ │ │ │ │ ├── conv_bwd_d_2d.json
│ │ │ │ │ ├── conv_bwd_d_3d.json
│ │ │ │ │ ├── conv_bwd_w_2d.json
│ │ │ │ │ ├── deconv.json
│ │ │ │ │ ├── deconv_bwd_d.json
│ │ │ │ │ ├── deconv_bwd_w.json
│ │ │ │ │ ├── dequantize_f8_e4m3.json
│ │ │ │ │ ├── dequantize_f8_e5m2.json
│ │ │ │ │ ├── dequantize_s8.json
│ │ │ │ │ ├── dequantize_u8.json
│ │ │ │ │ ├── dynamicdq.json
│ │ │ │ │ ├── dynamicdq_s4.json
│ │ │ │ │ ├── dynamicdq_u4.json
│ │ │ │ │ ├── dynamicq.json
│ │ │ │ │ ├── dynamicq_s4.json
│ │ │ │ │ ├── elu.json
│ │ │ │ │ ├── elu_bwd.json
│ │ │ │ │ ├── gelu.json
│ │ │ │ │ ├── genindex.json
│ │ │ │ │ ├── gnorm.json
│ │ │ │ │ ├── greaterequal.json
│ │ │ │ │ ├── hardsigmoid.json
│ │ │ │ │ ├── hardsigmoid_bwd.json
│ │ │ │ │ ├── interpolate.json
│ │ │ │ │ ├── interpolate_3d.json
│ │ │ │ │ ├── interpolate_bwd.json
│ │ │ │ │ ├── interpolate_bwd_2d.json
│ │ │ │ │ ├── lnorm.json
│ │ │ │ │ ├── lnorm_3d.json
│ │ │ │ │ ├── lnorm_3d_bwd.json
│ │ │ │ │ ├── lnorm_bwd.json
│ │ │ │ │ ├── lnorm_ks.json
│ │ │ │ │ ├── logsoftmax.json
│ │ │ │ │ ├── logsoftmax_bwd.json
│ │ │ │ │ ├── matmul_2d_4d.json
│ │ │ │ │ ├── maxpool.json
│ │ │ │ │ ├── maxpool_bwd.json
│ │ │ │ │ ├── prelu.json
│ │ │ │ │ ├── prelu_bwd.json
│ │ │ │ │ ├── prelu_bwd_dw_5d.json
│ │ │ │ │ ├── quantize.json
│ │ │ │ │ ├── quantize_f8_e4m3.json
│ │ │ │ │ ├── quantize_f8_e5m2.json
│ │ │ │ │ ├── reciprocal.json
│ │ │ │ │ ├── reducel1.json
│ │ │ │ │ ├── reducel2.json
│ │ │ │ │ ├── reducemax.json
│ │ │ │ │ ├── reducemean.json
│ │ │ │ │ ├── reducemin.json
│ │ │ │ │ ├── reduceprod.json
│ │ │ │ │ ├── reducesum.json
│ │ │ │ │ ├── relu.json
│ │ │ │ │ ├── relu_bwd.json
│ │ │ │ │ ├── reorder.json
│ │ │ │ │ ├── select.json
│ │ │ │ │ ├── softmax.json
│ │ │ │ │ ├── softmax_bwd.json
│ │ │ │ │ ├── softmax_bwd_d_3d.json
│ │ │ │ │ ├── softplus_bwd.json
│ │ │ │ │ ├── static_reshape.json
│ │ │ │ │ ├── static_transpose.json
│ │ │ │ │ └── typecast.json
│ │ │ │ ├── harness_bf16_all
│ │ │ │ ├── harness_bf16_ci
│ │ │ │ ├── harness_f16_all
│ │ │ │ ├── harness_f16_ci
│ │ │ │ ├── harness_f32_all
│ │ │ │ └── harness_f32_ci
│ │ │ ├── pattern
│ │ │ │ ├── f32
│ │ │ │ │ ├── avgpool_3d_chain_fusion.json
│ │ │ │ │ ├── binary_2d_post_ops_relu_fusion.json
│ │ │ │ │ ├── binary_2d_post_ops_sum_fusion.json
│ │ │ │ │ ├── binary_3d_post_ops_add_fusion.json
│ │ │ │ │ ├── binary_4d_post_ops_relu_fusion.json
│ │ │ │ │ ├── binary_4d_post_ops_sum_fusion.json
│ │ │ │ │ ├── binary_post_ops_chain_fusion.json
│ │ │ │ │ ├── binary_post_ops_fusion.json
│ │ │ │ │ ├── binary_post_ops_logistic_fusion.json
│ │ │ │ │ ├── bn_bwd_relu_bwd_fusion.json
│ │ │ │ │ ├── bn_relu_fusion.json
│ │ │ │ │ ├── conv_add_sigmoid_multiply_relu_fusion.json
│ │ │ │ │ ├── conv_add_swish_relu_fusion.json
│ │ │ │ │ ├── conv_bias_add_fusion.json
│ │ │ │ │ ├── conv_bias_mul_mul_depthwise_bias_swish_fusion_cpu.json
│ │ │ │ │ ├── conv_bias_post_ops_chain_fusion.json
│ │ │ │ │ ├── conv_bias_post_ops_fusion.json
│ │ │ │ │ ├── conv_bias_relu_depthwise_bias_relu_fusion_cpu.json
│ │ │ │ │ ├── conv_bias_sum_fusion.json
│ │ │ │ │ ├── conv_bias_sum_fusion_2.json
│ │ │ │ │ ├── conv_bias_swish_fusion.json
│ │ │ │ │ ├── conv_depthwise_fusion_cpu.json
│ │ │ │ │ ├── conv_post_ops_fusion.json
│ │ │ │ │ ├── convtranspose_post_ops_fusion.json
│ │ │ │ │ ├── interpolate_post_ops_chain_fusion.json
│ │ │ │ │ ├── interpolate_post_ops_chain_fusion_2.json
│ │ │ │ │ ├── interpolate_post_ops_chain_fusion_3.json
│ │ │ │ │ ├── interpolate_post_ops_chain_fusion_4.json
│ │ │ │ │ ├── lnorm_gelu.json
│ │ │ │ │ ├── matmul_bias_post_ops_chain_fusion.json
│ │ │ │ │ ├── matmul_bias_post_ops_clip_fusion.json
│ │ │ │ │ ├── matmul_bias_post_ops_elu_fusion.json
│ │ │ │ │ ├── matmul_post_ops_add_add_fusion.json
│ │ │ │ │ ├── matmul_post_ops_chain_fusion.json
│ │ │ │ │ ├── matmul_post_ops_clip_fusion.json
│ │ │ │ │ ├── matmul_post_ops_relu_add_fusion.json
│ │ │ │ │ ├── matmul_post_ops_sum_logistic_fusion.json
│ │ │ │ │ ├── matmul_post_ops_sum_relu_fusion.json
│ │ │ │ │ ├── matmul_post_ops_swish_fusion.json
│ │ │ │ │ ├── matmul_select.json
│ │ │ │ │ ├── maxpool_chain_fusion.json
│ │ │ │ │ ├── maxpool_sum_relu_fusion.json
│ │ │ │ │ ├── reciprocal_multiply_fusion.json
│ │ │ │ │ ├── reduction_post_ops_l1_chain_fusion.json
│ │ │ │ │ ├── reduction_post_ops_l2_fusion.json
│ │ │ │ │ ├── reduction_post_ops_max_chain_fusion.json
│ │ │ │ │ ├── reduction_post_ops_mean_fusion.json
│ │ │ │ │ ├── reduction_post_ops_min_chain_fusion.json
│ │ │ │ │ ├── reduction_post_ops_prod_chain_fusion.json
│ │ │ │ │ ├── reduction_post_ops_sum_chain_fusion.json
│ │ │ │ │ ├── shuffle_fusion.json
│ │ │ │ │ ├── softmax_post_ops_binary_fusion.json
│ │ │ │ │ ├── softmax_post_ops_unary_fusion.json
│ │ │ │ │ ├── unary_post_ops_elu_fusion.json
│ │ │ │ │ ├── unary_post_ops_gelu_fusion.json
│ │ │ │ │ ├── unary_post_ops_hardswish_fusion.json
│ │ │ │ │ ├── unary_post_ops_log_fusion.json
│ │ │ │ │ ├── unary_post_ops_round_fusion.json
│ │ │ │ │ ├── unary_post_ops_sqrt_fusion.json
│ │ │ │ │ ├── unary_post_ops_square_fusion.json
│ │ │ │ │ └── unary_post_ops_tanh_fusion.json
│ │ │ │ ├── f8
│ │ │ │ │ ├── f8_bf16_matmul_add_fusion.json
│ │ │ │ │ ├── f8_bf16_matmul_sum_add_mul_relu.json
│ │ │ │ │ ├── f8_conv_add_add_fusion.json
│ │ │ │ │ ├── f8_conv_bias_relu_fusion.json
│ │ │ │ │ ├── f8_conv_fwd.json
│ │ │ │ │ ├── f8_conv_post_ops_fusion.json
│ │ │ │ │ ├── f8_conv_post_ops_int8_add_fusion.json
│ │ │ │ │ ├── f8_f32_matmul_mul_add_fusion.json
│ │ │ │ │ ├── f8_matmul.json
│ │ │ │ │ └── f8_matmul_sum_add_mul_relu.json
│ │ │ │ ├── harness_bf16_all
│ │ │ │ ├── harness_bf16_ci
│ │ │ │ ├── harness_f16_all
│ │ │ │ ├── harness_f16_ci
│ │ │ │ ├── harness_f32_all
│ │ │ │ ├── harness_f32_ci
│ │ │ │ ├── harness_f8_all
│ │ │ │ ├── harness_f8_ci
│ │ │ │ ├── harness_int8_all
│ │ │ │ ├── harness_int8_ci
│ │ │ │ └── int8
│ │ │ │ │ ├── int8_avgpool_reshape_fusion.json
│ │ │ │ │ ├── int8_avgpool_transpose_fusion.json
│ │ │ │ │ ├── int8_bf16_conv_add_fusion.json
│ │ │ │ │ ├── int8_bf16_conv_add_relu_mul.json
│ │ │ │ │ ├── int8_bf16_conv_binary_add_fusion.json
│ │ │ │ │ ├── int8_bf16_conv_binary_add_fusion_2.json
│ │ │ │ │ ├── int8_bf16_gnorm_add_fusion.json
│ │ │ │ │ ├── int8_bf16_gnorm_relu_fusion.json
│ │ │ │ │ ├── int8_bf16_matmul.json
│ │ │ │ │ ├── int8_bf16_matmul_add_fusion.json
│ │ │ │ │ ├── int8_bf16_matmul_add_mul_relu.json
│ │ │ │ │ ├── int8_bf16_matmul_mul_add_fusion.json
│ │ │ │ │ ├── int8_bf16_matmul_mul_add_fusion_2.json
│ │ │ │ │ ├── int8_bf16_matmul_mul_w_smooth_quant_fusion.json
│ │ │ │ │ ├── int8_bf16_matmul_post_ops_fusion.json
│ │ │ │ │ ├── int8_bf16_matmul_relu_w_smooth_quant_fusion.json
│ │ │ │ │ ├── int8_bf16_matmul_sum_add_mul_relu.json
│ │ │ │ │ ├── int8_bf16_matmul_tc_add_fusion.json
│ │ │ │ │ ├── int8_bf16_matmul_tc_add_quant_fusion.json
│ │ │ │ │ ├── int8_bnorm_relu_fusion.json
│ │ │ │ │ ├── int8_concat_fusion.json
│ │ │ │ │ ├── int8_concat_fusion_3.json
│ │ │ │ │ ├── int8_conv_2d_fusion.json
│ │ │ │ │ ├── int8_conv_2d_fusion_2.json
│ │ │ │ │ ├── int8_conv_2d_fwd_i_fusion.json
│ │ │ │ │ ├── int8_conv_add_add_fusion.json
│ │ │ │ │ ├── int8_conv_add_mul_fusion.json
│ │ │ │ │ ├── int8_conv_bias_fusion.json
│ │ │ │ │ ├── int8_conv_bias_mish_fusion.json
│ │ │ │ │ ├── int8_conv_bias_relu_fusion.json
│ │ │ │ │ ├── int8_conv_bias_relu_fusion_2.json
│ │ │ │ │ ├── int8_conv_bias_relu_fusion_3.json
│ │ │ │ │ ├── int8_conv_post_ops_fusion.json
│ │ │ │ │ ├── int8_conv_post_ops_int8_add_fusion.json
│ │ │ │ │ ├── int8_conv_relu_fusion.json
│ │ │ │ │ ├── int8_convtranspose_post_ops_add_fusion.json
│ │ │ │ │ ├── int8_convtranspose_post_ops_chain_fusion.json
│ │ │ │ │ ├── int8_convtranspose_post_ops_fusion.json
│ │ │ │ │ ├── int8_convtranspose_post_ops_square_fusion.json
│ │ │ │ │ ├── int8_convtranspose_post_ops_sum_fusion.json
│ │ │ │ │ ├── int8_convtranspose_post_ops_sum_fusion_2.json
│ │ │ │ │ ├── int8_depthwise_conv_fusion.json
│ │ │ │ │ ├── int8_f32_matmul_mul_add_fusion.json
│ │ │ │ │ ├── int8_f32_matmul_mul_add_fusion_2.json
│ │ │ │ │ ├── int8_lnorm_gelu_quantize.json
│ │ │ │ │ ├── int8_lnorm_multiply_quantize.json
│ │ │ │ │ ├── int8_lnorm_tc_multiply_quantize.json
│ │ │ │ │ ├── int8_matmul_add_mul_fusion.json
│ │ │ │ │ ├── int8_matmul_add_mul_relu.json
│ │ │ │ │ ├── int8_matmul_bia_relu_fusion.json
│ │ │ │ │ ├── int8_matmul_bias_sum_fusion.json
│ │ │ │ │ ├── int8_matmul_logistic_fusion.json
│ │ │ │ │ ├── int8_matmul_mul_add_mul_fusion.json
│ │ │ │ │ ├── int8_matmul_post_ops_fusion.json
│ │ │ │ │ ├── int8_matmul_sum_add_mul_relu.json
│ │ │ │ │ ├── int8_maxpool_add_mul_fusion.json
│ │ │ │ │ ├── int8_reorder_fusion.json
│ │ │ │ │ ├── int8_reorder_fusion_2.json
│ │ │ │ │ ├── int8_reorder_fusion_3.json
│ │ │ │ │ └── int8_softmax_add.json
│ │ │ ├── test_graph_all
│ │ │ ├── test_graph_bf16
│ │ │ ├── test_graph_ci
│ │ │ ├── test_graph_f16
│ │ │ ├── test_graph_f32
│ │ │ ├── test_graph_f8
│ │ │ ├── test_graph_fusions
│ │ │ ├── test_graph_fusions_gpu
│ │ │ ├── test_graph_int8
│ │ │ ├── test_graph_op_gpu
│ │ │ └── test_graph_pattern_gpu
│ │ ├── ip
│ │ │ ├── harness_ip_gpt-j_2016-32_inf_lb_bfloat16
│ │ │ ├── harness_ip_gpt-j_2016-32_inf_lb_f32
│ │ │ ├── harness_ip_gpt-j_2016-32_inf_lb_float16
│ │ │ ├── harness_ip_gpt-j_2016-32_inf_sb_bfloat16
│ │ │ ├── harness_ip_gpt-j_2016-32_inf_sb_f32
│ │ │ ├── harness_ip_gpt-j_2016-32_inf_sb_float16
│ │ │ ├── harness_ip_gpt-j_32-32_inf_lb_bfloat16
│ │ │ ├── harness_ip_gpt-j_32-32_inf_lb_f32
│ │ │ ├── harness_ip_gpt-j_32-32_inf_lb_float16
│ │ │ ├── harness_ip_gpt-j_32-32_inf_sb_bfloat16
│ │ │ ├── harness_ip_gpt-j_32-32_inf_sb_f32
│ │ │ ├── harness_ip_gpt-j_32-32_inf_sb_float16
│ │ │ ├── harness_ip_regression
│ │ │ ├── harness_ip_sanitizers
│ │ │ ├── harness_ip_saturation
│ │ │ ├── harness_ip_smoke_ref
│ │ │ ├── harness_ip_tag
│ │ │ ├── harness_ip_tag_gpu
│ │ │ ├── option_set_fwks_ext_gpu
│ │ │ ├── option_set_fwks_key_gpu
│ │ │ ├── option_set_fwks_key_perf_gpu
│ │ │ ├── option_set_fwks_llm_gpu
│ │ │ ├── perf_ip_cpu
│ │ │ ├── perf_ip_inference_lb
│ │ │ ├── perf_ip_inference_sb
│ │ │ ├── perf_ip_knx
│ │ │ ├── perf_ip_training
│ │ │ ├── perf_ip_xe
│ │ │ ├── perf_ip_xe_hp
│ │ │ ├── perf_ip_xe_lp
│ │ │ ├── set_all
│ │ │ ├── set_gpu
│ │ │ ├── set_topologies
│ │ │ ├── shapes_0d
│ │ │ ├── shapes_0d_gpu
│ │ │ ├── shapes_1d
│ │ │ ├── shapes_3d
│ │ │ ├── shapes_alexnet
│ │ │ ├── shapes_bert
│ │ │ ├── shapes_bert_large
│ │ │ ├── shapes_ci
│ │ │ ├── shapes_dien_sb
│ │ │ ├── shapes_dlrm
│ │ │ ├── shapes_gnmt
│ │ │ ├── shapes_googlenet_v1
│ │ │ ├── shapes_googlenet_v3
│ │ │ ├── shapes_maskrcnn
│ │ │ ├── shapes_ncf
│ │ │ ├── shapes_regression
│ │ │ ├── shapes_resnet_50
│ │ │ ├── shapes_resnet_50_sparse
│ │ │ ├── shapes_rnn_t
│ │ │ ├── shapes_transformer_lt
│ │ │ ├── shapes_vgg16
│ │ │ ├── shapes_wd
│ │ │ ├── test_ip_acl
│ │ │ ├── test_ip_all
│ │ │ ├── test_ip_bf32_bfloat16
│ │ │ ├── test_ip_bfloat16
│ │ │ ├── test_ip_bfloat16_ymm
│ │ │ ├── test_ip_ci
│ │ │ ├── test_ip_float16
│ │ │ ├── test_ip_fp8
│ │ │ ├── test_ip_gpu
│ │ │ ├── test_ip_int8
│ │ │ ├── test_ip_large_gpu
│ │ │ └── test_ip_smoke
│ │ ├── lnorm
│ │ │ ├── option_set_all
│ │ │ ├── option_set_fwks_ext_gpu
│ │ │ ├── option_set_fwks_key_gpu
│ │ │ ├── shapes_ci
│ │ │ ├── test_lnorm_all
│ │ │ ├── test_lnorm_bfloat16
│ │ │ ├── test_lnorm_ci
│ │ │ ├── test_lnorm_float16
│ │ │ ├── test_lnorm_gpu
│ │ │ ├── test_lnorm_int8
│ │ │ └── test_lnorm_smoke
│ │ ├── lrn
│ │ │ ├── set_all
│ │ │ ├── shapes_0d
│ │ │ ├── shapes_2d
│ │ │ ├── shapes_3d
│ │ │ ├── shapes_ci
│ │ │ ├── shapes_topologies
│ │ │ ├── test_lrn_all
│ │ │ ├── test_lrn_bfloat16
│ │ │ ├── test_lrn_ci
│ │ │ ├── test_lrn_float16
│ │ │ ├── test_lrn_gpu
│ │ │ └── test_lrn_smoke
│ │ ├── matmul
│ │ │ ├── harness_matmul_3d_bcast
│ │ │ ├── harness_matmul_bert_inf_lb_bfloat16
│ │ │ ├── harness_matmul_bert_inf_lb_int8
│ │ │ ├── harness_matmul_bert_inf_sb_bfloat16
│ │ │ ├── harness_matmul_bert_inf_sb_int8
│ │ │ ├── harness_matmul_bert_tr_bfloat16
│ │ │ ├── harness_matmul_bert_tr_float16
│ │ │ ├── harness_matmul_data_tags
│ │ │ ├── harness_matmul_decompression
│ │ │ ├── harness_matmul_dropout
│ │ │ ├── harness_matmul_generated_ci
│ │ │ ├── harness_matmul_gpt-j_2016-32_inf_lb_bfloat16
│ │ │ ├── harness_matmul_gpt-j_2016-32_inf_lb_f32
│ │ │ ├── harness_matmul_gpt-j_2016-32_inf_lb_float16
│ │ │ ├── harness_matmul_gpt-j_2016-32_inf_sb_bfloat16
│ │ │ ├── harness_matmul_gpt-j_2016-32_inf_sb_f32
│ │ │ ├── harness_matmul_gpt-j_2016-32_inf_sb_float16
│ │ │ ├── harness_matmul_gpt-j_32-32_inf_lb_bfloat16
│ │ │ ├── harness_matmul_gpt-j_32-32_inf_lb_f32
│ │ │ ├── harness_matmul_gpt-j_32-32_inf_lb_float16
│ │ │ ├── harness_matmul_gpt-j_32-32_inf_sb_bfloat16
│ │ │ ├── harness_matmul_gpt-j_32-32_inf_sb_f32
│ │ │ ├── harness_matmul_gpt-j_32-32_inf_sb_float16
│ │ │ ├── harness_matmul_regression_bf16
│ │ │ ├── harness_matmul_regression_f32
│ │ │ ├── harness_matmul_regression_float16
│ │ │ ├── harness_matmul_regression_int8
│ │ │ ├── harness_matmul_runtime_f32
│ │ │ ├── harness_matmul_runtime_int8
│ │ │ ├── harness_matmul_smoke_ref
│ │ │ ├── harness_matmul_strides
│ │ │ ├── harness_matmul_transformer_lt_inf_lb_bfloat16
│ │ │ ├── harness_matmul_transformer_lt_inf_lb_int8
│ │ │ ├── harness_matmul_transformer_lt_inf_sb_bfloat16
│ │ │ ├── harness_matmul_transformer_lt_inf_sb_int8
│ │ │ ├── harness_matmul_transformer_lt_tr_bfloat16
│ │ │ ├── option_set_fp8_mixed
│ │ │ ├── option_set_fwks_ext_gpu
│ │ │ ├── option_set_fwks_key_gpu
│ │ │ ├── option_set_fwks_key_gpu_tf32
│ │ │ ├── option_set_fwks_key_perf_gpu
│ │ │ ├── option_set_fwks_llm_gpu
│ │ │ ├── perf_matmul_inference_batched
│ │ │ ├── perf_matmul_inference_lb
│ │ │ ├── perf_matmul_training
│ │ │ ├── shapes_2d
│ │ │ ├── shapes_2d_ci
│ │ │ ├── shapes_3d
│ │ │ ├── shapes_4bit
│ │ │ ├── shapes_4d
│ │ │ ├── shapes_bert
│ │ │ ├── shapes_bert_large
│ │ │ ├── shapes_converted_ip_inf_lb_alexnet
│ │ │ ├── shapes_converted_ip_inf_lb_dlrm
│ │ │ ├── shapes_converted_ip_inf_lb_gmnt
│ │ │ ├── shapes_converted_ip_inf_lb_googlenet
│ │ │ ├── shapes_converted_ip_inf_lb_maskrcnn
│ │ │ ├── shapes_converted_ip_inf_lb_ncf
│ │ │ ├── shapes_converted_ip_inf_lb_resnet
│ │ │ ├── shapes_converted_ip_inf_lb_rnn_t
│ │ │ ├── shapes_converted_ip_inf_lb_vgg16
│ │ │ ├── shapes_converted_ip_inf_lb_wd
│ │ │ ├── shapes_converted_ip_inf_sb_dien
│ │ │ ├── shapes_converted_ip_tr_alexnet_bwd_d
│ │ │ ├── shapes_converted_ip_tr_alexnet_bwd_w
│ │ │ ├── shapes_converted_ip_tr_alexnet_fwd
│ │ │ ├── shapes_converted_ip_tr_dlrm_bwd_d
│ │ │ ├── shapes_converted_ip_tr_dlrm_bwd_w
│ │ │ ├── shapes_converted_ip_tr_dlrm_fwd
│ │ │ ├── shapes_converted_ip_tr_gmnt_bwd_d
│ │ │ ├── shapes_converted_ip_tr_gmnt_bwd_w
│ │ │ ├── shapes_converted_ip_tr_gmnt_fwd
│ │ │ ├── shapes_converted_ip_tr_googlenet_bwd_d
│ │ │ ├── shapes_converted_ip_tr_googlenet_bwd_w
│ │ │ ├── shapes_converted_ip_tr_googlenet_fwd
│ │ │ ├── shapes_converted_ip_tr_maskrcnn_bwd_d
│ │ │ ├── shapes_converted_ip_tr_maskrcnn_bwd_w
│ │ │ ├── shapes_converted_ip_tr_maskrcnn_fwd
│ │ │ ├── shapes_converted_ip_tr_ncf_bwd_d
│ │ │ ├── shapes_converted_ip_tr_ncf_bwd_w
│ │ │ ├── shapes_converted_ip_tr_ncf_fwd
│ │ │ ├── shapes_converted_ip_tr_resnet_bwd_d
│ │ │ ├── shapes_converted_ip_tr_resnet_bwd_w
│ │ │ ├── shapes_converted_ip_tr_resnet_fwd
│ │ │ ├── shapes_converted_ip_tr_rnn_t_bwd_d
│ │ │ ├── shapes_converted_ip_tr_rnn_t_bwd_w
│ │ │ ├── shapes_converted_ip_tr_rnn_t_fwd
│ │ │ ├── shapes_converted_ip_tr_vgg16_bwd_d
│ │ │ ├── shapes_converted_ip_tr_vgg16_bwd_w
│ │ │ ├── shapes_converted_ip_tr_vgg16_fwd
│ │ │ ├── shapes_converted_ip_tr_wd_bwd_d
│ │ │ ├── shapes_converted_ip_tr_wd_bwd_w
│ │ │ ├── shapes_converted_ip_tr_wd_fwd
│ │ │ ├── shapes_mem_strided
│ │ │ ├── shapes_multidim
│ │ │ ├── shapes_sparse
│ │ │ ├── shapes_sparse_packed
│ │ │ ├── shapes_transformer
│ │ │ ├── test_matmul_all
│ │ │ ├── test_matmul_bf32_bf16
│ │ │ ├── test_matmul_bfloat16
│ │ │ ├── test_matmul_bfloat16_ymm
│ │ │ ├── test_matmul_ci
│ │ │ ├── test_matmul_float16
│ │ │ ├── test_matmul_fp4
│ │ │ ├── test_matmul_fp8
│ │ │ ├── test_matmul_gpu
│ │ │ ├── test_matmul_int8
│ │ │ ├── test_matmul_large_gpu
│ │ │ ├── test_matmul_llm_gpu
│ │ │ ├── test_matmul_multidims
│ │ │ ├── test_matmul_smoke
│ │ │ ├── test_matmul_sparse
│ │ │ ├── test_matmul_sparse_ci
│ │ │ └── test_matmul_sparse_gpu
│ │ ├── pool
│ │ │ ├── harness_pool_regression
│ │ │ ├── harness_pool_smoke_ref
│ │ │ ├── harness_pooling_different_dt
│ │ │ ├── option_set_fwks_ext_gpu
│ │ │ ├── option_set_fwks_key_gpu
│ │ │ ├── perf_pool_gpu
│ │ │ ├── set_all
│ │ │ ├── set_all_small
│ │ │ ├── set_topologies
│ │ │ ├── set_topologies_gpu
│ │ │ ├── shapes_1d
│ │ │ ├── shapes_2d
│ │ │ ├── shapes_2d_small
│ │ │ ├── shapes_3d
│ │ │ ├── shapes_3d_small
│ │ │ ├── shapes_3d_unet
│ │ │ ├── shapes_alexnet
│ │ │ ├── shapes_basic
│ │ │ ├── shapes_global_pooling
│ │ │ ├── shapes_googlenet_v1
│ │ │ ├── shapes_googlenet_v3
│ │ │ ├── shapes_i3d_resnet50_v1
│ │ │ ├── shapes_large_pool
│ │ │ ├── shapes_resnet_50
│ │ │ ├── test_pool_all
│ │ │ ├── test_pool_bfloat16
│ │ │ ├── test_pool_ci
│ │ │ ├── test_pool_float16
│ │ │ ├── test_pool_fp8
│ │ │ ├── test_pool_gpu
│ │ │ ├── test_pool_large_gpu
│ │ │ └── test_pool_smoke
│ │ ├── prelu
│ │ │ ├── option_set_all
│ │ │ ├── shapes_all
│ │ │ ├── shapes_ci
│ │ │ ├── test_prelu_all
│ │ │ ├── test_prelu_bfloat16
│ │ │ ├── test_prelu_ci
│ │ │ ├── test_prelu_float16
│ │ │ ├── test_prelu_gpu
│ │ │ └── test_prelu_smoke
│ │ ├── reduction
│ │ │ ├── harness_reduction_bf16
│ │ │ ├── harness_reduction_f16
│ │ │ ├── harness_reduction_f32
│ │ │ ├── harness_reduction_i8
│ │ │ ├── option_set_all
│ │ │ ├── option_set_all_algs
│ │ │ ├── option_set_all_algs_ci
│ │ │ ├── option_set_all_algs_int8
│ │ │ ├── option_set_all_algs_int8_ci
│ │ │ ├── option_set_fwks_ext_gpu
│ │ │ ├── option_set_fwks_key_gpu
│ │ │ ├── perf_reduction_gpu
│ │ │ ├── shapes_ci
│ │ │ ├── shapes_gpu_all
│ │ │ ├── shapes_nested_gpu
│ │ │ ├── test_reduction_all
│ │ │ ├── test_reduction_bfloat16
│ │ │ ├── test_reduction_ci
│ │ │ ├── test_reduction_float16
│ │ │ ├── test_reduction_gpu
│ │ │ └── test_reduction_smoke
│ │ ├── reorder
│ │ │ ├── harness_conv_reorders_gpu
│ │ │ ├── harness_reorder_amx
│ │ │ ├── harness_reorder_compensation
│ │ │ ├── harness_reorder_cross_engine_gpu
│ │ │ ├── harness_reorder_decompression
│ │ │ ├── harness_reorder_large
│ │ │ ├── harness_reorder_regression
│ │ │ ├── harness_reorder_runtime
│ │ │ ├── harness_reorder_saturation
│ │ │ ├── harness_reorder_scales
│ │ │ ├── option_set_fwks_ext_gpu
│ │ │ ├── option_set_fwks_key_gpu
│ │ │ ├── test_reorder_all
│ │ │ ├── test_reorder_bfloat16
│ │ │ ├── test_reorder_ci
│ │ │ ├── test_reorder_float16
│ │ │ ├── test_reorder_float8
│ │ │ ├── test_reorder_fp4
│ │ │ ├── test_reorder_gpu
│ │ │ ├── test_reorder_int4
│ │ │ └── test_reorder_smoke
│ │ ├── resampling
│ │ │ ├── option_set_fwks_key_gpu
│ │ │ ├── set_all
│ │ │ ├── shapes_1d
│ │ │ ├── shapes_2d
│ │ │ ├── shapes_3d
│ │ │ ├── shapes_ci
│ │ │ ├── shapes_maskrcnn
│ │ │ ├── test_resampling_all
│ │ │ ├── test_resampling_bfloat16
│ │ │ ├── test_resampling_ci
│ │ │ ├── test_resampling_float16
│ │ │ ├── test_resampling_gpu
│ │ │ └── test_resampling_smoke
│ │ ├── rnn
│ │ │ ├── harness_augru_bf32
│ │ │ ├── harness_augru_bfloat16
│ │ │ ├── harness_augru_float16
│ │ │ ├── harness_gru_bf32
│ │ │ ├── harness_gru_bfloat16
│ │ │ ├── harness_gru_f32
│ │ │ ├── harness_gru_float16
│ │ │ ├── harness_gru_int8
│ │ │ ├── harness_gru_regression
│ │ │ ├── harness_lstm_bf32
│ │ │ ├── harness_lstm_bfloat16
│ │ │ ├── harness_lstm_f32
│ │ │ ├── harness_lstm_float16
│ │ │ ├── harness_lstm_int8
│ │ │ ├── harness_rnn_bf32
│ │ │ ├── harness_rnn_bfloat16
│ │ │ ├── harness_rnn_f32
│ │ │ ├── harness_rnn_float16
│ │ │ ├── option_set_fwks_key_gpu
│ │ │ ├── option_set_gnmt_decoder
│ │ │ ├── option_set_gnmt_encoder
│ │ │ ├── option_set_large
│ │ │ ├── option_set_lstmp_large
│ │ │ ├── option_set_lstmp_small
│ │ │ ├── option_set_perf_inference_lb
│ │ │ ├── option_set_perf_inference_sb
│ │ │ ├── option_set_perf_training
│ │ │ ├── option_set_rnnt
│ │ │ ├── option_set_small
│ │ │ ├── perf_rnn_cpu
│ │ │ ├── perf_rnn_inference_lb
│ │ │ ├── perf_rnn_inference_sb
│ │ │ ├── perf_rnn_knx
│ │ │ ├── perf_rnn_training
│ │ │ ├── perf_rnn_xe
│ │ │ ├── perf_rnn_xe_hp
│ │ │ ├── perf_rnn_xe_lp
│ │ │ ├── shapes_deepspeech_2
│ │ │ ├── shapes_inference
│ │ │ ├── shapes_large
│ │ │ ├── shapes_large_gru
│ │ │ ├── shapes_lstmp_large
│ │ │ ├── shapes_lstmp_small
│ │ │ ├── shapes_rnn_t
│ │ │ ├── shapes_small
│ │ │ ├── shapes_small_gru
│ │ │ ├── shapes_training
│ │ │ ├── test_augru_all
│ │ │ ├── test_augru_bf32_bfloat16
│ │ │ ├── test_augru_bfloat16
│ │ │ ├── test_augru_ci
│ │ │ ├── test_augru_float16
│ │ │ ├── test_gru_all
│ │ │ ├── test_gru_bf32_bfloat16
│ │ │ ├── test_gru_bfloat16
│ │ │ ├── test_gru_ci
│ │ │ ├── test_gru_float16
│ │ │ ├── test_gru_int8
│ │ │ ├── test_lstm_all
│ │ │ ├── test_lstm_bf32_bfloat16
│ │ │ ├── test_lstm_bfloat16
│ │ │ ├── test_lstm_bfloat16_ymm
│ │ │ ├── test_lstm_ci
│ │ │ ├── test_lstm_f32
│ │ │ ├── test_lstm_float16
│ │ │ ├── test_lstm_int8
│ │ │ ├── test_rnn_all
│ │ │ ├── test_rnn_bf32_bfloat16
│ │ │ ├── test_rnn_bfloat16
│ │ │ ├── test_rnn_ci
│ │ │ ├── test_rnn_float16
│ │ │ └── test_rnn_gpu
│ │ ├── self
│ │ │ ├── test_self_ci
│ │ │ ├── test_self_f32
│ │ │ └── test_self_smoke
│ │ ├── shuffle
│ │ │ ├── option_set_all
│ │ │ ├── option_set_min
│ │ │ ├── option_set_perf
│ │ │ ├── perf_shuffle_cpu
│ │ │ ├── test_shuffle_all
│ │ │ ├── test_shuffle_bfloat16
│ │ │ ├── test_shuffle_ci
│ │ │ ├── test_shuffle_float16
│ │ │ ├── test_shuffle_gpu
│ │ │ └── test_shuffle_smoke
│ │ ├── softmax
│ │ │ ├── harness_softmax_regression
│ │ │ ├── option_set_fwks_ext_gpu
│ │ │ ├── option_set_fwks_key_gpu
│ │ │ ├── set_0d
│ │ │ ├── shapes_0d
│ │ │ ├── shapes_2d
│ │ │ ├── shapes_3d
│ │ │ ├── shapes_ci
│ │ │ ├── shapes_large
│ │ │ ├── shapes_large_axis
│ │ │ ├── shapes_nlp
│ │ │ ├── test_softmax_acl
│ │ │ ├── test_softmax_all
│ │ │ ├── test_softmax_bfloat16
│ │ │ ├── test_softmax_ci
│ │ │ ├── test_softmax_float16
│ │ │ ├── test_softmax_gpu
│ │ │ └── test_softmax_smoke
│ │ ├── sum
│ │ │ ├── option_set_fwks_ext_gpu
│ │ │ ├── option_set_fwks_key_gpu
│ │ │ ├── test_sum_all
│ │ │ ├── test_sum_bfloat16
│ │ │ ├── test_sum_ci
│ │ │ ├── test_sum_float16
│ │ │ ├── test_sum_gpu
│ │ │ └── test_sum_smoke
│ │ └── zeropad
│ │ │ ├── option_set_fwks_ext_gpu
│ │ │ ├── option_set_fwks_key_gpu
│ │ │ ├── set_dim1_block_3d
│ │ │ ├── set_dim1dim2_block_2d
│ │ │ ├── set_dim1dim2_block_3d
│ │ │ ├── set_dim2_block_3d
│ │ │ ├── set_dim2dim3_block_4d
│ │ │ ├── shapes_dim1_block_3d
│ │ │ ├── shapes_dim1dim2_block_2d
│ │ │ ├── shapes_dim1dim2_block_3d
│ │ │ ├── shapes_dim2_block_3d
│ │ │ ├── shapes_dim2dim3_block_4d
│ │ │ ├── test_zeropad_ci
│ │ │ └── test_zeropad_gpu
│ ├── ip
│ │ ├── bench_ip.cpp
│ │ ├── cfg.cpp
│ │ ├── ip.cpp
│ │ ├── ip.hpp
│ │ ├── ip_aux.cpp
│ │ └── ref_ip.cpp
│ ├── lnorm
│ │ ├── bench_lnorm.cpp
│ │ ├── lnorm.cpp
│ │ ├── lnorm.hpp
│ │ ├── lnorm_aux.cpp
│ │ └── ref_lnorm.cpp
│ ├── lrn
│ │ ├── bench_lrn.cpp
│ │ ├── lrn.cpp
│ │ ├── lrn.hpp
│ │ ├── lrn_aux.cpp
│ │ └── ref_lrn.cpp
│ ├── matmul
│ │ ├── bench_matmul.cpp
│ │ ├── cfg.cpp
│ │ ├── matmul.cpp
│ │ ├── matmul.hpp
│ │ ├── matmul_aux.cpp
│ │ └── ref_matmul.cpp
│ ├── pool
│ │ ├── bench_pool.cpp
│ │ ├── cfg.cpp
│ │ ├── pool.cpp
│ │ ├── pool.hpp
│ │ ├── pool_aux.cpp
│ │ └── ref_pool.cpp
│ ├── prelu
│ │ ├── bench_prelu.cpp
│ │ ├── prelu.cpp
│ │ ├── prelu.hpp
│ │ ├── prelu_aux.cpp
│ │ └── ref_prelu.cpp
│ ├── reduction
│ │ ├── bench_reduction.cpp
│ │ ├── reduction.cpp
│ │ ├── reduction.hpp
│ │ ├── reduction_aux.cpp
│ │ └── ref_reduction.cpp
│ ├── reorder
│ │ ├── bench_reorder.cpp
│ │ ├── cfg.cpp
│ │ ├── ref_reorder.cpp
│ │ ├── reorder.cpp
│ │ ├── reorder.hpp
│ │ └── reorder_aux.cpp
│ ├── resampling
│ │ ├── bench_resampling.cpp
│ │ ├── ref_resampling.cpp
│ │ ├── resampling.cpp
│ │ ├── resampling.hpp
│ │ └── resampling_aux.cpp
│ ├── rnn
│ │ ├── bench_rnn.cpp
│ │ ├── cells.hpp
│ │ ├── cfg.cpp
│ │ ├── gru_cell.cpp
│ │ ├── lbr_gru_cell.cpp
│ │ ├── lstm_cell.cpp
│ │ ├── ref_rnn_bwd.cpp
│ │ ├── ref_rnn_fwd.cpp
│ │ ├── rnn.cpp
│ │ ├── rnn.hpp
│ │ ├── rnn_aux.cpp
│ │ ├── rnn_aux.hpp
│ │ ├── rnn_cell.cpp
│ │ ├── rnn_task.hpp
│ │ ├── rnn_task_executor.hpp
│ │ └── rnn_utils.cpp
│ ├── self
│ │ ├── bnorm.cpp
│ │ ├── common.cpp
│ │ ├── compare.cpp
│ │ ├── conv.cpp
│ │ ├── graph_example.cpp
│ │ ├── memory.cpp
│ │ ├── norm.cpp
│ │ ├── res.cpp
│ │ ├── self.cpp
│ │ └── self.hpp
│ ├── shuffle
│ │ ├── bench_shuffle.cpp
│ │ ├── ref_shuffle.cpp
│ │ ├── shuffle.cpp
│ │ ├── shuffle.hpp
│ │ └── shuffle_aux.cpp
│ ├── softmax
│ │ ├── bench_softmax.cpp
│ │ ├── ref_softmax.cpp
│ │ ├── softmax.cpp
│ │ ├── softmax.hpp
│ │ └── softmax_aux.cpp
│ ├── sum
│ │ ├── bench_sum.cpp
│ │ ├── ref_sum.cpp
│ │ ├── sum.cpp
│ │ ├── sum.hpp
│ │ └── sum_aux.cpp
│ ├── utils
│ │ ├── bench_mode.cpp
│ │ ├── bench_mode.hpp
│ │ ├── cfg.hpp
│ │ ├── cold_cache.cpp
│ │ ├── cold_cache.hpp
│ │ ├── compare.cpp
│ │ ├── compare.hpp
│ │ ├── data_kind.cpp
│ │ ├── data_kind.hpp
│ │ ├── dims.cpp
│ │ ├── dims.hpp
│ │ ├── dnnl_query.cpp
│ │ ├── dnnl_query.hpp
│ │ ├── fill.cpp
│ │ ├── fill.hpp
│ │ ├── impl_filter.cpp
│ │ ├── impl_filter.hpp
│ │ ├── norm.hpp
│ │ ├── numeric.cpp
│ │ ├── numeric.hpp
│ │ ├── parallel.cpp
│ │ ├── parallel.hpp
│ │ ├── parser.cpp
│ │ ├── parser.hpp
│ │ ├── perf_report.cpp
│ │ ├── perf_report.hpp
│ │ ├── res.hpp
│ │ ├── settings.hpp
│ │ ├── stream_kind.cpp
│ │ ├── stream_kind.hpp
│ │ ├── task.hpp
│ │ ├── task_executor.hpp
│ │ ├── timer.cpp
│ │ ├── timer.hpp
│ │ └── wrapper.hpp
│ └── zeropad
│ │ ├── bench_zeropad.cpp
│ │ ├── zeropad.cpp
│ │ ├── zeropad.hpp
│ │ └── zeropad_aux.cpp
├── generate_c_symbols_refs.sh
├── gtests
│ ├── CMakeLists.txt
│ ├── api
│ │ ├── CMakeLists.txt
│ │ ├── test_engine.cpp
│ │ ├── test_memory.cpp
│ │ ├── test_memory_creation.cpp
│ │ ├── test_memory_desc.cpp
│ │ ├── test_memory_desc_ops.cpp
│ │ ├── test_memory_map.cpp
│ │ ├── test_namespace.cpp
│ │ ├── test_stream.cpp
│ │ └── test_submemory.cpp
│ ├── convolution_common.h
│ ├── dnnl_test_common.hpp
│ ├── dnnl_test_common_ocl.hpp
│ ├── dnnl_test_macros.hpp
│ ├── graph
│ │ ├── CMakeLists.txt
│ │ ├── api
│ │ │ ├── CMakeLists.txt
│ │ │ ├── api_test_main.cpp
│ │ │ ├── ocl
│ │ │ │ ├── CMakeLists.txt
│ │ │ │ ├── test_cpp_api_compiled_partition.cpp
│ │ │ │ ├── test_cpp_api_engine.cpp
│ │ │ │ └── test_cpp_api_tensor.cpp
│ │ │ ├── sycl
│ │ │ │ ├── CMakeLists.txt
│ │ │ │ ├── test_cpp_api_compiled_partition.cpp
│ │ │ │ ├── test_cpp_api_engine.cpp
│ │ │ │ └── test_cpp_api_tensor.cpp
│ │ │ ├── test_api_common.cpp
│ │ │ ├── test_api_common.h
│ │ │ ├── test_api_common.hpp
│ │ │ ├── test_c_api_add_op.cpp
│ │ │ ├── test_c_api_compile.cpp
│ │ │ ├── test_c_api_compile_parametrized.cpp
│ │ │ ├── test_c_api_constant_cache.cpp
│ │ │ ├── test_c_api_filter.cpp
│ │ │ ├── test_c_api_graph.cpp
│ │ │ ├── test_c_api_logical_tensor.cpp
│ │ │ ├── test_c_api_op.cpp
│ │ │ ├── test_cpp_api_compile.cpp
│ │ │ ├── test_cpp_api_constant_cache.cpp
│ │ │ ├── test_cpp_api_engine.cpp
│ │ │ ├── test_cpp_api_graph.cpp
│ │ │ ├── test_cpp_api_logical_tensor.cpp
│ │ │ ├── test_cpp_api_op.cpp
│ │ │ ├── test_cpp_api_partition.cpp
│ │ │ └── test_cpp_api_tensor.cpp
│ │ ├── test_allocator.cpp
│ │ ├── test_allocator.hpp
│ │ └── unit
│ │ │ ├── CMakeLists.txt
│ │ │ ├── backend
│ │ │ ├── CMakeLists.txt
│ │ │ ├── dnnl
│ │ │ │ ├── CMakeLists.txt
│ │ │ │ ├── dnnl_test_common.hpp
│ │ │ │ ├── ref_func.hpp
│ │ │ │ ├── test_batch_norm.cpp
│ │ │ │ ├── test_binary_op.cpp
│ │ │ │ ├── test_bmm.cpp
│ │ │ │ ├── test_common.cpp
│ │ │ │ ├── test_compiled_partition.cpp
│ │ │ │ ├── test_concat.cpp
│ │ │ │ ├── test_constant_cache.cpp
│ │ │ │ ├── test_convolution.cpp
│ │ │ │ ├── test_convtranspose.cpp
│ │ │ │ ├── test_dequantize.cpp
│ │ │ │ ├── test_dnnl_infer_shape_cpu.cpp
│ │ │ │ ├── test_dnnl_utils_cpu.cpp
│ │ │ │ ├── test_eltwise.cpp
│ │ │ │ ├── test_fusion_info_cpu.cpp
│ │ │ │ ├── test_graph_cpu.cpp
│ │ │ │ ├── test_group_norm.cpp
│ │ │ │ ├── test_insert_ops_cpu.cpp
│ │ │ │ ├── test_internal_attrs_cpu.cpp
│ │ │ │ ├── test_interpolate.cpp
│ │ │ │ ├── test_large_partition.cpp
│ │ │ │ ├── test_layer_norm.cpp
│ │ │ │ ├── test_layout_id_cpu.cpp
│ │ │ │ ├── test_layout_propagator_cpu.cpp
│ │ │ │ ├── test_logical_tensor_cpu.cpp
│ │ │ │ ├── test_matmul.cpp
│ │ │ │ ├── test_memory_planning_cpu.cpp
│ │ │ │ ├── test_mqa_decomp.cpp
│ │ │ │ ├── test_op_executable.cpp
│ │ │ │ ├── test_op_schema_cpu.cpp
│ │ │ │ ├── test_partition_cpu.cpp
│ │ │ │ ├── test_pass.cpp
│ │ │ │ ├── test_pool.cpp
│ │ │ │ ├── test_prelu.cpp
│ │ │ │ ├── test_quantize.cpp
│ │ │ │ ├── test_reduce.cpp
│ │ │ │ ├── test_reorder.cpp
│ │ │ │ ├── test_scratchpad.cpp
│ │ │ │ ├── test_sdp_decomp.cpp
│ │ │ │ ├── test_select.cpp
│ │ │ │ ├── test_softmax.cpp
│ │ │ │ ├── test_subgraph_pass.cpp
│ │ │ │ ├── test_thread_local_cache_cpu.cpp
│ │ │ │ └── test_typecast.cpp
│ │ │ └── fake
│ │ │ │ ├── CMakeLists.txt
│ │ │ │ ├── test_compiled_partition.cpp
│ │ │ │ ├── test_fake_backend.cpp
│ │ │ │ ├── test_graph.cpp
│ │ │ │ ├── test_partition.cpp
│ │ │ │ └── test_pass.cpp
│ │ │ ├── interface
│ │ │ ├── CMakeLists.txt
│ │ │ ├── sycl
│ │ │ │ ├── CMakeLists.txt
│ │ │ │ └── test_allocator.cpp
│ │ │ ├── test_allocator.cpp
│ │ │ ├── test_backend_cpu.cpp
│ │ │ ├── test_compiled_partition.cpp
│ │ │ ├── test_graph_cpu.cpp
│ │ │ ├── test_logical_tensor_cpu.cpp
│ │ │ ├── test_op_cpu.cpp
│ │ │ ├── test_op_def_constraint_cpu.cpp
│ │ │ ├── test_op_schema_cpu.cpp
│ │ │ ├── test_partition_hashing.cpp
│ │ │ ├── test_shape_infer_cpu.cpp
│ │ │ ├── test_tensor.cpp
│ │ │ └── test_value_cpu.cpp
│ │ │ ├── unit_test_common.cpp
│ │ │ ├── unit_test_common.hpp
│ │ │ ├── unit_test_main.cpp
│ │ │ ├── utils.hpp
│ │ │ └── utils
│ │ │ ├── CMakeLists.txt
│ │ │ ├── test_allocator.cpp
│ │ │ ├── test_attribute_value_cpu.cpp
│ │ │ ├── test_debug_cpu.cpp
│ │ │ ├── test_json_cpu.cpp
│ │ │ ├── test_pattern_matcher_cpu.cpp
│ │ │ └── test_utils_cpu.cpp
│ ├── in
│ │ ├── convolution_attr.h
│ │ ├── convolution_simple.h
│ │ ├── gemm_in.h
│ │ └── layer_normalization.h
│ ├── internals
│ │ ├── CMakeLists.txt
│ │ ├── sdpa_internal.hpp
│ │ ├── test_bcast_strategy.cpp
│ │ ├── test_bfloat16.cpp
│ │ ├── test_brgemm.cpp
│ │ ├── test_comparison_operators.cpp
│ │ ├── test_dnnl_threading.cpp
│ │ ├── test_env_vars_dnnl.cpp
│ │ ├── test_env_vars_onednn.cpp
│ │ ├── test_float8.cpp
│ │ ├── test_nibble.cpp
│ │ ├── test_sdpa.cpp
│ │ ├── test_utils.cpp
│ │ └── test_utils.hpp
│ ├── main.cpp
│ ├── ocl
│ │ ├── CMakeLists.txt
│ │ └── api
│ │ │ ├── CMakeLists.txt
│ │ │ ├── test_engine.cpp
│ │ │ ├── test_memory_buffer.cpp
│ │ │ ├── test_memory_usm.cpp
│ │ │ └── test_stream.cpp
│ ├── regression
│ │ ├── CMakeLists.txt
│ │ └── test_binary_stride.cpp
│ ├── sycl
│ │ ├── CMakeLists.txt
│ │ └── api
│ │ │ ├── CMakeLists.txt
│ │ │ ├── test_engine.cpp
│ │ │ ├── test_memory_buffer.cpp
│ │ │ ├── test_memory_usm.cpp
│ │ │ └── test_stream.cpp
│ ├── test_batch_normalization.cpp
│ ├── test_binary.cpp
│ ├── test_concat.cpp
│ ├── test_concurrency.cpp
│ ├── test_convolution_backward_data_common.hpp
│ ├── test_convolution_backward_data_f32.cpp
│ ├── test_convolution_backward_weights_common.hpp
│ ├── test_convolution_backward_weights_f32.cpp
│ ├── test_convolution_eltwise_forward_common.hpp
│ ├── test_convolution_eltwise_forward_f32.cpp
│ ├── test_convolution_eltwise_forward_x8s8f32s32.cpp
│ ├── test_convolution_format_any.cpp
│ ├── test_convolution_forward_common.hpp
│ ├── test_convolution_forward_f32.cpp
│ ├── test_convolution_forward_u8s8fp.cpp
│ ├── test_convolution_forward_u8s8s32.cpp
│ ├── test_cross_engine_reorder.cpp
│ ├── test_deconvolution.cpp
│ ├── test_eltwise.cpp
│ ├── test_gemm_bf16bf16bf16.cpp
│ ├── test_gemm_bf16bf16f32.cpp
│ ├── test_gemm_common.hpp
│ ├── test_gemm_data_preparation.hpp
│ ├── test_gemm_f16.cpp
│ ├── test_gemm_f16f16f32.cpp
│ ├── test_gemm_f32.cpp
│ ├── test_gemm_params.hpp
│ ├── test_gemm_s8s8s32.cpp
│ ├── test_gemm_s8u8s32.cpp
│ ├── test_gemm_u8s8s32.cpp
│ ├── test_gemm_u8u8s32.cpp
│ ├── test_gemm_validation.hpp
│ ├── test_global_scratchpad.cpp
│ ├── test_group_normalization.cpp
│ ├── test_iface_attr.cpp
│ ├── test_iface_attr_quantization.cpp
│ ├── test_iface_binary_bcast.cpp
│ ├── test_iface_gpu_only.cpp
│ ├── test_iface_handle.cpp
│ ├── test_iface_pd.cpp
│ ├── test_iface_pd_iter.cpp
│ ├── test_iface_primitive_cache.cpp
│ ├── test_iface_runtime_dims.cpp
│ ├── test_iface_sparse.cpp
│ ├── test_iface_threadpool.cpp
│ ├── test_iface_weights_format.cpp
│ ├── test_iface_wino_convolution.cpp
│ ├── test_inner_product_backward_data.cpp
│ ├── test_inner_product_backward_weights.cpp
│ ├── test_inner_product_forward.cpp
│ ├── test_ip_formats.cpp
│ ├── test_isa_hints.cpp
│ ├── test_isa_iface.cpp
│ ├── test_isa_mask.cpp
│ ├── test_layer_normalization.cpp
│ ├── test_lrn.cpp
│ ├── test_malloc.cpp
│ ├── test_malloc.hpp
│ ├── test_matmul.cpp
│ ├── test_persistent_cache_api.cpp
│ ├── test_pooling_backward.cpp
│ ├── test_pooling_forward.cpp
│ ├── test_prelu.cpp
│ ├── test_primitive_cache_mt.cpp
│ ├── test_reduction.cpp
│ ├── test_reorder.cpp
│ ├── test_reorder_common.hpp
│ ├── test_reorder_formats.cpp
│ ├── test_resampling.cpp
│ ├── test_rnn_forward.cpp
│ ├── test_shuffle.cpp
│ ├── test_softmax.cpp
│ └── test_sum.cpp
├── noexcept
│ ├── CMakeLists.txt
│ └── main.cpp
├── other
│ └── subproject
│ │ ├── CMakeLists.txt
│ │ └── main.c
├── test_isa_common.hpp
├── test_thread.cpp
└── test_thread.hpp
└── third_party
├── .clang-format
├── .clang-tidy
├── gtest
├── CMakeLists.txt
├── LICENSE
├── gtest-death-test.h
├── gtest-matchers.h
├── gtest-message.h
├── gtest-param-test.h
├── gtest-printers.h
├── gtest-spi.h
├── gtest-test-part.h
├── gtest-typed-test.h
├── gtest.h
├── gtest_pred_impl.h
├── gtest_prod.h
├── internal
│ ├── custom
│ │ ├── README.md
│ │ ├── gtest-port.h
│ │ ├── gtest-printers.h
│ │ └── gtest.h
│ ├── gtest-death-test-internal.h
│ ├── gtest-filepath.h
│ ├── gtest-internal.h
│ ├── gtest-param-util.h
│ ├── gtest-port-arch.h
│ ├── gtest-port.h
│ ├── gtest-string.h
│ └── gtest-type-util.h
└── src
│ ├── gtest-all.cc
│ ├── gtest-death-test.cc
│ ├── gtest-filepath.cc
│ ├── gtest-internal-inl.h
│ ├── gtest-matchers.cc
│ ├── gtest-port.cc
│ ├── gtest-printers.cc
│ ├── gtest-test-part.cc
│ ├── gtest-typed-test.cc
│ ├── gtest.cc
│ └── gtest_main.cc
├── ittnotify
├── LICENSE.BSD
├── README.md
├── disable_warnings.h
├── ittnotify.h
├── ittnotify_config.h
├── ittnotify_static.c
├── ittnotify_static.h
├── ittnotify_types.h
├── ittptmark64.S
├── ittptmark64.asm
├── jitprofiling.c
├── jitprofiling.h
└── legacy
│ └── ittnotify.h
├── level_zero
├── layers
│ ├── zel_tracing_api.h
│ ├── zel_tracing_ddi.h
│ └── zel_tracing_register_cb.h
├── loader
│ └── ze_loader.h
├── ze_api.h
├── ze_ddi.h
├── ze_ddi_common.h
├── ze_intel_gpu.h
├── ze_stypes.h
├── zes_api.h
├── zes_ddi.h
├── zet_api.h
└── zet_ddi.h
├── mdapi
└── metrics_discovery_api.h
├── ngen
├── COPYRIGHT
├── ngen.hpp
├── ngen_asm.hpp
├── ngen_auto_swsb.hpp
├── ngen_compiler_fix.hpp
├── ngen_config_internal.hpp
├── ngen_core.hpp
├── ngen_debuginfo.hpp
├── ngen_decoder.hpp
├── ngen_elf.hpp
├── ngen_emulation.hpp
├── ngen_gen12.hpp
├── ngen_gen8.hpp
├── ngen_interface.hpp
├── ngen_level_zero.hpp
├── ngen_opencl.hpp
├── ngen_pseudo.hpp
├── ngen_register_allocator.cpp
├── ngen_register_allocator.hpp
├── ngen_register_decl.hpp
├── ngen_registers.hpp
├── ngen_sycl.hpp
├── ngen_utils.hpp
└── npack
│ ├── elf_structs.hpp
│ ├── hash.hpp
│ ├── neo_packager.hpp
│ └── neo_structs.hpp
├── spdlog
├── README.md
├── common-inl.h
├── common.h
├── details
│ ├── backtracer-inl.h
│ ├── backtracer.h
│ ├── circular_q.h
│ ├── console_globals.h
│ ├── file_helper-inl.h
│ ├── file_helper.h
│ ├── fmt_helper.h
│ ├── log_msg-inl.h
│ ├── log_msg.h
│ ├── log_msg_buffer-inl.h
│ ├── log_msg_buffer.h
│ ├── null_mutex.h
│ ├── os-inl.h
│ ├── os.h
│ ├── periodic_worker-inl.h
│ ├── periodic_worker.h
│ ├── registry-inl.h
│ ├── registry.h
│ ├── synchronous_factory.h
│ └── windows_include.h
├── fmt
│ ├── bundled
│ │ ├── base.h
│ │ ├── core.h
│ │ ├── format-inl.h
│ │ └── format.h
│ └── fmt.h
├── formatter.h
├── logger-inl.h
├── logger.h
├── mdc.h
├── pattern_formatter-inl.h
├── pattern_formatter.h
├── sinks
│ ├── ansicolor_sink-inl.h
│ ├── ansicolor_sink.h
│ ├── base_sink-inl.h
│ ├── base_sink.h
│ ├── basic_file_sink-inl.h
│ ├── basic_file_sink.h
│ ├── null_sink.h
│ ├── ostream_sink.h
│ ├── rotating_file_sink-inl.h
│ ├── rotating_file_sink.h
│ ├── sink-inl.h
│ ├── sink.h
│ ├── wincolor_sink-inl.h
│ └── wincolor_sink.h
├── spdlog-inl.h
├── spdlog.h
├── tweakme.h
└── version.h
├── xbyak
├── COPYRIGHT
├── xbyak.h
├── xbyak_bin2hex.h
├── xbyak_mnemonic.h
└── xbyak_util.h
└── xbyak_aarch64
├── CMakeLists.txt
├── src
├── err_impl.h
├── util_impl.cpp
├── util_impl.h
├── util_impl_linux.h
├── util_impl_mac.h
├── util_impl_windows.h
├── xbyak_aarch64_impl.cpp
├── xbyak_aarch64_impl.h
└── xbyak_aarch64_mnemonic.h
└── xbyak_aarch64
├── xbyak_aarch64.h
├── xbyak_aarch64_adr.h
├── xbyak_aarch64_code_array.h
├── xbyak_aarch64_err.h
├── xbyak_aarch64_gen.h
├── xbyak_aarch64_inner.h
├── xbyak_aarch64_label.h
├── xbyak_aarch64_meta_mnemonic.h
├── xbyak_aarch64_mnemonic_def.h
├── xbyak_aarch64_perf.h
├── xbyak_aarch64_reg.h
├── xbyak_aarch64_util.h
└── xbyak_aarch64_version.h
/.github/ISSUE_TEMPLATE/documentation.md:
--------------------------------------------------------------------------------
1 | ---
2 | name: Request a documentation change
3 | about: Use this template to report documentation issue or request documentation changes
4 | title: ''
5 | labels: 'documentation'
6 | assignees: ''
7 | ---
8 |
9 | # Summary
10 | Include a short summary of the issue or request. Sections below provide
11 | guidance on what factors are considered important for a documentation
12 | issue.
13 |
14 | # URLs
15 | Include pointers to documents that are impacted.
16 |
17 | # Additional details
18 | Provide detailed description of the expected changes in documentation
19 | and suggestions you have.
20 |
--------------------------------------------------------------------------------
/.github/ISSUE_TEMPLATE/question.md:
--------------------------------------------------------------------------------
1 | ---
2 | name: Ask a question
3 | about: Use this template for everything that is not a bug or a feature request
4 | title: ''
5 | labels: 'question'
6 | assignees: ''
7 | ---
8 |
--------------------------------------------------------------------------------
/.github/automation/aarch64/ci.json:
--------------------------------------------------------------------------------
1 | {
2 | "dependencies": {
3 | "acl": "v52.1.0",
4 | "gcc": "13",
5 | "clang": "17",
6 | "onednn-base": "v3.7"
7 | }
8 | }
9 |
--------------------------------------------------------------------------------
/.github/codeql-config.yml:
--------------------------------------------------------------------------------
1 | paths:
2 | - ./
3 | - .github/
4 |
--------------------------------------------------------------------------------
/cmake/template.vcxproj.user:
--------------------------------------------------------------------------------
1 | <?xml version="1.0" encoding="utf-8"?>
2 | <Project ToolsVersion="14.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
3 | <PropertyGroup>
4 | <LocalDebuggerEnvironment>PATH=@CTESTCONFIG_PATH@;$(PATH)</LocalDebuggerEnvironment>
5 | <DebuggerFlavor>WindowsLocalDebugger</DebuggerFlavor>
6 | </PropertyGroup>
7 | </Project>
8 |
--------------------------------------------------------------------------------
/doc/advanced/design/mem_fmt_blk.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/advanced/design/mem_fmt_blk.png
--------------------------------------------------------------------------------
/doc/advanced/design/mem_fmt_img1.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/advanced/design/mem_fmt_img1.png
--------------------------------------------------------------------------------
/doc/advanced/design/mem_fmt_img2.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/advanced/design/mem_fmt_img2.png
--------------------------------------------------------------------------------
/doc/advanced/design/mem_fmt_padded_blk.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/advanced/design/mem_fmt_padded_blk.png
--------------------------------------------------------------------------------
/doc/advanced/design/strides.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/advanced/design/strides.png
--------------------------------------------------------------------------------
/doc/build/system_requirements.md:
--------------------------------------------------------------------------------
1 | System Requirements {#dev_guide_system_requirements}
2 | ====================================================
3 |
4 | oneDNN supports a broad list of hardware platforms, operating systems, and compilers.
5 | For details, see [oneDNN System Requirements](https://github.com/uxlfoundation/oneDNN?tab=readme-ov-file#system-requirements).
6 |
--------------------------------------------------------------------------------
/doc/environment.yml:
--------------------------------------------------------------------------------
1 | name: onednn-doc
2 | channels:
3 | - conda-forge
4 | - nodefaults
5 | dependencies:
6 | - python=3.9.5
7 | - pip=21.1.2
8 | - doxyrest=2.1.2
9 | - doxygen=1.8.14
10 | - graphviz=2.40.1
11 | - sphinx=4.0.2
12 | - sphinx-book-theme=0.0.41
13 | - sphinx-copybutton=0.5.2
14 |
--------------------------------------------------------------------------------
/doc/graph/fusion_patterns/images/binary_pattern.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/fusion_patterns/images/binary_pattern.png
--------------------------------------------------------------------------------
/doc/graph/fusion_patterns/images/compressed_sdpa_pattern.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/fusion_patterns/images/compressed_sdpa_pattern.png
--------------------------------------------------------------------------------
/doc/graph/fusion_patterns/images/conv_bwd_pattern.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/fusion_patterns/images/conv_bwd_pattern.png
--------------------------------------------------------------------------------
/doc/graph/fusion_patterns/images/conv_pattern.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/fusion_patterns/images/conv_pattern.png
--------------------------------------------------------------------------------
/doc/graph/fusion_patterns/images/convtranspose_pattern.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/fusion_patterns/images/convtranspose_pattern.png
--------------------------------------------------------------------------------
/doc/graph/fusion_patterns/images/epilogue_subgraph_conv.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/fusion_patterns/images/epilogue_subgraph_conv.png
--------------------------------------------------------------------------------
/doc/graph/fusion_patterns/images/epilogue_subgraph_general_1.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/fusion_patterns/images/epilogue_subgraph_general_1.png
--------------------------------------------------------------------------------
/doc/graph/fusion_patterns/images/epilogue_subgraph_general_2.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/fusion_patterns/images/epilogue_subgraph_general_2.png
--------------------------------------------------------------------------------
/doc/graph/fusion_patterns/images/epilogue_subgraph_matmul.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/fusion_patterns/images/epilogue_subgraph_matmul.png
--------------------------------------------------------------------------------
/doc/graph/fusion_patterns/images/f2f_conversion.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/fusion_patterns/images/f2f_conversion.png
--------------------------------------------------------------------------------
/doc/graph/fusion_patterns/images/f2q_conversion_general.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/fusion_patterns/images/f2q_conversion_general.png
--------------------------------------------------------------------------------
/doc/graph/fusion_patterns/images/f2q_conversion_quantized_conv.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/fusion_patterns/images/f2q_conversion_quantized_conv.png
--------------------------------------------------------------------------------
/doc/graph/fusion_patterns/images/f2q_conversion_quantized_matmul.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/fusion_patterns/images/f2q_conversion_quantized_matmul.png
--------------------------------------------------------------------------------
/doc/graph/fusion_patterns/images/f2q_conversion_softmax.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/fusion_patterns/images/f2q_conversion_softmax.png
--------------------------------------------------------------------------------
/doc/graph/fusion_patterns/images/fp-gated-mlp.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/fusion_patterns/images/fp-gated-mlp.png
--------------------------------------------------------------------------------
/doc/graph/fusion_patterns/images/gated-mlp-swish.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/fusion_patterns/images/gated-mlp-swish.png
--------------------------------------------------------------------------------
/doc/graph/fusion_patterns/images/gqa.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/fusion_patterns/images/gqa.png
--------------------------------------------------------------------------------
/doc/graph/fusion_patterns/images/interpolate_pattern.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/fusion_patterns/images/interpolate_pattern.png
--------------------------------------------------------------------------------
/doc/graph/fusion_patterns/images/matmul_pattern.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/fusion_patterns/images/matmul_pattern.png
--------------------------------------------------------------------------------
/doc/graph/fusion_patterns/images/norm_pattern.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/fusion_patterns/images/norm_pattern.png
--------------------------------------------------------------------------------
/doc/graph/fusion_patterns/images/pool_pattern.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/fusion_patterns/images/pool_pattern.png
--------------------------------------------------------------------------------
/doc/graph/fusion_patterns/images/q2f_conversion_quantized_conv_matmul.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/fusion_patterns/images/q2f_conversion_quantized_conv_matmul.png
--------------------------------------------------------------------------------
/doc/graph/fusion_patterns/images/q2f_conversion_quantized_convtranspose.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/fusion_patterns/images/q2f_conversion_quantized_convtranspose.png
--------------------------------------------------------------------------------
/doc/graph/fusion_patterns/images/quantized_conv_pattern.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/fusion_patterns/images/quantized_conv_pattern.png
--------------------------------------------------------------------------------
/doc/graph/fusion_patterns/images/quantized_convtranspose_pattern.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/fusion_patterns/images/quantized_convtranspose_pattern.png
--------------------------------------------------------------------------------
/doc/graph/fusion_patterns/images/quantized_matmul_pattern.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/fusion_patterns/images/quantized_matmul_pattern.png
--------------------------------------------------------------------------------
/doc/graph/fusion_patterns/images/reduction_pattern.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/fusion_patterns/images/reduction_pattern.png
--------------------------------------------------------------------------------
/doc/graph/fusion_patterns/images/sdpa-mask-1.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/fusion_patterns/images/sdpa-mask-1.png
--------------------------------------------------------------------------------
/doc/graph/fusion_patterns/images/sdpa-mask-2.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/fusion_patterns/images/sdpa-mask-2.png
--------------------------------------------------------------------------------
/doc/graph/fusion_patterns/images/sdpa-mask-3.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/fusion_patterns/images/sdpa-mask-3.png
--------------------------------------------------------------------------------
/doc/graph/fusion_patterns/images/sdpa-mask-4.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/fusion_patterns/images/sdpa-mask-4.png
--------------------------------------------------------------------------------
/doc/graph/fusion_patterns/images/sdpa-reorder.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/fusion_patterns/images/sdpa-reorder.png
--------------------------------------------------------------------------------
/doc/graph/fusion_patterns/images/sdpa.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/fusion_patterns/images/sdpa.png
--------------------------------------------------------------------------------
/doc/graph/fusion_patterns/images/softmax_pattern.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/fusion_patterns/images/softmax_pattern.png
--------------------------------------------------------------------------------
/doc/graph/fusion_patterns/images/unary_pattern.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/fusion_patterns/images/unary_pattern.png
--------------------------------------------------------------------------------
/doc/graph/programming_model/images/bf16_programming.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/programming_model/images/bf16_programming.jpg
--------------------------------------------------------------------------------
/doc/graph/programming_model/images/img_graph_programming_model.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/programming_model/images/img_graph_programming_model.png
--------------------------------------------------------------------------------
/doc/graph/programming_model/images/int8_programming.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/programming_model/images/int8_programming.jpg
--------------------------------------------------------------------------------
/doc/graph/rst/graph_programming_model.rst:
--------------------------------------------------------------------------------
1 | Programming Model
2 | #################
3 |
4 | .. toctree::
5 | :maxdepth: 1
6 |
7 | dev_guide_graph_basic_concepts
8 | dev_guide_graph_low_precision
9 |
--------------------------------------------------------------------------------
/doc/graph/rst/images/other_pattern.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/graph/rst/images/other_pattern.png
--------------------------------------------------------------------------------
/doc/performance_considerations/benchdnn.md:
--------------------------------------------------------------------------------
1 | Benchmarking Performance {#dev_guide_benchdnn}
2 | ==============================================
3 |
4 | oneDNN has a built-in benchmarking program called benchdnn.
5 |
6 | For a complete description of the available options and working examples, see
7 | the [benchdnn readme](https://github.com/uxlfoundation/oneDNN/blob/main/tests/benchdnn/README.md#benchdnn).
8 |
--------------------------------------------------------------------------------
/doc/performance_considerations/vtune.md:
--------------------------------------------------------------------------------
1 | Profiling with VTune(TM) Profiler {#dev_guide_vtune}
2 | ========================================================
3 |
4 | See @ref dev_guide_profilers
5 |
--------------------------------------------------------------------------------
/doc/primitives/images/unrolled_stack_rnn.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/primitives/images/unrolled_stack_rnn.jpg
--------------------------------------------------------------------------------
/doc/programming_model/images/img_depthwise_fusion.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/programming_model/images/img_depthwise_fusion.jpg
--------------------------------------------------------------------------------
/doc/programming_model/images/img_dnnl_object_snapshot.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/programming_model/images/img_dnnl_object_snapshot.jpg
--------------------------------------------------------------------------------
/doc/programming_model/images/img_dnnl_programming_flow.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/programming_model/images/img_dnnl_programming_flow.jpg
--------------------------------------------------------------------------------
/doc/programming_model/images/img_overview_flow.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/programming_model/images/img_overview_flow.jpg
--------------------------------------------------------------------------------
/doc/programming_model/images/img_programming_model.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/programming_model/images/img_programming_model.png
--------------------------------------------------------------------------------
/doc/rst/advanced_topics.rst:
--------------------------------------------------------------------------------
1 | Advanced Topics
2 | #####################
3 |
4 | .. toctree::
5 | :maxdepth: 1
6 |
7 | dev_guide_transition_to_dnnl
8 | dev_guide_understanding_memory_formats
9 | dev_guide_int8_computations
10 | dev_guide_primitive_cache
11 | dev_guide_persistent_cache
12 | dev_guide_threadpool
13 | dev_guide_sparsity
14 | dev_guide_experimental
15 |
--------------------------------------------------------------------------------
/doc/rst/build_and_link.rst:
--------------------------------------------------------------------------------
1 | Build and Link oneDNN
2 | #####################
3 |
4 | .. toctree::
5 | :maxdepth: 1
6 |
7 | dev_guide_system_requirements
8 | dev_guide_build
9 | dev_guide_build_options
10 | dev_guide_link
--------------------------------------------------------------------------------
/doc/rst/graph_extension.rst:
--------------------------------------------------------------------------------
1 | Graph Extension
2 | ###############
3 |
4 | .. toctree::
5 | :maxdepth: 1
6 |
7 | graph_programming_model
8 | graph_supported_operations
9 | graph_fusion_patterns
10 | dev_guide_graph_dump
11 | dev_guide_constant_tensor_cache
12 |
--------------------------------------------------------------------------------
/doc/rst/interop_with_dpcpp_and_opencl.rst:
--------------------------------------------------------------------------------
1 | Interoperability with DPC++ and OpenCL
2 | ########################################
3 |
4 | .. toctree::
5 | :maxdepth: 1
6 |
7 | dev_guide_opencl_interoperability.rst
8 | dev_guide_dpcpp_interoperability.rst
9 |
--------------------------------------------------------------------------------
/doc/rst/performance_profiling_and_inspection.rst:
--------------------------------------------------------------------------------
1 | Performance Profiling and Inspection
2 | ########################################
3 |
4 | .. toctree::
5 | :maxdepth: 1
6 |
7 | dev_guide_verbose
8 | dev_guide_performance_settings
9 | dev_guide_benchdnn
10 | dev_guide_profilers
11 | dev_guide_inspecting_jit
12 | page_performance_profiling_cpp
13 | dev_guide_cpu_dispatcher_control
14 | dev_guide_cpu_isa_hints
15 | dev_guide_verbose_table
16 |
--------------------------------------------------------------------------------
/doc/rst/programming_model.rst:
--------------------------------------------------------------------------------
1 | oneDNN Concepts
2 | ###############
3 |
4 | .. toctree::
5 | :maxdepth: 1
6 |
7 | page_memory_format_propagation_cpp
8 | dev_guide_inference_and_training_aspects
9 | dev_guide_attributes
10 | dev_guide_data_types
11 | page_cross_engine_reorder_cpp
12 | dev_guide_c_and_cpp_apis
13 | interop_with_dpcpp_and_opencl
--------------------------------------------------------------------------------
/doc/rst/supported_primitives.rst:
--------------------------------------------------------------------------------
1 | Supported Primitives
2 | #####################
3 |
4 | .. toctree::
5 | :maxdepth: 1
6 |
7 | dev_guide_convolution
8 | dev_guide_inner_product
9 | dev_guide_matmul
10 | dev_guide_rnn
11 | dev_guide_batch_normalization
12 | dev_guide_binary
13 | dev_guide_concat
14 | dev_guide_eltwise
15 | dev_guide_group_normalization
16 | dev_guide_layer_normalization
17 | dev_guide_lrn
18 | dev_guide_pooling
19 | dev_guide_prelu
20 | dev_guide_resampling
21 | dev_guide_shuffle
22 | dev_guide_softmax
23 | dev_guide_sum
24 | dev_guide_reorder
25 | dev_guide_reduction
26 |
--------------------------------------------------------------------------------
/doc/rst/ukernels.rst:
--------------------------------------------------------------------------------
1 | Ukernels
2 | #####################
3 |
4 | .. toctree::
5 | :maxdepth: 1
6 |
7 | dev_guide_ukernel_basic_concepts.rst
8 | dev_guide_ukernel_brgemm.rst
9 | dev_guide_ukernel_transform.rst
10 | page_cpu_brgemm_example_cpp.rst
11 |
--------------------------------------------------------------------------------
/doc/sphinx/_static/favicons.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/sphinx/_static/favicons.png
--------------------------------------------------------------------------------
/doc/sphinx/_static/oneAPI-rgb-rev-100.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/sphinx/_static/oneAPI-rgb-rev-100.png
--------------------------------------------------------------------------------
/doc/usage_models/images/img_bf16_diagram.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/usage_models/images/img_bf16_diagram.png
--------------------------------------------------------------------------------
/doc/usage_models/images/img_diagram.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/usage_models/images/img_diagram.png
--------------------------------------------------------------------------------
/doc/usage_models/images/img_inference_scope.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/usage_models/images/img_inference_scope.jpg
--------------------------------------------------------------------------------
/doc/usage_models/images/img_multiscalar.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/usage_models/images/img_multiscalar.png
--------------------------------------------------------------------------------
/doc/usage_models/images/img_singlescalar.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/usage_models/images/img_singlescalar.png
--------------------------------------------------------------------------------
/doc/usage_models/images/img_training_inference_scope.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/uxlfoundation/oneDNN/da36b74f751a4caf34c739bfc56754cdaa31555e/doc/usage_models/images/img_training_inference_scope.jpg
--------------------------------------------------------------------------------
/doc/usage_models/training.md:
--------------------------------------------------------------------------------
1 | Training {#dev_guide_training}
2 | ==============================
3 |
4 | NEW_CONTENT_GOES_HERE
5 |
6 | ## fp32 Training
7 |
8 | NEW_CONTENT_GOES_HERE
9 |
10 | ## bfp16 Training
11 |
12 | NEW_CONTENT_GOES_HERE
13 |
--------------------------------------------------------------------------------
/pyproject.toml:
--------------------------------------------------------------------------------
1 | [tool.black]
2 | line-length = 80
3 | include = 'scripts\/.*\.pyi?#39;
4 |
--------------------------------------------------------------------------------
/src/cpu/jit_utils/linux_perf/README.md:
--------------------------------------------------------------------------------
1 | This is an implementation of jitdump format used by linux perf. The
2 | [spec](https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/plain/tools/perf/Documentation/jitdump-specification.txt)
3 |
--------------------------------------------------------------------------------
/src/gpu/intel/jit/gemm/.clang-tidy:
--------------------------------------------------------------------------------
1 | Checks: '-*,misc-definitions-in-headers'
2 | CheckOptions:
3 | - { key: HeaderFileExtensions, value: "x" }
4 |
--------------------------------------------------------------------------------
/src/gpu/intel/microkernels/.clang-tidy:
--------------------------------------------------------------------------------
1 | Checks: '-*,misc-definitions-in-headers'
2 | CheckOptions:
3 | - { key: HeaderFileExtensions, value: "x" }
4 |
--------------------------------------------------------------------------------
/tests/benchdnn/doc/knobs_dir.md:
--------------------------------------------------------------------------------
1 | # Direction
2 |
3 | **Benchdnn** renames the library propagation kind abstraction into "direction".
4 | The following direction values are supported:
5 |
6 | | Prop kind | Description
7 | | :--- | :---
8 | | FWD_B | dnnl_forward_training w/ bias
9 | | FWD_D | dnnl_forward_training w/o bias
10 | | FWD_I | dnnl_forward_inference
11 | | BWD_D | dnnl_backward_data
12 | | BWD_WB | dnnl_backward_weights w/ bias
13 | | BWD_W | dnnl_backward_weights w/o bias
14 | | BWD_DW | dnnl_backward
15 |
16 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/binary/harness_binary_regression:
--------------------------------------------------------------------------------
1 | # repeated sum with varying scale
2 | --reset --attr-post-ops=sum+relu+sum:2 8x8x3x5:8x8x3x5_n"multisum"
3 |
4 | # Curious edge case in GPU JIT-reorder-based binary
5 | --reset --alg=ADD --stag=ABcd32a16b:ABcd32a16b --dtag=acdb --sdt=f16:f16 --ddt=f16 64x168x42x42:64x168x42x42
6 |
7 | # Mixed src1/post-op src broadcast
8 | --reset --attr-post-ops=add:f32:2 1x17:1x1
9 |
10 | # per_w broadcasting strategy
11 | --reset --attr-post-ops=mul:f32:4+add:f32:4 --alg=add 1x20x768:1x20x1
12 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/binary/option_set_src0_bcast:
--------------------------------------------------------------------------------
1 | --stag=abx:abx 1x5x1x1:3x5x6x9
2 | 4x1x4x4:4x4x1x4
3 | 1x1x1x1:16x12x2x2
4 | 1x12:12x1
5 |
6 | --stag=aBx8b:aBx16b 2x16x1x1:2x16x5x7
7 |
8 | --stag=aBx16b:axb 1x16x5x7:2x16x5x7
9 | 2x16x1x1:2x16x1x1
10 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/binary/shapes_ci:
--------------------------------------------------------------------------------
1 | 3x5x6x9:3x5x6x9
2 | 5x3x2x9:1x3x2x9
3 | 32x17x2x3:32x17
4 | 32x17x2x3:1x17
5 | 15x12x3x5:15
6 | 15x12x3x5:1
7 | 5x3x1x9:1x3x2x9
8 | 12x12:1x12
9 | 12x1:1x12
10 | 2x3x48:1x3x48
11 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/binary/shapes_perf_1st_conv:
--------------------------------------------------------------------------------
1 | 64x3x224x224:1x3x1x1
2 | 256x3x224x224:1x3x1x1
3 | 512x3x224x224:1x3x1x1
4 | 1024x3x224x224:1x3x1x1
5 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/binary/shapes_perf_scaleshift:
--------------------------------------------------------------------------------
1 | 1024x2048x10x10:1x2048x1x1
2 | 1024x1024x19x19:1x1024x1x1
3 | 1024x512x38x38:1x512x1x1
4 | 1024x256x75x75:1x256x1x1
5 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/binary/test_binary_all:
--------------------------------------------------------------------------------
1 | # all
2 | --reset
3 |
4 | --batch=harness_binary_f32
5 | --batch=harness_binary_bf16
6 | --batch=harness_binary_f16
7 | --batch=harness_binary_i8
8 | --batch=harness_binary_different_dt
9 | --batch=harness_binary_regression
10 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/binary/test_binary_bfloat16:
--------------------------------------------------------------------------------
1 | # bf16 (for legacy infra support)
2 | --reset
3 |
4 | --batch=harness_binary_bf16
5 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/binary/test_binary_different_dt_ci:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --inplace=false # Different src and dst data types does not support in-place mode.
4 | --ddt=s8,u8,f32,s32 --sdt=s8:u8,u8:s8,s8:f32,f32:u8,f32:f32,f32:s32,s32:f32
5 | --alg=ADD,MUL,MAX,MIN,DIV,SUB,GE,GT,LE,LT,EQ,NE,SELECT
6 | --stag=abx:any,axb:any
7 | --batch=shapes_ci
8 |
9 | --alg=ADD
10 | --attr-post-ops=,add:f32:per_oc+sum:2+linear:2:1:3
11 | --attr-scales=,src:common:0.5+src1:common:0.25
12 | --batch=shapes_ci
13 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/binary/test_binary_float16:
--------------------------------------------------------------------------------
1 | # f16
2 | --reset
3 |
4 | --batch=harness_binary_f16
5 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/bnorm/perf_bnorm_gpu:
--------------------------------------------------------------------------------
1 | --reset --batch=option_set_fwks_key_gpu
2 | --reset --batch=option_set_fwks_ext_gpu
3 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/bnorm/set_nd:
--------------------------------------------------------------------------------
1 | --batch=shapes_1d
2 | --batch=shapes_2d
3 | --batch=shapes_3d
4 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/bnorm/shapes_1d:
--------------------------------------------------------------------------------
1 | # random 1d problems
2 |
3 | mb1ic16iw1n"bnorm1d:1"
4 | mb1ic32iw30n"bnorm1d:2"
5 | mb32ic128iw10n"bnorm1d:3"
6 | mb32ic27iw7n"bnorm1d:4"
7 | mb9ic128iw1n"bnorm1d:5"
8 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/bnorm/shapes_2d:
--------------------------------------------------------------------------------
1 | # random 2d problems
2 |
3 | mb1ic1ih1_n"bnorm_2d:1"
4 | mb1ic3ih1_n"bnorm_2d:2"
5 | mb1ic256ih28_n"bnorm_2d:3"
6 | mb1ic257ih28_n"bnorm_2d:4"
7 | mb1ic1024ih28iw17_n"bnorm_2d:5"
8 | mb2ic128ih2_n"bnorm_2d:catch_cancel"
9 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/bnorm/shapes_3d:
--------------------------------------------------------------------------------
1 | # random 3d problems
2 |
3 | mb1ic16id40_n"bnorm_3d:1"
4 | mb32ic128id12ih10iw8_n"bnorm_3d:2"
5 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/bnorm/shapes_ci:
--------------------------------------------------------------------------------
1 | # random problems
2 |
3 | ic23_n"bnorm_ci_0d:1"
4 | ic32iw32_n"bnorm_ci_1d:1"
5 | mb5ic33iw27_n"bnorm_ci_1d:2"
6 | ic256ih28_n"bnorm_ci_2d:1" # Used in smoke validation, don't change the name
7 | mb4ic200ih17iw16_n"bnorm_ci_2d:2" # Used in smoke validation, don't change the name
8 | ic16id12_n"bnorm_ci_3d:1"
9 | mb1ic19id4ih6iw6_n"bnorm_ci_3d:2"
10 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/bnorm/shapes_large:
--------------------------------------------------------------------------------
1 | # Large minibatch
2 | mb3072ic64ih112
3 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/bnorm/shapes_regressions:
--------------------------------------------------------------------------------
1 | # Blocking bugs
2 | mb2ic251ih28 # 2-core system w/ 1.5MB LLC slices
3 | mb2ic64ih224 # 2-core system w/ 2.5MB LLC slices
4 | mb2ic64ih56 # 4-core system w/ 1.5MB LLC slices
5 | mb8ic322ih16 # 4-core system w/ 2.5MB LLC slices
6 | mb88ic600ih16 # 36-core system w/ 1.5MB LLC slices
7 | mb88ic980ih16 # 44-core system w/ 2.5MB LLC slices
8 | mb88ic800ih16 # 48-core system w/ 1.5MB LLC slices
9 | mb88ic900ih16 # 56-core system w/ 1.5MB LLC slices
10 |
11 | # Spatial threading bugs for bfloat16
12 | mb12ic4ih8
13 | mb1ic24ih14iw14
14 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/bnorm/test_bnorm_regressions:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --inplace=true
4 | --tag=axb,abx,aBx8b,aBx16b
5 |
6 | # training
7 | --dir=FWD_D,BWD_DW
8 | --dt=f32,bf16
9 | --flags=,G,CH,R,GCHR
10 | --batch=shapes_regressions
11 |
12 | # inference
13 | --dir=FWD_I
14 |
15 | --dt=f32,bf16
16 | --flags=,GCH
17 | --attr-post-ops=,relu
18 | --batch=shapes_regressions
19 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/bnorm/test_bnorm_regressions_large:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --skip-impl=ref
4 |
5 | --tag=axb,abx,aBx16b --dir=FWD_D,BWD_DW --flags=CHR --batch=shapes_large
6 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/bnorm/test_bnorm_smoke:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --match=.*bnorm_ci_2d.* # Use 2d problems only from shapes_ci
4 | --inplace=false
5 | --tag=axb
6 |
7 | # training
8 | --dir=FWD_D,BWD_DW
9 | --dt=f32,bf16,f16
10 | --flags=,G,CHR
11 | --batch=shapes_ci
12 | ## no scale or shift support for backward_data
13 | --dir=BWD_D
14 | --flags=,G,R
15 | --batch=shapes_ci
16 |
17 | # inference
18 | --dir=FWD_I
19 |
20 | --dt=f32,bf16,f16
21 | --flags=,G,CH
22 | --attr-post-ops=,relu
23 | --batch=shapes_ci
24 | --attr-post-ops=
25 |
26 | --dt=f16
27 | --flags=,G,CH
28 | --batch=shapes_ci
29 |
30 | --dt=s8
31 | --flags=G,GCH
32 | --batch=shapes_ci
33 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/brgemm/harness_brgemm_f32:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --dt=f32
4 | --bia_dt=undef,f32
5 | --beta=0,1
6 | --attr-post-ops=,sum:2,relu
7 | --ld=,:160:96
8 | --batch=option_set_f32
9 |
10 | # Separate cases for non-default alpha
11 | --reset
12 | --dt=f32
13 | --alpha=2
14 | --batch=shapes_2d_no_tail_f32
15 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/brgemm/harness_brgemm_fpmath:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --attr-fpmath=bf16
4 |
5 | # f32
6 | --dt=f32
7 | --bia_dt=undef,f32
8 | --beta=0,1
9 | --batch=option_set_f32
10 |
11 | # Separate cases for non-default alpha
12 | --reset
13 | --alpha=2
14 | --batch=shapes_2d_no_tail_f32
15 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/brgemm/option_set_bf16:
--------------------------------------------------------------------------------
1 | # Incorporates all meaningful shapes with bs included.
2 | # bs > 1 is applicable when K is divisible by K-block size.
3 |
4 | --bs=1,7,16
5 |
6 | --batch=shapes_2d_no_tail_bf16
7 |
8 | --batch=shapes_2d_tail_n_bf16
9 |
10 | --bs=1
11 |
12 | --batch=shapes_2d_tail_k_bf16
13 | --batch=shapes_2d_big_k_bf16
14 |
15 | --batch=shapes_2d_tail_k_tail_n_bf16
16 | --batch=shapes_2d_big_k_tail_n_bf16
17 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/brgemm/option_set_f32:
--------------------------------------------------------------------------------
1 | # Incorporates all meaningful shapes with bs included.
2 | # bs > 1 is applicable when K is divisible by K-block size.
3 |
4 | --bs=1,7,16
5 |
6 | --batch=shapes_2d_no_tail_f32
7 |
8 | --batch=shapes_2d_tail_n_f32
9 |
10 | --bs=1
11 |
12 | --batch=shapes_2d_tail_k_f32
13 | --batch=shapes_2d_big_k_f32
14 |
15 | --batch=shapes_2d_tail_k_tail_n_f32
16 | --batch=shapes_2d_big_k_tail_n_f32
17 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/brgemm/option_set_int8:
--------------------------------------------------------------------------------
1 | # Incorporates all meaningful shapes with bs included.
2 | # bs > 1 is applicable when K is divisible by K-block size.
3 |
4 | --bs=1,7,16
5 |
6 | --batch=shapes_2d_no_tail_int8
7 |
8 | --batch=shapes_2d_tail_n_int8
9 |
10 | --bs=1
11 |
12 | --batch=shapes_2d_tail_k_int8
13 | --batch=shapes_2d_big_k_int8
14 |
15 | --batch=shapes_2d_tail_k_tail_n_int8
16 | --batch=shapes_2d_big_k_tail_n_int8
17 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/brgemm/shapes_2d_big_k_bf16:
--------------------------------------------------------------------------------
1 |
2 |
3 | 1x256:256x16_n"bf16:big_k:0"
4 | 1x256:256x32_n"bf16:big_k:1"
5 | 1x256:256x48_n"bf16:big_k:2"
6 | 1x256:256x64_n"bf16:big_k:3"
7 |
8 |
9 | 13x256:256x16_n"bf16:big_k:4"
10 | 13x256:256x32_n"bf16:big_k:5"
11 | 13x256:256x48_n"bf16:big_k:6"
12 | 13x256:256x64_n"bf16:big_k:7"
13 |
14 |
15 | 16x256:256x16_n"bf16:big_k:8"
16 | 16x256:256x32_n"bf16:big_k:9"
17 | 16x256:256x48_n"bf16:big_k:10"
18 | 16x256:256x64_n"bf16:big_k:11"
19 |
20 |
21 | 64x256:256x16_n"bf16:big_k:12"
22 | 64x256:256x32_n"bf16:big_k:13"
23 | 64x256:256x48_n"bf16:big_k:14"
24 | 64x256:256x64_n"bf16:big_k:15"
25 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/brgemm/shapes_2d_big_k_f32:
--------------------------------------------------------------------------------
1 |
2 |
3 | 1x136:136x16_n"f32:big_k:0"
4 | 1x136:136x32_n"f32:big_k:1"
5 | 1x136:136x48_n"f32:big_k:2"
6 | 1x136:136x64_n"f32:big_k:3"
7 |
8 |
9 | 13x136:136x16_n"f32:big_k:4"
10 | 13x136:136x32_n"f32:big_k:5"
11 | 13x136:136x48_n"f32:big_k:6"
12 | 13x136:136x64_n"f32:big_k:7"
13 |
14 |
15 | 16x136:136x16_n"f32:big_k:8"
16 | 16x136:136x32_n"f32:big_k:9"
17 | 16x136:136x48_n"f32:big_k:10"
18 | 16x136:136x64_n"f32:big_k:11"
19 |
20 |
21 | 64x136:136x16_n"f32:big_k:12"
22 | 64x136:136x32_n"f32:big_k:13"
23 | 64x136:136x48_n"f32:big_k:14"
24 | 64x136:136x64_n"f32:big_k:15"
25 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/brgemm/shapes_2d_big_k_int8:
--------------------------------------------------------------------------------
1 |
2 |
3 | 1x512:512x16_n"int8:big_k:0"
4 | 1x512:512x32_n"int8:big_k:1"
5 | 1x512:512x48_n"int8:big_k:2"
6 | 1x512:512x64_n"int8:big_k:3"
7 |
8 |
9 | 13x512:512x16_n"int8:big_k:4"
10 | 13x512:512x32_n"int8:big_k:5"
11 | 13x512:512x48_n"int8:big_k:6"
12 | 13x512:512x64_n"int8:big_k:7"
13 |
14 |
15 | 16x512:512x16_n"int8:big_k:8"
16 | 16x512:512x32_n"int8:big_k:9"
17 | 16x512:512x48_n"int8:big_k:10"
18 | 16x512:512x64_n"int8:big_k:11"
19 |
20 |
21 | 64x512:512x16_n"int8:big_k:12"
22 | 64x512:512x32_n"int8:big_k:13"
23 | 64x512:512x48_n"int8:big_k:14"
24 | 64x512:512x64_n"int8:big_k:15"
25 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/brgemm/test_brgemm_all:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --batch=test_brgemm_ci
4 |
5 | --batch=test_brgemm_f32
6 |
7 | --batch=test_brgemm_bf16
8 |
9 | --batch=test_brgemm_f16
10 |
11 | --batch=test_brgemm_f8
12 |
13 | --batch=test_brgemm_int8
14 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/brgemm/test_brgemm_f32:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --batch=harness_brgemm_f32
4 |
5 | --batch=harness_brgemm_fpmath
6 |
7 | --reset
8 | --brgemm-attr=generate_skip_accumulation:1
9 | --beta=0,1
10 | --bia_dt=f32
11 | --dt=f32
12 | --batch=option_set_f32
13 |
14 | # ukernel wtag support
15 | --reset
16 | --wtag=ba
17 | --dt=f32
18 | --batch=option_set_f32
19 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/brgemm/test_brgemm_f8:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --batch=harness_brgemm_f8
4 |
5 | # ukernel wtag support
6 | --reset
7 | --wtag=ba
8 | --dt=f8_e5m2,f8_e4m3
9 | --batch=option_set_int8
10 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/brgemm/test_brgemm_regression:
--------------------------------------------------------------------------------
1 | # Incorrect mask comparison in lazy hw config, need two problems to run back-to-back
2 | --reset --dt=bf16 128x448:448x32 42x608:608x32
3 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/brgemm/test_brgemm_smoke:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --match=.*64:64.* # Use shapes from ci with K=64 only
4 |
5 | --bs=16
6 | --dt=f32
7 | --bia_dt=undef,f32
8 | --attr-post-ops=,relu
9 | --batch=shapes_2d_no_tail_f32
10 |
11 | --dt=f16,bf16
12 | --batch=shapes_2d_no_tail_bf16
13 |
14 | --dt=u8:s8:f32,s8:s8:s32
15 | --attr-scales=,src:common:0.5+wei:per_oc
16 | --attr-zero-points=,src:common:128+dst:common:-1
17 | --batch=shapes_2d_no_tail_int8
18 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/concat/test_concat_ci:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --dtag=undef,any,abx,axb
4 | --sdt=f32,bf16,f16,s32,s8,u8
5 | --ddt=f32,bf16,f16,s32,s8,u8
6 | --stag=abx:abx,axb:axb
7 | --axis=1
8 | 2x16x3x4:2x16x3x4
9 | 3x5x5x17:3x4x5x17
10 | --axis=0
11 | 32x4x5x17:16x4x5x17
12 | 1x16x3x4:2x16x3x4
13 |
14 | --stag=abx:abx:abx,axb:axb:axb
15 | --axis=1
16 | --attr-scales=,msrc0:common:1.5,msrc0:common:1.5+msrc1:common:2.5
17 | 6x48x3x4x5:6x32x3x4x5:6x16x3x4x5
18 | 6x48x3x4x5:6x31x3x4x5:6x16x3x4x5
19 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/concat/test_concat_smoke:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --dtag=undef
4 | --sdt=f32,bf16,f16,s8,u8
5 | --ddt=f32,bf16,f16,s8,u8
6 | --stag=axb:axb
7 | --axis=1
8 | 2x16x3x4:2x16x3x4
9 | 3x5x5x17:3x4x5x17
10 |
11 | --stag=axb:axb:axb
12 | --axis=1
13 | --attr-scales=,msrc0:common:1.5+msrc1:common:2.5
14 | 6x48x3x4x5:6x32x3x4x5:6x16x3x4x5
15 | 6x48x3x4x5:6x31x3x4x5:6x16x3x4x5
16 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/harness_conv_arbitrary_dst:
--------------------------------------------------------------------------------
1 | --reset
2 | --mb=2,16
3 | --dir=FWD_B
4 |
5 | # mixed halfs
6 | --dt=bf16,bf16:bf16:f32,f16:f16:f32
7 | --batch=set_gpu
8 |
9 | #int8dst
10 | --dt=f16:f16:s8,f16:f16:u8,bf16:bf16:s8,bf16:bf16:u8,f32:f32:s8,f32:f32:u8
11 | --batch=set_gpu
12 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/harness_conv_auto:
--------------------------------------------------------------------------------
1 | # auto algo
2 | --reset --dt=f32 --alg=auto
3 | --dir=FWD_B --batch=shapes_auto
4 | --dir=BWD_D --batch=shapes_auto
5 | --dir=BWD_WB --batch=shapes_auto
6 | --dt=u8:s8:s8
7 | --dir=FWD_B --batch=shapes_auto
8 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/harness_conv_deepbench:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --dir=FWD_B,BWD_D,BWD_W
4 | --batch=shapes_deepbench_inference_device
5 | --batch=shapes_deepbench_inference_server
6 | --batch=shapes_deepbench_training
7 |
8 | --dir=FWD_B
9 | --attr-post-ops=relu
10 | --batch=shapes_deepbench_inference_device
11 | --batch=shapes_deepbench_inference_server
12 | --batch=shapes_deepbench_training
13 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/harness_conv_dilated_int8:
--------------------------------------------------------------------------------
1 | # Dilated Int8
2 | --reset
3 | --mb=2
4 | --dir=FWD_D
5 | --dt=u8:s8:u8,s8:s8:u8
6 | --batch=shapes_dilated_rfcn
7 | --match=.*fc6.* --batch=shapes_ssd_300_voc0712
8 |
9 | --reset
10 | --dir=FWD_D --mb=2
11 | --skip-impl=ref,x64:gemm
12 | --dt=u8:s8:s32
13 | --batch=set_dilated-conv_1st
14 | --batch=set_dilated-conv
15 | --batch=set_dilated-conv_3d
16 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/harness_conv_f32:
--------------------------------------------------------------------------------
1 | # f32
2 | --reset --dt=f32
3 | --mb=2 # for fwd and bwd_d reduce mb
4 | --dir=FWD_B --batch=set_conv_all
5 | --dir=BWD_D --batch=set_conv_all
6 | --dir=BWD_WB --batch=set_conv_all
7 | --mb=0 # for bwd_w use the actual mb for 1 topology
8 | --dir=BWD_WB --batch=shapes_resnet_50
9 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/harness_conv_saturation_int8:
--------------------------------------------------------------------------------
1 | # Test that saturation is handled properly (lightweight test)
2 | --reset
3 | --mb=2
4 | --dir=FWD_B
5 | --dt=u8:s8:u8,u8:s8:s8,u8:s8:s32
6 | --attr-scales=src:common:0.25+wei:common:0.5+dst:common:4294967295
7 | ic16oc16ih4oh4kh1ph0 # jit 1x1
8 | ic16oc16ih4oh4kh3ph0 # jit
9 | ic16oc16ih4oh7kh3ph3 # gemm impl
10 |
11 | --reset
12 | --dt=u8:s8:u8
13 | --attr-scales=src:common:0.25+wei:common:0.5+dst:common:4294967295
14 | --attr-post-ops=round
15 | ic16oc16_ih5oh5kh3ph1
16 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/harness_conv_tags:
--------------------------------------------------------------------------------
1 | # Test different combinations of memory format tags for input tensors
2 | --reset
3 | --mb=2
4 |
5 | # Source or destination tensor may be in plain layout
6 | --stag=any,abx,axb
7 | --dtag=any,abx,axb
8 |
9 | # Training: bf16 and f32
10 | --dt=bf16,f32
11 | --dir=FWD_B,BWD_D,BWD_WB
12 | --batch=shapes_basic
13 | # Inference: int8
14 | --dt=u8:s8:u8
15 | --dir=FWD_B
16 | --batch=shapes_basic
17 |
18 | # Winograd
19 | --alg=wino
20 | --match=.*k[dhw]3.*
21 | --dt=f32 # Training: f32 (no bf16 for Winograd)
22 | --dir=FWD_B,BWD_D,BWD_WB
23 | --batch=shapes_basic
24 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/perf_conv_bdw_1sock:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --dt=f32
4 | --dir=FWD_D,BWD_D,BWD_W
5 | --alg=direct
6 |
7 | --mb=0,20,40,80
8 | --batch=set_perf_cpu_large_mb
9 |
10 | --mb=1,4,8
11 | --batch=set_perf_cpu_all_mb
12 |
13 | --reset
14 | ## int8
15 | --dt=u8:s8:u8
16 | --dir=FWD_D
17 | --alg=direct
18 |
19 | --mb=0,20,40,80
20 | --batch=set_perf_cpu_large_mb
21 |
22 | --mb=1,4,8
23 | --batch=set_perf_cpu_all_mb
24 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/perf_conv_clx_1sock:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | # Direct
4 | ## f32
5 | --dt=f32
6 | --dir=FWD_D,BWD_D,BWD_W
7 | --alg=direct
8 |
9 | --mb=0,26,52,104
10 | --batch=set_perf_cpu_large_mb
11 |
12 | --mb=1,4,8
13 | --batch=set_perf_cpu_all_mb
14 |
15 | --reset
16 | ## int8
17 | --dt=u8:s8:u8
18 | --dir=FWD_D
19 | --alg=direct
20 |
21 | --mb=0,26,52,104
22 | --batch=set_perf_cpu_large_mb
23 |
24 | --mb=1,4,8
25 | --batch=set_perf_cpu_all_mb
26 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/perf_conv_skx_1sock:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | # Direct
4 | ## f32
5 | --dt=f32
6 | --dir=FWD_D,BWD_D,BWD_W
7 | --alg=direct
8 |
9 | --mb=0,16,32,64
10 | --batch=set_perf_cpu_large_mb
11 |
12 | --mb=1,4,8
13 | --batch=set_perf_cpu_all_mb
14 |
15 | --reset
16 | ## int8
17 | --dt=u8:s8:u8
18 | --dir=FWD_D
19 | --alg=direct
20 |
21 | --mb=0,16,32,64
22 | --batch=set_perf_cpu_large_mb
23 |
24 | --mb=1,4,8
25 | --batch=set_perf_cpu_all_mb
26 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/perf_conv_xe:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | # Forward
4 |
5 | # f32
6 | --dir=FWD_B
7 | --dt=f32
8 | --mb=16,64
9 | --batch=set_perf_gpu_large_mb
10 |
11 | --dir=FWD_B
12 | --dt=f32
13 | --mb=1
14 | --batch=set_perf_gpu_all_mb
15 |
16 | # f16
17 | --dt=f16
18 | --dir=FWD_B
19 |
20 | --mb=16,64
21 | --batch=set_perf_gpu_large_mb
22 |
23 | --mb=1
24 | --batch=set_perf_gpu_all_mb
25 |
26 | # Backward
27 |
28 | # f32
29 | --dt=f32
30 | --dir=BWD_D,BWD_W
31 | --mb=64
32 | --batch=set_perf_gpu_large_mb
33 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/perf_conv_xe_hp:
--------------------------------------------------------------------------------
1 | --batch=perf_conv_xe_lp
2 |
3 | --reset
4 |
5 | # Forward, bf16
6 |
7 | --dir=FWD_B
8 | --dt=bf16
9 | --mb=16,64
10 | --batch=set_perf_gpu_large_mb
11 |
12 | --dir=FWD_B
13 | --dt=bf16
14 | --mb=1
15 | --batch=set_perf_gpu_all_mb
16 |
17 | # Backward, bf16
18 |
19 | --dt=bf16
20 | --dir=BWD_D,BWD_W
21 | --mb=64
22 | --batch=set_perf_gpu_large_mb
23 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/perf_conv_xe_lp:
--------------------------------------------------------------------------------
1 | --batch=perf_conv_xe
2 |
3 | --reset
4 |
5 | # Forward, int8
6 | --dt=u8:s8:s8
7 | --dir=FWD_B
8 |
9 | --mb=16,64
10 | --batch=set_perf_gpu_large_mb
11 |
12 | --mb=1
13 | --batch=set_perf_gpu_all_mb
14 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/set_conv_3d:
--------------------------------------------------------------------------------
1 | # 3D Convolutions
2 | # ngroups = 1; ic = oc = 16
3 | # 1x1 Filter
4 | --batch=shapes_3d_1x1_unit-stride_no-padding
5 | --batch=shapes_3d_1x1_unit-stride_padding
6 | --batch=shapes_3d_1x1_strided_no-padding
7 | --batch=shapes_3d_1x1_strided_padding
8 | # N-sized Filter
9 | # ngroups = 1; ic = 1; oc = 16
10 | --batch=shapes_3d_1st_strided_padding
11 | # ngroups = 1; ic = oc = 16
12 | --batch=shapes_3d_unit-stride_no-padding
13 | --batch=shapes_3d_unit-stride_padding
14 | --batch=shapes_3d_strided_no-padding
15 | --batch=shapes_3d_strided_padding
16 | # Depth + Height
17 | --batch=shapes_3d_2d_strided_padding
18 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/set_conv_all:
--------------------------------------------------------------------------------
1 | --batch=shapes_1d
2 | --batch=shapes_1d_wavenet
3 | --batch=shapes_3d_unet
4 | --batch=shapes_alexnet
5 | --batch=shapes_vgg_19
6 | --batch=shapes_resnet_50
7 | --batch=shapes_googlenet_v1
8 | --batch=shapes_googlenet_v2
9 | --batch=shapes_googlenet_v3
10 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/set_dilated-conv:
--------------------------------------------------------------------------------
1 | # Dilated Convolution
2 | # 1D
3 | --batch=shapes_dilated_1d_strided_no-padding
4 | --batch=shapes_dilated_1d_unit-stride_no-padding
5 | --batch=shapes_dilated_1d_unit-stride_padding
6 | --batch=shapes_dilated_1d_strided_padding
7 | # 2D
8 | --batch=shapes_dilated_2d_strided_no-padding
9 | --batch=shapes_dilated_2d_strided_padding
10 | --batch=shapes_dilated_2d_unit-stride_no-padding
11 | --batch=shapes_dilated_2d_unit-stride_padding
12 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/set_dilated-conv_1st:
--------------------------------------------------------------------------------
1 | # Dilated Convolutions
2 | # ic = 1 for 1st convolution code-path
3 | --batch=shapes_dilated_1d_1st_strided_padding
4 | --batch=shapes_dilated_2d_1st_strided_padding
5 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/set_dilated-conv_3d:
--------------------------------------------------------------------------------
1 | # 3D Dilated Convolution
2 | # N-sized Filter
3 | --batch=shapes_dilated_3d_strided_no-padding
4 | --batch=shapes_dilated_3d_strided_padding
5 | --batch=shapes_dilated_3d_unit-stride_no-padding
6 | --batch=shapes_dilated_3d_unit-stride_padding
7 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/set_fastrcnn:
--------------------------------------------------------------------------------
1 | # FastRCNN
2 | --batch=shapes_fastrcnn_p1
3 | --batch=shapes_fastrcnn_p2
4 | #--batch=shapes_fastrcnn_p3 # no unique conv. shapes
5 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/set_gpu:
--------------------------------------------------------------------------------
1 | --batch=shapes_1d_wavenet
2 | --batch=shapes_alexnet
3 | --batch=shapes_resnet_50
4 | --batch=shapes_3d_gpu
5 | --batch=shapes_mobilenet_dw
6 | --batch=shapes_tails_gpu
7 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/set_maskrcnn:
--------------------------------------------------------------------------------
1 | # MaskRCNN
2 | --batch=shapes_maskrcnn_p1
3 | --batch=shapes_maskrcnn_p2
4 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/set_perf_cpu_all_mb:
--------------------------------------------------------------------------------
1 | --batch=set_perf_cpu_large_mb
2 | --batch=set_perf_cpu_small_mb
3 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/set_perf_cpu_inference_only:
--------------------------------------------------------------------------------
1 | --batch=shapes_ssd_resnet34_inference
2 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/set_perf_cpu_large_mb:
--------------------------------------------------------------------------------
1 | --batch=shapes_1d_wavenet
2 | --batch=shapes_alexnet
3 | --batch=shapes_densnet
4 | --batch=shapes_efficientdet
5 | --batch=shapes_googlenet_v1
6 | --batch=shapes_googlenet_v2
7 | --batch=shapes_googlenet_v3
8 | --batch=shapes_mobilenet
9 | --batch=shapes_mobilenet_dw
10 | --batch=shapes_resnet_50
11 | --batch=shapes_resnet_50_sparse
12 | --batch=shapes_ssd_300_voc0712
13 | --batch=shapes_ssd_mobilenet
14 | --batch=shapes_ssd_resnet34_training
15 | --batch=shapes_unet
16 | --batch=shapes_vgg_19
17 | --batch=shapes_yolov2
18 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/set_perf_cpu_small_mb:
--------------------------------------------------------------------------------
1 | --batch=set_maskrcnn
2 | --batch=set_fastrcnn
3 | --batch=shapes_3d_unet
4 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/set_perf_gpu_all_mb:
--------------------------------------------------------------------------------
1 | --batch=set_perf_gpu_large_mb
2 | --batch=set_perf_gpu_small_mb
3 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/set_perf_gpu_large_mb:
--------------------------------------------------------------------------------
1 | --batch=shapes_1d_wavenet
2 | --batch=shapes_a3c
3 | --batch=shapes_alexnet
4 | --batch=shapes_densnet
5 | --batch=shapes_ffn
6 | --batch=shapes_googlenet_v3
7 | --batch=shapes_mobilenet
8 | --batch=shapes_mobilenet_dw
9 | --batch=shapes_pointnet
10 | --batch=shapes_resnet_50
11 | --batch=shapes_resnet_50_sparse
12 | --batch=shapes_unet
13 | --batch=shapes_vgg_19
14 | --batch=shapes_yolov2
15 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/set_perf_gpu_small_mb:
--------------------------------------------------------------------------------
1 | --batch=set_maskrcnn
2 | --batch=set_fastrcnn
3 | --batch=shapes_3d_unet
4 | --batch=shapes_cosmictagger
5 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/set_topologies_inference_only:
--------------------------------------------------------------------------------
1 | --batch=shapes_ssd_resnet34_inference
2 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/shapes_3d_gpu:
--------------------------------------------------------------------------------
1 | #test cases for 3D convolution
2 |
3 | ic16oc16_ih13kh3ph1_iw50kw3pw1_id10kd3pd1_n"3d_conv_pad:1"
4 | ic64oc64_ih10kh3ph1_iw20kw3pw1_id15kd3pd1_n"3d_conv_pad:2"
5 | ic256oc256_ih7kh3ph1_iw9kw3pw1_id11kd3pd1_n"3d_conv_pad:3"
6 | ic256oc256_ih7kh1ph1_iw9kw1pw1_id11kd1pd1_n"3d_conv_pad:4"
7 | ic16oc16_ih13kh3ph0_iw50kw3pw0_id10kd3pd0_n"3d_conv:1"
8 | ic16oc16_ih13kh1ph0_iw50kw1pw0_id10kd1pd0_n"3d_conv:2"
9 | ic256oc256_ih7kh3ph0_iw9kw3pw0_id11kd3pd0_n"3d_conv:3"
10 | ic256oc256_ih7kh1ph0_iw9kw1pw0_id11kd1pd0_n"3d_conv:4"
11 |
12 | #mb1_ic16oc16_id1od2kd1sd1dd1pd0_ih2oh3kh2sh1dh1ph2_iw5ow3kw3sw2dw0pw0
13 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/shapes_a3c:
--------------------------------------------------------------------------------
1 | # A3C
2 |
3 | mb1_g1ic4oc16_ih84oh20kh8sh4dh0ph0_iw84ow20kw8sw4dw0pw0_n"a3c:conv1"
4 | mb1_g1ic16oc32_ih20oh9kh4sh2dh0ph0_iw20ow9kw4sw2dw0pw0_n"a3c:conv2"
5 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/shapes_alexnet:
--------------------------------------------------------------------------------
1 | # alexnet
2 |
3 | g1mb256ic3ih227iw227oc96oh55ow55kh11kw11sh4sw4ph0pw0n"alexnet:conv1"
4 | g2mb256ic96ih27oc256oh27kh5ph2n"alexnet:conv2"
5 | mb256ic256ih13oc384oh13kh3ph1n"alexnet:conv3"
6 | g2mb256ic384ih13oc384oh13kh3ph1n"alexnet:conv4"
7 | g2mb256ic384ih13oc256oh13kh3ph1n"alexnet:conv5"
8 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/shapes_auto:
--------------------------------------------------------------------------------
1 | mb2_ic3oc64_ih300kh3oh300ph1n"ssd_300_voc0712:conv1_1"
2 | mb9_ic3oc64_ih300kh3oh300ph1n"ssd_300_voc0712:conv1_1"
3 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/shapes_dilated_rfcn:
--------------------------------------------------------------------------------
1 | # Dilated Convolution from RFCN
2 |
3 | mb1ic512ih38iw57oc512oh38ow57kh3kw3ph3pw3dh2dw2n"dilated_rfcn"
4 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/shapes_ffn:
--------------------------------------------------------------------------------
1 | # FFN
2 |
3 | mb4ic2id33oc32od33kd3pd1n"ffn:0*2"
4 | mb4ic32id33oc1od33kd1pd0n"ffn:1*3"
5 | mb4ic32id33oc32od33kd3pd1n"ffn:2*69"
6 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/shapes_fused_large_src:
--------------------------------------------------------------------------------
1 | # large-mb, small-ic shape to exceed combined L2 cache and target jit impls
2 | mb448_ic16oc64_ih56oh56kh1sh1dh0ph0n"fused_large_src"
3 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/shapes_fused_mobilenet_stride_2:
--------------------------------------------------------------------------------
1 | ic32oc64_ih112oh112kh1sh1dh0ph0n"MobileNet_v1_fused_stride_2:conv1"
2 | ic128oc128_ih56oh56kh1sh1dh0ph0n"MobileNet_v1_fused_stride_2:conv2"
3 | ic256oc256_ih28oh28kh1sh1dh0ph0n"MobileNet_v1_fused_stride_2:conv3"
4 | ic512oc512_ih14oh14kh1sh1dh0ph0n"MobileNet_v1_fused_stride_2:conv4"
5 |
6 | ic16oc96_ih112oh112kh1sh1dh0ph0n"MobileNet_v2_fused_stride_2:conv1"
7 | ic24oc144_ih56oh56kh1sh1dh0ph0n"MobileNet_v2_fused_stride_2:conv2"
8 | ic64oc384_ih28oh28kh1sh1dh0ph0n"MobileNet_v2_fused_stride_2:conv3"
9 | ic96oc576_ih14oh14kh1sh1dh0ph0n"MobileNet_v2_fused_stride_2:conv4"
10 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/shapes_large_conv:
--------------------------------------------------------------------------------
1 | # Large iw
2 | mb1ic32oc1iw134217732kw3
3 | mb1ic1oc1iw4294967311kw3
4 |
5 | # Large ic
6 | mb1ic4294967311oc1iw1kw1
7 | mb1ic134217732oc1iw27kw3
8 |
9 | # Larg oc
10 | mb1ic1oc1342177321iw27kw3
11 | mb1ic1oc4294967311iw1kw1
12 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/shapes_large_padding:
--------------------------------------------------------------------------------
1 | ic16oc16_iw1kw3ow100pw100
2 | ic1oc64_ih55oh18kh30sh3ph29dh2
3 | ic1oc64ih7oh7kh3ph1sh1dh0
4 | ic1oc64iw100ow98kw5pw50sw2dw0
5 | ic2oc64ih30oh37kh14ph10sh1dh2
6 | ic3oc64id28od36kd3pd5sd1dd0
7 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/shapes_pointnet:
--------------------------------------------------------------------------------
1 | # PointNet
2 |
3 | mb10ic1088ih1iw15000oc512oh1ow15000kh1kw1ph0pw0n"pointnet:0*300"
4 | mb10ic128ih1iw15000oc1024oh1ow15000kh1kw1ph0pw0n"pointnet:1*900"
5 | mb10ic128ih1iw15000oc5oh1ow15000kh1kw1ph0pw0n"pointnet:2*300"
6 | mb10ic256ih1iw15000oc128oh1ow15000kh1kw1ph0pw0n"pointnet:3*300"
7 | mb10ic4ih1iw15000oc64oh1ow15000kh1kw1ph0pw0n"pointnet:4*500"
8 | mb10ic512ih1iw15000oc256oh1ow15000kh1kw1ph0pw0n"pointnet:5*300"
9 | mb10ic64ih1iw15000oc128oh1ow15000kh1kw1ph0pw0n"pointnet:6*900"
10 | mb10ic64ih1iw15000oc64oh1ow15000kh1kw1ph0pw0n"pointnet:7*900"
11 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/shapes_regression_gemm:
--------------------------------------------------------------------------------
1 | # ResNext50
2 | mb2_g32ic128oc128_ih56oh56kh3sh1dh0ph1_iw56ow56kw3sw1dw0pw1
3 |
4 | # Faster RCNN
5 | mb1_g64ic256oc256_ih240oh240kh3sh1dh0ph1_iw352ow352kw3sw1dw0pw1
6 |
7 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/shapes_segnet:
--------------------------------------------------------------------------------
1 | # Minibatch 4 is found in the original prototxt
2 | # Not clear if that's what should be used for perf measurements
3 |
4 | mb4ih240iw320ic64oc64oh240ow320kh7kw7ph3pw3sh1sw1n"conv2"
5 | mb4ih120iw160ic64oc64oh120ow160kh7kw7ph3pw3sh1sw1n"conv3"
6 | mb4ih60iw80ic64oc64oh60ow80kh7kw7ph3pw3sh1sw1n"conv4"
7 | mb4ih480iw640ic64oc64oh480ow640kh7kw7ph3pw3sh1sw1n"conv_decode1"
8 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/shapes_vgg_11:
--------------------------------------------------------------------------------
1 | # vgg_11
2 |
3 | mb64ic3ih100oc32oh100kh3ph1n"vgg_11:conv1_1"
4 | mb64ic32ih100oc64oh100kh3ph1n"vgg_11:conv1_2"
5 | mb64ic64ih50oc64oh50kh3ph1n"vgg_11:conv2_1"
6 | mb64ic64ih50oc128oh50kh3ph1n"vgg_11:conv2_2"
7 | mb64ic128ih25oc96oh25kh3ph1n"vgg_11:conv3_1"
8 | mb64ic96ih25oc192oh25kh3ph1n"vgg_11:conv3_2"
9 | mb64ic192ih13oc128oh13kh3ph1n"vgg_11:conv4_1"
10 | mb64ic128ih13oc256oh13kh3ph1n"vgg_11:conv4_2"
11 | mb64ic256ih7oc160oh7kh3ph1n"vgg_11:conv5_1"
12 | mb64ic160ih7oc320oh7kh3ph1n"vgg_11:conv5_2"
13 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/test_conv_3d:
--------------------------------------------------------------------------------
1 | # f32 3-D Convolutions
2 | --reset --dt=f32
3 | --mb=2
4 | --skip-impl=ref,x64:gemm # ! test jit version only
5 | --dir=FWD_B,BWD_D,BWD_WB
6 | --batch=shapes_3d
7 | --batch=set_conv_3d
8 |
9 | # i8 3-D Convolution
10 | --reset
11 | --skip-impl=ref,x64:gemm # ! test jit version only
12 | --mb=2
13 | --dt=u8:s8:s8,s8:s8:s32,u8:s8:s8
14 | --batch=shapes_3d
15 | --batch=set_conv_3d
16 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/test_conv_3d_f32_plain:
--------------------------------------------------------------------------------
1 | # f32 3-D Convolutions
2 | --reset --dt=f32
3 | --stag=axb --dtag=axb
4 | --mb=2
5 | --skip-impl=ref,x64:gemm # ! test jit version only
6 | --dir=FWD_B,BWD_D,BWD_WB
7 | --batch=shapes_3d
8 | --batch=set_conv_3d
9 |
10 | --stag=abx --dtag=abx
11 | --batch=shapes_3d
12 | --batch=set_conv_3d
13 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/test_conv_all_topologies:
--------------------------------------------------------------------------------
1 | # Test All Topologies with F32 Configuration
2 | --reset --dt=f32
3 | --skip-impl=ref
4 | --mb=2
5 | --dir=FWD_B,BWD_D,BWD_WB
6 | --batch=set_all_topologies
7 |
8 | --dir=FWD_B
9 | --batch=set_topologies_inference_only
10 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/test_conv_all_topologies_f32_plain:
--------------------------------------------------------------------------------
1 | # Test All Topologies with F32 Configuration
2 | --reset --dt=f32
3 | --stag=axb --dtag=axb
4 | --skip-impl=ref
5 | --mb=2
6 | --dir=FWD_B,BWD_D,BWD_WB
7 | --batch=set_all_topologies
8 |
9 | --dir=FWD_B
10 | --batch=set_topologies_inference_only
11 |
12 | --stag=abx --dtag=abx
13 | --dir=FWD_B,BWD_D,BWD_WB
14 | --batch=set_all_topologies
15 |
16 | --dir=FWD_B
17 | --batch=set_topologies_inference_only
18 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/test_conv_attrs:
--------------------------------------------------------------------------------
1 | # f32
2 | --reset
3 | --mb=2
4 | --dir=FWD_B
5 |
6 | --dt=f32
7 | --skip-impl=ref,x64:gemm # ! test jit version only
8 | --batch=option_set_combined_postops
9 | --skip-impl=
10 |
11 | --dt=f32,f16,bf16,u8:s8:u8
12 | --skip-impl=ref
13 | --batch=option_set_all_eltwise_postops
14 | --skip-impl=
15 |
16 | # f32_wino
17 | --reset
18 | --mb=2
19 | --dir=FWD_B
20 |
21 | --dt=f32
22 | --alg=wino
23 | --attr-post-ops=sum+relu
24 | --batch=shapes_resnet_50
25 |
26 | # Depthwise fusion
27 | --batch=harness_conv_fused_depthwise
28 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/test_conv_attrs_f32_plain:
--------------------------------------------------------------------------------
1 | # f32
2 | --reset --dt=f32
3 | --mb=2
4 | --stag=axb --dtag=axb
5 | --dir=FWD_B
6 |
7 | --skip-impl=ref,x64:gemm # ! test jit version only
8 | --batch=option_set_combined_postops
9 |
10 | --stag=abx --dtag=abx
11 | --batch=option_set_all_eltwise_postops
12 |
13 | --stag=axb --dtag=axb
14 | --skip-impl=ref
15 | --batch=option_set_all_eltwise_postops
16 |
17 | --stag=abx --dtag=abx
18 | --batch=option_set_all_eltwise_postops
19 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/test_conv_bfloat16_ymm:
--------------------------------------------------------------------------------
1 | # Currently, only gemm routine implements bf16_ymm hint
2 | # Therefore test bfloat16 GeMM specific functionality
3 |
4 | # global benchdnn knob, will not be reset again
5 | --cpu-isa-hints=prefer_ymm
6 | --reset
7 |
8 | --batch=test_conv_gemm_bfloat16
9 | --batch=test_conv_gemm_bfloat16_nxc
10 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/test_conv_depthwise:
--------------------------------------------------------------------------------
1 | # f32
2 | --reset
3 | --mb=2
4 | --dir=FWD_D,BWD_D,BWD_WB,FWD_I --batch=shapes_mobilenet_dw
5 | --dir=FWD_D,BWD_D,BWD_WB,FWD_I --batch=shapes_regression_dw
6 |
7 | # post-ops
8 | --dir=FWD_D
9 | --attr-post-ops=relu,sum,sum+relu+add:f32:per_tensor,add:f32:per_oc
10 | --batch=shapes_mobilenet_dw
11 |
12 | --reset --dt=f32
13 | --mb=2
14 | --skip-impl=ref,x64:gemm
15 | --dir=FWD_B,BWD_D,BWD_WB,FWD_I
16 | --batch=set_conv_dw
17 | --batch=shapes_dw_minibatch_2d-spatial
18 | --batch=shapes_dw_minibatch_channel_2d-spatial
19 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/test_conv_dilated:
--------------------------------------------------------------------------------
1 | # dilated f32
2 | --reset
3 | --mb=2
4 | --dt=f32
5 | --dir=FWD_B,BWD_D,BWD_WB
6 | --batch=shapes_dilated --batch=shapes_dilated_rfcn
7 | --match=.*fc6.* --batch=shapes_ssd_300_voc0712
8 |
9 | --reset --dt=f32
10 | --mb=2
11 | --skip-impl=ref,x64:gemm # ! test jit version only
12 | --dir=FWD_B,BWD_D,BWD_WB
13 | --batch=set_dilated-conv_1st
14 | --batch=set_dilated-conv
15 | --batch=set_dilated-conv_3d
16 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/test_conv_dt:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | # f32
4 | --batch=harness_conv_f32
5 |
6 | # tails
7 | --reset
8 | --dir=FWD_B,BWD_D,BWD_WB --batch=shapes_tails
9 |
10 | # bf16
11 | --batch=test_conv_bfloat16
12 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/test_conv_dt_plain:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | # f32
4 | --batch=harness_conv_f32_plain
5 |
6 | # tails
7 | --reset
8 | --skip-impl=ref
9 | --stag=axb --dtag=axb
10 | --dir=FWD_B,BWD_D,BWD_WB --batch=shapes_tails
11 | --stag=abx --dtag=abx
12 | --batch=shapes_tails
13 |
14 | # bf16
15 | --batch=test_conv_bfloat16_nxc
16 |
17 | # f16
18 | --batch=test_conv_float16_nxc
19 |
20 | # fp8
21 | --batch=test_conv_fp8_nxc
22 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/test_conv_fp4:
--------------------------------------------------------------------------------
1 | --reset --mb=1 --dt=f4_e2m1:f4_e2m1:bf16,f4_e3m0:f4_e3m0:bf16 --dir=fwd_d --batch=shapes_4bit
2 | --reset --mb=1 --dt=f16:f4_e2m1:f4_e2m1,f32:f4_e3m0:f4_e3m0 --dir=bwd_d --batch=shapes_4bit
3 | --reset --mb=1 --dt=f4_e2m1:f16:f4_e2m1,f4_e3m0:f32:f4_e3m0 --dir=bwd_w --batch=shapes_4bit
4 | --reset --mb=1 --dt=f4_e2m1,f4_e3m0 --dir=fwd_d --batch=shapes_4bit
5 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/test_conv_function:
--------------------------------------------------------------------------------
1 | # Test functional convolution features
2 |
3 | --reset
4 |
5 | # auto algo
6 | --batch=harness_conv_auto
7 |
8 | # memory-tags
9 | --batch=harness_conv_tags
10 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/test_conv_gemm_bfloat16:
--------------------------------------------------------------------------------
1 | # Test Bfloat16 GeMM specific functionality
2 | --reset
3 | --mb=2
4 | --skip-impl=ref
5 | --dir=FWD_B
6 | --dt=bf16:bf16:f32 --batch=shapes_gemm
7 |
8 | --dir=FWD_D
9 | --dt=bf16 --batch=shapes_gemm
10 |
11 | --dir=BWD_D
12 | --dt=f32:bf16:bf16 --batch=shapes_gemm
13 |
14 | --dir=BWD_WB
15 | --dt=bf16:f32:bf16 --batch=shapes_gemm
16 |
17 | --dir=FWD_B,BWD_D,BWD_W
18 | --dt=bf16
19 | --stag=abx --dtag=abx
20 | --batch=shapes_gemm
21 | --batch=shapes_3d_2d_strided_padding
22 | --batch=shapes_dilated_3d_strided_padding
23 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/test_conv_gemm_bfloat16_nxc:
--------------------------------------------------------------------------------
1 | # bfloat16
2 | --reset
3 | --mb=2
4 | --stag=axb --dtag=axb
5 | --skip-impl=ref
6 |
7 | --dir=FWD_B
8 | --dt=bf16:bf16:f32,bf16
9 | --batch=shapes_gemm
10 |
11 | --dir=BWD_D
12 | --dt=f32:bf16:bf16,bf16
13 | --batch=shapes_gemm
14 |
15 | --dir=BWD_WB
16 | --mb=0
17 | --dt=bf16:f32:bf16,bf16
18 | --batch=shapes_gemm
19 |
20 | # PostOps + Attributes
21 | --dir=FWD_B
22 | --dt=bf16:bf16:f32,bf16
23 | --mb=2
24 | --attr-post-ops=sum,relu,sum+tanh:1:1:2.5 --batch=shapes_gemm
25 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/test_conv_gemm_dt:
--------------------------------------------------------------------------------
1 | # Test GeMM specific functionality
2 |
3 | # f32
4 | --reset --dt=f32
5 | --mb=2 # for fwd and bwd_d reduce mb
6 | --dir=FWD_B,BWD_D,BWD_WB --batch=shapes_gemm
7 |
8 | --stag=abx --dtag=abx
9 | --batch=shapes_3d_2d_strided_padding
10 | --batch=shapes_dilated_3d_strided_padding
11 |
12 | # bf16
13 | --batch=test_conv_gemm_bfloat16
14 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/test_conv_gpu_ci:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --stag=any
4 | --dtag=any
5 | --batch=option_gpu_ci
6 |
7 | --stag=axb
8 | --dtag=axb
9 | --batch=option_gpu_ci
10 |
11 | --stag=any
12 | --dtag=axb
13 | --dir=FWD_B
14 | --dt=f64,f32,bf16,f16,s8,f8_e4m3,f8_e5m2
15 | --batch=shapes_ci_gpu
16 |
17 | --dir=BWD_D,BWD_W
18 | --dt=f64,f32,bf16
19 | --batch=shapes_ci_gpu
20 |
21 | --stag=axb
22 | --dtag=any
23 | --dir=FWD_B
24 | --dt=f64,f32,bf16,f16,s8,f8_e4m3,f8_e5m2
25 | --batch=shapes_ci_gpu
26 |
27 | --dir=BWD_D,BWD_W
28 | --dt=f64,f32,bf16
29 | --batch=shapes_ci_gpu
30 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/test_conv_int8:
--------------------------------------------------------------------------------
1 | # Test Int8 Convolutions
2 | --reset
3 |
4 | --batch=harness_conv_saturation_int8
5 |
6 | --batch=harness_conv_int8
7 |
8 | --batch=harness_conv_dilated_int8
9 |
10 | --batch=harness_conv_attrs_int8
11 |
12 | --batch=harness_conv_attrs_int8_asymmetric
13 |
14 | --batch=harness_conv_depthwise_int8
15 |
16 | --batch=harness_conv_zero_points
17 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/test_conv_regression:
--------------------------------------------------------------------------------
1 | # f32
2 | --reset --dt=f32
3 | --dir=FWD_B,BWD_D,BWD_WB
4 | --batch=shapes_regression_small_spatial
5 | --batch=shapes_regression_padding
6 | --batch=shapes_regression_gemm
7 | --batch=shapes_regression_1x1
8 | --stag=axb --dtag=axb --batch=shapes_regression_1x1
9 |
10 | --reset --dt=f32
11 | --dir=FWD_B
12 | --attr-post-ops=relu
13 | --batch=shapes_regression_small_spatial
14 | --batch=shapes_regression_padding
15 |
16 | #
17 | --batch=harness_conv_regression_general
18 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/test_conv_wino_f32:
--------------------------------------------------------------------------------
1 | # f32 wino
2 | --reset
3 | --dt=f32
4 | --alg=wino
5 | --match=.*kh3[^0-9].* # only 3x3 convolutions so far
6 | --mb=2 # for fwd and bwd_d reduce mb
7 | --dir=FWD_I,FWD_B,BWD_D,BWD_WB
8 | --batch=set_conv_all
9 | --batch=shapes_regression_padding
10 |
11 | --mb=0
12 | --batch=shapes_tails
13 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/conv/test_conv_wino_gpu:
--------------------------------------------------------------------------------
1 | # f16 wino
2 | --reset --dt=f16,f32 --stag=any,nhwc --alg=wino
3 | --match=.*[^k][^d][0-9]kh3[^0-9].* # only 3x3 convolutions so far
4 | --mb=2,32 # for fwd and bwd_d reduce mb
5 | --batch=set_conv_all
6 | --batch=shapes_regression_padding
7 | --batch=shapes_tails
8 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/deconv/harness_deconv_regression_general_f32:
--------------------------------------------------------------------------------
1 | # f32 regression : long accumulation chains
2 | --reset --dt=f32 --dir=bwd_w mb28_ic16oc16_id10od10kd3
3 |
4 | # test brgdeconv strided shape that caused segfault
5 | --reset
6 | --skip-impl=ref,x64:gemm
7 | --dt=f32
8 | --dir=fwd_b
9 | mb56_ic2oc3_ih6oh18kh3sh3dh0ph1_iw5ow15kw3sw3dw0pw1_n"2d_strided_with_bias"
10 |
11 | # test brgdeconv strided shape that kd/kh block != kd/kh
12 | --reset
13 | --skip-impl=ref,x64:gemm
14 | --dt=f32
15 | --dir=FWD_B
16 | mb1ic16ih16oc128oh32kh64sh2ph1
17 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/deconv/set_all:
--------------------------------------------------------------------------------
1 | --batch=shapes_1d
2 | --batch=shapes_3d
3 | --batch=shapes_2d
4 | --batch=shapes_dilated
5 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/deconv/shapes_1d:
--------------------------------------------------------------------------------
1 | # 1D
2 | ic8iw5oc8ow2kw3pw3dw2n"deconv1d:1"
3 | g1mb256oc3ow227ic96iw55kw11sw4pw0n"alexnet:deconv1"
4 | g2mb256oc96ow27ic256iw27kw5pw2n"alexnet:deconv2"
5 | mb256oc256ow13ic384iw13kw3pw1n"alexnet:deconv3"
6 | g1mb96ic64iw112oc3ow224kw7sw2pw3n"googlenet_v1:conv1/7x7_s2"
7 | mb1_g1oc3ic64_ow1030iw512kw7sw2dw0pw0_n"masknet_p1:deconv1"
8 | g1mb50ic256iw28oc512ow56kw1sw2pw0n"resnet_50:res3a_branch1"
9 | mb9_ic1oc1_ih1oh1kh1sh1dh0ph0_iw55ow55kw3sw1dw0pw1n"pytorch_unittest"
10 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/deconv/test_deconv_all:
--------------------------------------------------------------------------------
1 | # f32
2 | --reset
3 | --skip-impl=ref
4 | --mb=2
5 |
6 | --dir=FWD_B
7 | --attr-post-ops=,sum,linear:2:1,sum:1.5+add:f32:per_oc+relu
8 | --batch=set_all
9 |
10 | --dir=BWD_D,BWD_W,BWD_WB
11 | --attr-post-ops=
12 | --batch=set_all
13 |
14 | # int8
15 | --batch=test_deconv_int8
16 |
17 | # bf16
18 | --batch=test_deconv_bfloat16
19 |
20 | # f16
21 | --batch=test_deconv_float16_nxc
22 |
23 | # Regression
24 | --batch=harness_deconv_regression_general_f32
25 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/deconv/test_deconv_all_f32_nxc:
--------------------------------------------------------------------------------
1 | # f32 nxc
2 | --reset
3 | --skip-impl=ref
4 | --mb=2
5 | --stag=axb --dtag=axb
6 |
7 | --dir=FWD_B
8 | --attr-post-ops=,sum+prelu:per_oc,linear:2:1,sum:1.5+add:f32:per_oc+relu
9 | --batch=set_all
10 |
11 | --dir=BWD_D,BWD_W,BWD_WB
12 | --attr-post-ops=
13 | --batch=set_all
14 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/deconv/test_deconv_bfloat16:
--------------------------------------------------------------------------------
1 | # bf16
2 | --reset
3 | --skip-impl=ref
4 | --mb=2
5 |
6 | --dt=bf16
7 | --dir=FWD_B
8 | --attr-post-ops=,sum,linear:2:1,sum:1.5+add:f32:per_oc+relu
9 | --batch=set_all
10 |
11 | --dir=BWD_D,BWD_W,BWD_WB
12 | --attr-post-ops=
13 | --batch=set_all
14 |
15 | --dt=bf16:bf16:f32 --dir=FWD_B --batch=set_all
16 | --dt=f32:bf16:bf16 --dir=BWD_D --batch=set_all
17 | --dt=bf16:f32:bf16 --dir=BWD_WB --batch=set_all
18 |
19 | # Test Deconv w/bias through GeMM
20 | --reset
21 | --skip-impl=ref
22 | --mb=2
23 |
24 | --dt=bf16
25 | --dir=FWD_B,BWD_WB g16_ic32ih4iw8_oc64oh3ow8_kh3kw3sh1sw1ph0pw0n"gemm_shape"
26 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/deconv/test_deconv_bfloat16_ymm:
--------------------------------------------------------------------------------
1 | # bf16
2 |
3 | # global benchdnn knob, will not be reset again
4 | --cpu-isa-hints=prefer_ymm
5 | --batch=test_deconv_bfloat16
6 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/deconv/test_deconv_int8:
--------------------------------------------------------------------------------
1 | # int8
2 | --reset
3 | --skip-impl=ref
4 | --mb=2
5 | --dir=FWD_I,FWD_B
6 | --batch=set_all
7 | --batch=shapes_1x1
8 |
9 | --batch=harness_deconv_regression_general_int8
10 | --batch=harness_deconv_attrs_int8
11 | --batch=harness_deconv_attrs_int8_asymmetric
12 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/eltwise/harness_eltwise_large_buffer:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | # Use inplace to reduce memory consumption, most eltwise implementations do not
4 | # use different implementations for inplace. Use smallest supported data type to
5 | # minimize memory usage as well.
6 |
7 | --inplace=true
8 |
9 | --dt=u8
10 | --dir=FWD_D
11 | --attr-post-ops=,add:f32:per_tensor
12 | --batch=shapes_large_buffer
13 |
14 | --reset
15 | --dt=bf16
16 | --dir=BWD_D
17 | --batch=shapes_large_buffer
18 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/eltwise/harness_eltwise_regression:
--------------------------------------------------------------------------------
1 | # tag `a` regression check
2 | --reset
3 | --skip-impl=
4 |
5 | --dir=FWD_D
6 | --dt=f32
7 | --tag=a
8 | --attr-post-ops=mul:f32
9 | 16
10 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/eltwise/harness_eltwise_saturation:
--------------------------------------------------------------------------------
1 | # int linear saturation check
2 | --reset
3 |
4 | --dt=u8
5 | --tag=abx,axb
6 | --alpha=1,4294967295
7 | --beta=1,4294967295
8 | --alg=linear
9 | --batch=shapes_eltwise
10 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/eltwise/option_set_all_algs_int8:
--------------------------------------------------------------------------------
1 | # Integer algorithm coverage based on alpha and beta validity
2 | --alpha=0,0.25,-0.25 --beta=0
3 | --alg=relu
4 | --batch=shapes_eltwise
5 |
6 | --alpha=0,0.25,-0.25 --beta=0,0.25,-0.25
7 | --alg=linear
8 | --batch=shapes_eltwise
9 |
10 | --alpha=0,0.25,-0.25 --beta=0,0.25,-0.25
11 | --alg=clip
12 | --batch=shapes_eltwise
13 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/eltwise/option_set_all_algs_int8_ci:
--------------------------------------------------------------------------------
1 | # Integer algorithm coverage based on alpha and beta validity
2 | --alpha=-2 --beta=0
3 | --alg=relu
4 | --batch=shapes_ci
5 |
6 | --alpha=1 --beta=2
7 | --alg=linear
8 | --batch=shapes_ci
9 |
10 | --alpha=-2 --beta=3
11 | --alg=clip
12 | --batch=shapes_ci
13 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/eltwise/shapes_ci:
--------------------------------------------------------------------------------
1 | 16x16x2x1_n"eltwise_ci_2d:0" # Used in smoke validation, don't change the name
2 | 3x15x2x5x3_n"eltwise_ci_3d:0"
3 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/eltwise/shapes_eltwise:
--------------------------------------------------------------------------------
1 | 5x16x3
2 | 16x64x1x1
3 | 3x7x3x2
4 | 2x32x3x2
5 | 32x5x2x3
6 | 2x16x5x2x3
7 | 3x17x2x5x3
8 |
9 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/eltwise/shapes_large_buffer:
--------------------------------------------------------------------------------
1 | # Test dimensions with 2 ^ 32 + 1 size to catch integer overflow
2 | 4294967297n"large_dim1"
3 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/eltwise/test_eltwise_bfloat16:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --inplace=true,false
4 | --dt=bf16
5 | --tag=abx,axb,aBx8b,aBx16b,ABx16a16b
6 |
7 | --dir=FWD_D,BWD_D
8 | --batch=option_set_all_algs
9 |
10 | --dir=FWD_D
11 | --attr-post-ops=add:f32+mul:f32:per_oc
12 | --batch=option_set_all_algs
13 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/eltwise/test_eltwise_ci:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --inplace=true,false
4 |
5 | --dt=f32,bf16,f16,f8_e5m2,f8_e4m3
6 | --tag=abx,axb
7 | --dir=FWD_D
8 | --attr-post-ops=,mul:s8:per_oc
9 | --batch=option_set_all_algs_ci
10 | --dir=BWD_D
11 | --attr-post-ops=
12 | --batch=option_set_all_algs_ci
13 |
14 | --dir=FWD_I
15 | --dt=s32,s8,u8
16 | --attr-post-ops=,mul:f32
17 | --batch=option_set_all_algs_int8_ci
18 |
19 |
20 | --dt=f64
21 | --dir=BWD_D,FWD_I
22 | --attr-post-ops=
23 | --batch=option_set_all_algs_ci
24 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/eltwise/test_eltwise_float16:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --inplace=true,false
4 | --dt=f16
5 | --tag=abx,axb
6 |
7 | --dir=FWD_D,BWD_D
8 | --batch=option_set_all_algs
9 |
10 | --dir=FWD_D
11 | --attr-post-ops=add:f32+mul:f32:per_oc
12 | --batch=option_set_all_algs
13 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/eltwise/test_eltwise_float8:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --inplace=true,false
4 | --dt=f8_e5m2,f8_e4m3
5 | --tag=abx,axb
6 |
7 | --dir=FWD_D,BWD_D
8 | --batch=option_set_all_algs
9 |
10 | --dir=FWD_D
11 | --attr-post-ops=add:f32+mul:f32:per_oc
12 | --batch=option_set_all_algs
13 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/eltwise/test_eltwise_smoke:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --match=.*eltwise_ci_2d.* # Use 2d problems only from shapes_ci
4 | --inplace=false
5 |
6 | --dt=f32,bf16,f16
7 | --tag=axb
8 | --dir=FWD_D,BWD_D
9 | --batch=option_set_all_algs_ci
10 |
11 | --dir=FWD_I
12 | --dt=s8,u8
13 | --batch=option_set_all_algs_int8_ci
14 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/gnorm/shapes_all:
--------------------------------------------------------------------------------
1 | # Instance normalization
2 | # Unet-3D (id is divided by 10 for the first 2 cases)
3 | g32mb2ic32id16ih224iw224
4 | g256mb6ic256id2ih28iw28
5 | g320mb2ic320id10ih14iw14
6 | g128mb2ic128id6ih7iw7
7 |
8 | # Group normalization
9 | g1mb2ic2iw40
10 | g2mb2ic8ih30iw40
11 | g5mb2ic10id9ih10iw10
12 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/gnorm/shapes_ci:
--------------------------------------------------------------------------------
1 | # Instance normalization
2 | g2mb2ic2iw4
3 | g7mb2ic7ih3iw4
4 | g5mb2ic5id9ih1iw10
5 |
6 | # Group normalization
7 | g1mb2ic2iw4
8 | g2mb2ic8ih3iw4
9 | g5mb2ic10id9ih1iw10
10 | g5mb2ic40id2ih1iw5
11 | g5mb2ic80id2ih1iw5
12 | g1mb1ic128ih2iw2
13 | g2mb2ic30ih1iw1
14 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/graph/complex_fusion/harness_mlp_all:
--------------------------------------------------------------------------------
1 | --batch=complex_fusion/harness_mlp_ci
2 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/graph/complex_fusion/harness_mlp_ci:
--------------------------------------------------------------------------------
1 | --reset --dt=bf16,f16 --case=complex_fusion/mlp/gated-mlp-f32.json
2 |
3 | # WA1: use smaller problem to pass correctness check for f32 on pvc.
4 | # WA2: use subtract binary to avoid precision issue for f32 on xe-lpg.
5 | --reset --in-shapes=0:1x128+1:128x256+4:128x256+13:256x128 --op-kind=12:Subtract --case=complex_fusion/mlp/gated-mlp-f32.json
6 |
7 | # f16-int4 case
8 | --reset --case=complex_fusion/mlp/gated-mlp-int4.json
9 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/graph/pattern/harness_bf16_ci:
--------------------------------------------------------------------------------
1 | --reset --dt=bf16 --case=pattern/f32/conv_post_ops_fusion.json
2 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/graph/pattern/harness_f16_ci:
--------------------------------------------------------------------------------
1 | --reset --dt=f16 --case=pattern/f32/conv_post_ops_fusion.json
2 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/graph/pattern/harness_f32_ci:
--------------------------------------------------------------------------------
1 | --reset --in-shapes=0:2x64x112x112+1:32x64x2x2 --case=pattern/f32/conv_post_ops_fusion.json
2 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/graph/pattern/harness_f8_all:
--------------------------------------------------------------------------------
1 | --reset --case=pattern/f8/f8_conv_add_add_fusion.json
2 | --reset --case=pattern/f8/f8_conv_fwd.json
3 | --reset --case=pattern/f8/f8_conv_post_ops_fusion.json
4 | --reset --case=pattern/f8/f8_conv_post_ops_int8_add_fusion.json
5 | --reset --case=pattern/f8/f8_conv_bias_relu_fusion.json
6 | --reset --case=pattern/f8/f8_matmul.json
7 | --reset --case=pattern/f8/f8_bf16_matmul_add_fusion.json
8 | --reset --case=pattern/f8/f8_bf16_matmul_sum_add_mul_relu.json
9 | --reset --case=pattern/f8/f8_matmul_sum_add_mul_relu.json
10 | --reset --case=pattern/f8/f8_f32_matmul_mul_add_fusion.json
11 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/graph/pattern/harness_f8_ci:
--------------------------------------------------------------------------------
1 | --reset --case=pattern/f8/f8_conv_fwd.json
2 | --reset --case=pattern/f8/f8_matmul.json
3 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/graph/pattern/harness_int8_ci:
--------------------------------------------------------------------------------
1 | --reset --mb=2 --case=pattern/int8/int8_conv_post_ops_fusion.json
2 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/graph/test_graph_all:
--------------------------------------------------------------------------------
1 | --batch=test_graph_f32
2 | --batch=test_graph_bf16
3 | --batch=test_graph_f16
4 | --batch=test_graph_int8
5 | --batch=test_graph_f8
6 | --batch=test_graph_fusions
7 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/graph/test_graph_bf16:
--------------------------------------------------------------------------------
1 | --batch=op/harness_bf16_all
2 | --batch=pattern/harness_bf16_all
3 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/graph/test_graph_ci:
--------------------------------------------------------------------------------
1 | --batch=op/harness_f32_ci
2 | --batch=op/harness_f16_ci
3 | --batch=op/harness_bf16_ci
4 | --batch=pattern/harness_f32_ci
5 | --batch=pattern/harness_f16_ci
6 | --batch=pattern/harness_bf16_ci
7 | --batch=pattern/harness_int8_ci
8 | --batch=pattern/harness_f8_ci
9 | --batch=complex_fusion/harness_mha_ci
10 | --batch=complex_fusion/harness_mlp_ci
11 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/graph/test_graph_f16:
--------------------------------------------------------------------------------
1 | --batch=op/harness_f16_all
2 | --batch=pattern/harness_f16_all
3 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/graph/test_graph_f32:
--------------------------------------------------------------------------------
1 | --batch=op/harness_f32_all
2 | --batch=pattern/harness_f32_all
3 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/graph/test_graph_f8:
--------------------------------------------------------------------------------
1 | --batch=pattern/harness_f8_all
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/graph/test_graph_fusions:
--------------------------------------------------------------------------------
1 | --batch=complex_fusion/harness_mha_all
2 | --batch=complex_fusion/harness_mlp_all
3 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/graph/test_graph_fusions_gpu:
--------------------------------------------------------------------------------
1 | --batch=complex_fusion/harness_mha_all
2 | --batch=complex_fusion/harness_mha_ci
3 | --batch=complex_fusion/harness_mlp_all
4 | --batch=complex_fusion/harness_mlp_ci
5 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/graph/test_graph_int8:
--------------------------------------------------------------------------------
1 | --batch=pattern/harness_int8_all
2 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/graph/test_graph_op_gpu:
--------------------------------------------------------------------------------
1 | --batch=op/harness_f32_ci
2 | --batch=op/harness_f16_ci
3 | --batch=op/harness_bf16_ci
4 |
5 | --batch=op/harness_f32_all
6 | --batch=op/harness_f16_all
7 | --batch=op/harness_bf16_all
8 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/graph/test_graph_pattern_gpu:
--------------------------------------------------------------------------------
1 | --batch=pattern/harness_f32_ci
2 | --batch=pattern/harness_f16_ci
3 | --batch=pattern/harness_bf16_ci
4 | --batch=pattern/harness_int8_ci
5 | --batch=pattern/harness_f8_ci
6 |
7 | --batch=pattern/harness_f32_all
8 | --batch=pattern/harness_f16_all
9 | --batch=pattern/harness_bf16_all
10 | --batch=pattern/harness_int8_all
11 | --batch=pattern/harness_f8_all
12 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/harness_ip_gpt-j_2016-32_inf_lb_bfloat16:
--------------------------------------------------------------------------------
1 | # GPT-J
2 | # Input token size = 2016
3 | # Output token size = 32
4 | # Batch Size = 12
5 | # Num Beams = 4
6 | # M = num_beams * batch_size = 48
7 |
8 | --reset
9 | --dir=FWD_I
10 | --dt=bf16
11 |
12 | mb96768ic4096oc4096_n"gptj:gemm0*112"
13 | mb96768ic4096oc16384_n"gptj:gemm3*28"
14 | mb96768ic16384oc4096_n"gptj:gemm4*28"
15 | mb96768ic4096oc50400_n"gptj:gemm5*1"
16 | mb48ic4096oc4096_n"gptj:gemm6*3472"
17 | mb48ic4096oc16384_n"gptj:gemm9*868"
18 | mb48ic16384oc4096_n"gptj:gemm10*868"
19 | mb48ic4096oc50400_n"gptj:gemm11*31"
20 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/harness_ip_gpt-j_2016-32_inf_lb_f32:
--------------------------------------------------------------------------------
1 | # GPT-J
2 | # Input token size = 2016
3 | # Output token size = 32
4 | # Batch Size = 6
5 | # Num Beams = 4
6 | # M = num_beams * batch_size = 24
7 |
8 | --reset
9 | --dir=FWD_I
10 | --dt=f32
11 |
12 | mb48384ic4096oc4096_n"gptj:gemm0*112"
13 | mb48384ic4096oc16384_n"gptj:gemm3*28"
14 | mb48384ic16384oc4096_n"gptj:gemm4*28"
15 | mb48384ic4096oc50400_n"gptj:gemm5*1"
16 | mb24ic4096oc4096_n"gptj:gemm6*3472"
17 | mb24ic4096oc16384_n"gptj:gemm9*868"
18 | mb24ic16384oc4096_n"gptj:gemm10*868"
19 | mb24ic4096oc50400_n"gptj:gemm11*31"
20 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/harness_ip_gpt-j_2016-32_inf_lb_float16:
--------------------------------------------------------------------------------
1 | # GPT-J
2 | # Input token size = 2016
3 | # Output token size = 32
4 | # Batch Size = 12
5 | # Num Beams = 4
6 | # M = num_beams * batch_size = 48
7 |
8 | --reset
9 | --dir=FWD_I
10 | --dt=f16
11 |
12 | mb96768ic4096oc4096_n"gptj:gemm0*112"
13 | mb96768ic4096oc16384_n"gptj:gemm3*28"
14 | mb96768ic16384oc4096_n"gptj:gemm4*28"
15 | mb96768ic4096oc50400_n"gptj:gemm5*1"
16 | mb48ic4096oc4096_n"gptj:gemm6*3472"
17 | mb48ic4096oc16384_n"gptj:gemm9*868"
18 | mb48ic16384oc4096_n"gptj:gemm10*868"
19 | mb48ic4096oc50400_n"gptj:gemm11*31"
20 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/harness_ip_gpt-j_2016-32_inf_sb_bfloat16:
--------------------------------------------------------------------------------
1 | # GPT-J
2 | # Input token size = 2016
3 | # Output token size = 32
4 | # Batch Size = 1
5 | # Num Beams = 4
6 | # M = num_beams * batch_size = 4
7 |
8 | --reset
9 | --dir=FWD_I
10 | --dt=bf16
11 |
12 | mb8064ic4096oc4096_n"gptj:gemm0*112"
13 | mb8064ic4096oc16384_n"gptj:gemm3*28"
14 | mb8064ic16384oc4096_n"gptj:gemm4*28"
15 | mb8064ic4096oc50400_n"gptj:gemm5*1"
16 | mb4ic4096oc4096_n"gptj:gemm6*3472"
17 | mb4ic4096oc16384_n"gptj:gemm9*868"
18 | mb4ic16384oc4096_n"gptj:gemm10*868"
19 | mb4ic4096oc50400_n"gptj:gemm11*31"
20 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/harness_ip_gpt-j_2016-32_inf_sb_f32:
--------------------------------------------------------------------------------
1 | # GPT-J
2 | # Input token size = 2016
3 | # Output token size = 32
4 | # Batch Size = 1
5 | # Num Beams = 4
6 | # M = num_beams * batch_size = 4
7 |
8 | --reset
9 | --dir=FWD_I
10 | --dt=f32
11 |
12 | mb8064ic4096oc4096_n"gptj:gemm0*112"
13 | mb8064ic4096oc16384_n"gptj:gemm3*28"
14 | mb8064ic16384oc4096_n"gptj:gemm4*28"
15 | mb8064ic4096oc50400_n"gptj:gemm5*1"
16 | mb4ic4096oc4096_n"gptj:gemm6*3472"
17 | mb4ic4096oc16384_n"gptj:gemm9*868"
18 | mb4ic16384oc4096_n"gptj:gemm10*868"
19 | mb4ic4096oc50400_n"gptj:gemm11*31"
20 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/harness_ip_gpt-j_2016-32_inf_sb_float16:
--------------------------------------------------------------------------------
1 | # GPT-J
2 | # Input token size = 2016
3 | # Output token size = 32
4 | # Batch Size = 1
5 | # Num Beams = 4
6 | # M = num_beams * batch_size = 4
7 |
8 | --reset
9 | --dir=FWD_I
10 | --dt=f16
11 |
12 | mb8064ic4096oc4096_n"gptj:gemm0*112"
13 | mb8064ic4096oc16384_n"gptj:gemm3*28"
14 | mb8064ic16384oc4096_n"gptj:gemm4*28"
15 | mb8064ic4096oc50400_n"gptj:gemm5*1"
16 | mb4ic4096oc4096_n"gptj:gemm6*3472"
17 | mb4ic4096oc16384_n"gptj:gemm9*868"
18 | mb4ic16384oc4096_n"gptj:gemm10*868"
19 | mb4ic4096oc50400_n"gptj:gemm11*31"
20 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/harness_ip_gpt-j_32-32_inf_lb_bfloat16:
--------------------------------------------------------------------------------
1 | # GPT-J
2 | # Input token size = 32
3 | # Output token size = 32
4 | # Batch Size = 428
5 | # Num Beams = 4
6 | # M = num_beams * batch_size = 1712
7 |
8 | --reset
9 | --dir=FWD_I
10 | --dt=bf16
11 |
12 | mb54784ic4096oc4096_n"gptj:gemm0*112"
13 | mb54784ic4096oc16384_n"gptj:gemm3*28"
14 | mb54784ic16384oc4096_n"gptj:gemm4*28"
15 | mb54784ic4096oc50400_n"gptj:gemm5*1"
16 | mb1712ic4096oc4096_n"gptj:gemm6*3472"
17 | mb1712ic4096oc16384_n"gptj:gemm9*868"
18 | mb1712ic16384oc4096_n"gptj:gemm10*868"
19 | mb1712ic4096oc50400_n"gptj:gemm11*31"
20 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/harness_ip_gpt-j_32-32_inf_lb_f32:
--------------------------------------------------------------------------------
1 | # GPT-J
2 | # Input token size = 32
3 | # Output token size = 32
4 | # Batch Size = 214
5 | # Num Beams = 4
6 | # M = num_beams * batch_size = 856
7 |
8 | --reset
9 | --dir=FWD_I
10 | --dt=f32
11 |
12 | mb27392ic4096oc4096_n"gptj:gemm0*112"
13 | mb27392ic4096oc16384_n"gptj:gemm3*28"
14 | mb27392ic16384oc4096_n"gptj:gemm4*28"
15 | mb27392ic4096oc50400_n"gptj:gemm5*1"
16 | mb856ic4096oc4096_n"gptj:gemm6*3472"
17 | mb856ic4096oc16384_n"gptj:gemm9*868"
18 | mb856ic16384oc4096_n"gptj:gemm10*868"
19 | mb856ic4096oc50400_n"gptj:gemm11*31"
20 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/harness_ip_gpt-j_32-32_inf_lb_float16:
--------------------------------------------------------------------------------
1 | # GPT-J
2 | # Input token size = 32
3 | # Output token size = 32
4 | # Batch Size = 428
5 | # Num Beams = 4
6 | # M = num_beams * batch_size = 1712
7 |
8 | --reset
9 | --dir=FWD_I
10 | --dt=f16
11 |
12 | mb54784ic4096oc4096_n"gptj:gemm0*112"
13 | mb54784ic4096oc16384_n"gptj:gemm3*28"
14 | mb54784ic16384oc4096_n"gptj:gemm4*28"
15 | mb54784ic4096oc50400_n"gptj:gemm5*1"
16 | mb1712ic4096oc4096_n"gptj:gemm6*3472"
17 | mb1712ic4096oc16384_n"gptj:gemm9*868"
18 | mb1712ic16384oc4096_n"gptj:gemm10*868"
19 | mb1712ic4096oc50400_n"gptj:gemm11*31"
20 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/harness_ip_gpt-j_32-32_inf_sb_bfloat16:
--------------------------------------------------------------------------------
1 | # GPT-J
2 | # Input token size = 32
3 | # Output token size = 32
4 | # Batch Size = 1
5 | # Num Beams = 4
6 | # M = num_beams * batch_size = 4
7 |
8 | --reset
9 | --dir=FWD_I
10 | --dt=bf16
11 |
12 | mb128ic4096oc4096_n"gptj:gemm0*112"
13 | mb128ic4096oc16384_n"gptj:gemm3*28"
14 | mb128ic16384oc4096_n"gptj:gemm4*28"
15 | mb128ic4096oc50400_n"gptj:gemm5*1"
16 | mb4ic4096oc4096_n"gptj:gemm6*3472"
17 | mb4ic4096oc16384_n"gptj:gemm9*868"
18 | mb4ic16384oc4096_n"gptj:gemm10*868"
19 | mb4ic4096oc50400_n"gptj:gemm11*31"
20 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/harness_ip_gpt-j_32-32_inf_sb_f32:
--------------------------------------------------------------------------------
1 | # GPT-J
2 | # Input token size = 32
3 | # Output token size = 32
4 | # Batch Size = 1
5 | # Num Beams = 4
6 | # M = num_beams * batch_size = 4
7 |
8 | --reset
9 | --dir=FWD_I
10 | --dt=f32
11 |
12 | mb128ic4096oc4096_n"gptj:gemm0*112"
13 | mb128ic4096oc16384_n"gptj:gemm3*28"
14 | mb128ic16384oc4096_n"gptj:gemm4*28"
15 | mb128ic4096oc50400_n"gptj:gemm5*1"
16 | mb4ic4096oc4096_n"gptj:gemm6*3472"
17 | mb4ic4096oc16384_n"gptj:gemm9*868"
18 | mb4ic16384oc4096_n"gptj:gemm10*868"
19 | mb4ic4096oc50400_n"gptj:gemm11*31"
20 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/harness_ip_gpt-j_32-32_inf_sb_float16:
--------------------------------------------------------------------------------
1 | # GPT-J
2 | # Input token size = 32
3 | # Output token size = 32
4 | # Batch Size = 1
5 | # Num Beams = 4
6 | # M = num_beams * batch_size = 4
7 |
8 | --reset
9 | --dir=FWD_I
10 | --dt=f16
11 |
12 | mb128ic4096oc4096_n"gptj:gemm0*112"
13 | mb128ic4096oc16384_n"gptj:gemm3*28"
14 | mb128ic16384oc4096_n"gptj:gemm4*28"
15 | mb128ic4096oc50400_n"gptj:gemm5*1"
16 | mb4ic4096oc4096_n"gptj:gemm6*3472"
17 | mb4ic4096oc16384_n"gptj:gemm9*868"
18 | mb4ic16384oc4096_n"gptj:gemm10*868"
19 | mb4ic4096oc50400_n"gptj:gemm11*31"
20 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/harness_ip_regression:
--------------------------------------------------------------------------------
1 | # repeated sum with varying scale
2 | --reset --attr-post-ops=sum+relu+sum:2 ic64oc64_n"multisum"
3 |
4 | # large oc to trigger use_buffer_b with tail
5 | --reset --dir=BWD_W mb32ic16oc2049_n"large_oc_for_use_buffer_b_with_tail"
6 |
7 | # f16->s8 is not supported by reference, gemm based ip handles this case,
8 | # but it requires src and weights to have compatible tags.
9 | # This test case makes sure ip chooses weights tag correctly for plain src tag
10 | # when shapes contain ones.
11 | --reset --dt=f16:f16:s8 --stag=abcd mb128ic768ih1oc768_n"f16_s8_plain_src_tag"
12 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/harness_ip_sanitizers:
--------------------------------------------------------------------------------
1 | # shapes to help sanitizers catch bugs
2 |
3 | --reset
4 | --dt=f32,bf16,s8:s8:s8
5 | --dir=FWD_B
6 |
7 | # mb * ic will overflow for int data type in below shape
8 | mb2613184_ic1536oc16_n"huge_mb_ic"
9 |
10 | --dt=f32,bf16
11 | --dir=BWD_D,BWD_WB
12 | mb2613184_ic1536oc16_n"huge_mb_ic"
13 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/harness_ip_saturation:
--------------------------------------------------------------------------------
1 | --reset
2 | --dir=FWD_D
3 | --dt=u8:s8:s32,u8:s8:s8,u8:s8:u8
4 | --attr-scales=wei:common:4294967295
5 | ic16oc16ih1
6 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/harness_ip_tag_gpu:
--------------------------------------------------------------------------------
1 | --reset
2 | --attr-scratchpad=user
3 | --dt=f32
4 | --dir=FWD_B,FWD_I
5 | --dtag=abx
6 | --mb=16,17
7 |
8 | --wtag=ABx16a16b
9 | ic2048ih2id2iw1oc1000
10 |
11 | --stag=aBx16b
12 | --wtag=any
13 | ic2048id2ih2iw2oc1000
14 |
15 | --stag=ABx16a16b,ABcd8a8b,aBx16b,aBx8b
16 | --attr-post-ops=sum:0.5+relu:0.5
17 | ic2048ih2iw2oc1000
18 |
19 | --stag=aBx8b,aBx4b
20 | ic2048iw2oc1000
21 |
22 | --stag=aBx16b
23 | --attr-post-ops=linear:0.5:1.5:2
24 | ic2048id2ih2iw2oc1000
25 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/option_set_fwks_llm_gpu:
--------------------------------------------------------------------------------
1 | --reset --allow-enum-tags-only=0 --dir=FWD_I --dt=f16:f16:f32 --bia-dt=undef --stag=abcd --wtag=any --dtag=ab mb1024ic896ih1iw1oc151936_n"a12f80214bc2625a81ecaa173e76b977*2"
2 | --reset --allow-enum-tags-only=0 --dir=FWD_I --dt=f16:f16:f32 --bia-dt=undef --stag=abcd --wtag=any --dtag=ab mb32ic896ih1iw1oc151936_n"a72ef24774e776b6f286de263d897013*2"
3 | --reset --allow-enum-tags-only=0 --dir=FWD_I --dt=f16:f16:f32 --bia-dt=undef --stag=abcd --wtag=any --dtag=ab mb8ic896ih1iw1oc151936_n"fed31de8b7be65676854064135d375c7*508"
4 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/perf_ip_cpu:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | # Forward
4 | --dir=FWD_B
5 | --dt=f32,u8:s8:s32,s8:s8:s8
6 | --mb=1,64,256,1024,2048
7 | --batch=set_topologies
8 |
9 | # Backward
10 | --dt=f32
11 | --dir=BWD_D,BWD_WB
12 | --mb=64,256,1024,2048
13 | --batch=set_topologies
14 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/perf_ip_inference_sb:
--------------------------------------------------------------------------------
1 | # inference_sb -- inference with small batch size
2 | # f32
3 | --reset
4 | --dir=FWD_B
5 | --mb=1
6 | --batch=shapes_gnmt
7 | --batch=shapes_wd
8 | --batch=shapes_resnet_50
9 | --batch=shapes_resnet_50_sparse
10 | --batch=shapes_googlenet_v1
11 | --batch=shapes_googlenet_v3
12 | --batch=shapes_vgg16
13 | --batch=shapes_ncf
14 | --batch=shapes_alexnet
15 | --batch=shapes_bert
16 | --batch=shapes_bert_large
17 | --mb=0 --batch=shapes_rnn_t
18 | --mb=128 --batch=shapes_dlrm
19 | --mb=0 --batch=shapes_transformer_lt
20 | --mb=0 --batch=shapes_dien_sb
21 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/perf_ip_knx:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | # Forward
4 | --dir=FWD_B
5 | --dt=f32
6 | --mb=1,64,256,1024,2048
7 | --batch=set_topologies
8 |
9 | # Backward
10 | --dt=f32
11 | --dir=BWD_D,BWD_WB
12 | --mb=64,256,1024,2048
13 | --batch=set_topologies
14 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/perf_ip_xe:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | # Forward
4 |
5 | # f32
6 | --dir=FWD_B
7 | --dt=f32
8 |
9 | --mb=1,32,64,128
10 | --batch=set_gpu
11 |
12 | --mb=0
13 | --batch=shapes_1d
14 |
15 | # f16
16 | --dir=FWD_B
17 | --dt=f16
18 |
19 | --mb=1,32,64,128
20 | --batch=set_gpu
21 |
22 | --mb=0
23 | --batch=shapes_1d
24 |
25 | # Backward
26 |
27 | # f32
28 | --dir=BWD_D,BWD_WB
29 | --dt=f32
30 |
31 | --mb=32,64,128
32 | --batch=set_gpu
33 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/perf_ip_xe_hp:
--------------------------------------------------------------------------------
1 | --batch=perf_ip_xe_lp
2 |
3 | --reset
4 |
5 | # Forward, bf16
6 |
7 | --dir=FWD_B
8 | --dt=bf16
9 |
10 | --mb=1,32,64,128
11 | --batch=set_gpu
12 |
13 | --mb=0
14 | --batch=shapes_1d
15 |
16 | # Backward, bf16
17 |
18 | --dir=BWD_D,BWD_WB
19 | --dt=bf16
20 |
21 | --mb=32,64,128
22 | --batch=set_gpu
23 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/perf_ip_xe_lp:
--------------------------------------------------------------------------------
1 | --batch=perf_ip_xe
2 |
3 | --reset
4 |
5 | # Forward, int8
6 | --dir=FWD_B
7 | --dt=u8:s8:s32
8 |
9 | --mb=1,32,64,128
10 | --batch=set_gpu
11 |
12 | --mb=0
13 | --batch=shapes_1d
14 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/set_all:
--------------------------------------------------------------------------------
1 | --batch=shapes_resnet_50
2 | --batch=shapes_googlenet_v1
3 | --batch=shapes_googlenet_v3
4 | --batch=shapes_resnet_50_sparse
5 | --batch=shapes_vgg16
6 | --batch=shapes_3d
7 | --batch=shapes_wd
8 | --batch=shapes_maskrcnn
9 | --batch=shapes_rnn_t
10 | --batch=shapes_alexnet
11 | --batch=shapes_bert
12 | --batch=shapes_bert_large
13 | --batch=shapes_dlrm
14 | --batch=shapes_gnmt
15 | --batch=shapes_ncf
16 | --batch=shapes_transformer_lt
17 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/set_gpu:
--------------------------------------------------------------------------------
1 | --batch=shapes_resnet_50
2 | --batch=shapes_googlenet_v1
3 | --batch=shapes_googlenet_v3
4 | --batch=shapes_resnet_50_sparse
5 | --batch=shapes_vgg16
6 | --batch=shapes_3d
7 | --batch=shapes_wd
8 | --batch=shapes_maskrcnn
9 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/set_topologies:
--------------------------------------------------------------------------------
1 | --batch=shapes_alexnet
2 | --batch=shapes_bert
3 | --batch=shapes_bert_large
4 | --batch=shapes_dlrm
5 | --batch=shapes_gnmt
6 | --batch=shapes_googlenet_v1
7 | --batch=shapes_googlenet_v3
8 | --batch=shapes_maskrcnn
9 | --batch=shapes_ncf
10 | --batch=shapes_resnet_50
11 | --batch=shapes_resnet_50_sparse
12 | --batch=shapes_transformer_lt
13 | --batch=shapes_vgg16
14 | --batch=shapes_wd
15 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/shapes_0d_gpu:
--------------------------------------------------------------------------------
1 | mb64ic1oc2
2 | mb18ic7oc1
3 | mb16ic1oc5
4 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/shapes_1d:
--------------------------------------------------------------------------------
1 | # 1D
2 | mb128ic128iw4oc1024n"googlenet_v1_1d:ip1"
3 | mb64ic512iw7oc4096n"VGG16_1d:ip1"
4 | mb32ic64iw3oc1000n"1d:ip1"
5 | mb32ic512iw5oc1000n"1d:ip2"
6 | mb256ic128iw5oc128n"1d:ip3"
7 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/shapes_3d:
--------------------------------------------------------------------------------
1 | ic64id2ih3iw3oc1000n"3d:ip1"
2 | ic512id5ih5iw5oc1000n"3d:ip2"
3 | ic128id5ih5iw5oc128n"3d:ip3"
4 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/shapes_alexnet:
--------------------------------------------------------------------------------
1 | ic9216oc4096n"Alexnet:ip1"
2 | ic4096oc4096n"Alexnet:ip2"
3 | ic4096oc1000n"Alexnet:ip3"
4 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/shapes_bert:
--------------------------------------------------------------------------------
1 | ic768oc2304n"BERT:1"
2 | ic768oc768n"BERT:2"
3 | ic768oc3072n"BERT:3"
4 | ic3072oc768n"BERT:4"
5 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/shapes_ci:
--------------------------------------------------------------------------------
1 | --batch=shapes_dlrm
2 | --batch=shapes_ncf
3 | --batch=shapes_resnet_50
4 | --batch=shapes_rnn_t
5 | --batch=shapes_bert
6 | --batch=shapes_bert_large
7 | --batch=shapes_gnmt
8 | --batch=shapes_wd
9 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/shapes_dlrm:
--------------------------------------------------------------------------------
1 | ic13oc512n"DLRM:0"
2 | ic512oc256n"DLRM:1*2"
3 | ic256oc128n"DLRM:2"
4 | ic479oc1024n"DLRM:3"
5 | ic1024oc1024n"DLRM:4"
6 | ic1024oc512n"DLRM:5"
7 | # ic512oc256n"DLRM:6"
8 | ic256oc1n"DLRM:7"
9 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/shapes_gnmt:
--------------------------------------------------------------------------------
1 | ic512oc512n"GNMT:0"
2 | ic1024oc1024n"GNMT:1"
3 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/shapes_googlenet_v1:
--------------------------------------------------------------------------------
1 | ic128ih4iw4oc1024n"googlenet_v1:ip1"
2 | ic1024ih1iw1oc1000n"googlenet_v1:ip2"
3 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/shapes_googlenet_v3:
--------------------------------------------------------------------------------
1 | ic2048ih1oc1000n"inceptionv3:ip1"
2 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/shapes_maskrcnn:
--------------------------------------------------------------------------------
1 | mb1000ic256ih7iw7oc1024n"masknet:ip1"
2 | mb1000ic1024oc1024n"masknet:ip2"
3 | mb1000ic1024oc324n"masknet:ip3"
4 | mb1000ic1024oc81n"masknet:ip4"
5 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/shapes_ncf:
--------------------------------------------------------------------------------
1 | ic256oc256n"NCF:0"
2 | ic256oc128n"NCF:1"
3 | ic128oc64n"NCF:2"
4 | ic128oc1n"NCF:3"
5 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/shapes_regression:
--------------------------------------------------------------------------------
1 | mb1_ic16oc26_n"small_oc_block"
2 | mb128ic2200oc2200
3 | mb128ic1500oc1500
4 | mb1120ic1024oc2046
5 | mb1000ic1000oc1111
6 | mb1ic16001oc101
7 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/shapes_resnet_50:
--------------------------------------------------------------------------------
1 | ic2048oc1000n"resnet:ip1"
2 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/shapes_resnet_50_sparse:
--------------------------------------------------------------------------------
1 | ic2048oc1000n"resnet_sparse:ip1"
2 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/shapes_rnn_t:
--------------------------------------------------------------------------------
1 | # mb = 16 * num_cores for throughput inference / training
2 | # and mb = 16 for real time inference
3 | mb16ic240oc4096n"RNN-T:Encoder_cell1_Input*2"
4 | mb16ic1024oc4096n"RNN-T:Encoder_cell1_Hidden*11"
5 | mb16ic2048oc4096n"RNN-T:Encoder_cell3_Input*1"
6 | mb16ic320oc1280n"RNN-T:Prediction_Input*12"
7 | mb16ic1344oc512n"RNN-T:JointNet_Linear1*3"
8 | mb16ic512oc29n"RNN-T:JointNet_Linear2*3"
9 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/shapes_vgg16:
--------------------------------------------------------------------------------
1 | ic512ih7iw7oc4096n"VGG16:ip1"
2 | ic4096oc4096n"VGG16:ip2"
3 | ic4096oc81n"VGG16:ip3"
4 | ic4096oc324n"VGG16:ip4"
5 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/shapes_wd:
--------------------------------------------------------------------------------
1 | # Used in smoke validation, don't change the name
2 | ic845oc1024n"WnD:0"
3 | ic1024oc512n"WnD:1"
4 | ic512oc256n"WnD:2"
5 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/test_ip_all:
--------------------------------------------------------------------------------
1 | # f32
2 | --reset
3 | --dir=FWD_B,BWD_D,BWD_WB
4 | --batch=set_all --batch=shapes_0d
5 |
6 | --dir=FWD_B
7 | --attr-post-ops=sum:0.5+relu:0.5+add:f32:per_oc,prelu:per_oc,sum:2:0:s32
8 | --mb=2 --batch=set_all
9 | --mb=0 --batch=shapes_0d
10 |
11 | --batch=harness_ip_regression
12 | --batch=harness_ip_tag
13 |
14 | # int8
15 | --batch=test_ip_int8
16 |
17 | # fp8
18 | --batch=test_ip_fp8
19 |
20 | # bf16
21 | --batch=test_ip_bfloat16
22 |
23 | # bf32
24 | --batch=test_ip_bf32_bfloat16
25 |
26 | # f16
27 | --batch=test_ip_float16
28 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/ip/test_ip_bfloat16_ymm:
--------------------------------------------------------------------------------
1 | # global benchdnn knob, will not be reset again
2 | --cpu-isa-hints=prefer_ymm
3 | --batch=test_ip_bfloat16
4 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/lnorm/shapes_ci:
--------------------------------------------------------------------------------
1 | 15x3_n"lnorm_ci_0d:0" # Used in smoke validation, don't change the name
2 | 30x300_n"lnorm_ci_0d:1" # Used in smoke validation, don't change the name
3 | 256x768_n"lnorm_ci_0d:2" # Used in smoke validation, don't change the name
4 | 257x768_n"lnorm_ci_0d:3" # Used in smoke validation, don't change the name
5 | 128x1x1024_n"lnorm_ci_1d:0"
6 | 6x2x128x1024_n"lnorm_ci_2d:0"
7 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/lnorm/test_lnorm_all:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | # f32
4 | --inplace=true,false
5 | --dt=f32
6 |
7 | --dir=FWD_D,FWD_I
8 | --flags=,CH,G,GCH,M,GCHM
9 | --batch=option_set_all
10 |
11 | --dir=BWD_D
12 | --flags=,G,GM
13 | --batch=option_set_all
14 |
15 | --dir=BWD_DW
16 | --flags=CH,GCH,M,GCHM
17 | --batch=option_set_all
18 |
19 | # bf16
20 | --batch=test_lnorm_bfloat16
21 |
22 | # int8
23 | --batch=test_lnorm_int8
24 |
25 | # F16
26 | --batch=test_lnorm_float16
27 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/lnorm/test_lnorm_bfloat16:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --inplace=true,false
4 | --dt=bf16,bf16:f32,f32:bf16
5 |
6 | --dir=FWD_D,FWD_I
7 | --flags=,CH,G,GCH,M,GCHM
8 | --batch=option_set_all
9 |
10 | --dir=BWD_D
11 | --flags=,G,GM
12 | --batch=option_set_all
13 |
14 | --dir=BWD_DW
15 | --flags=CH,GCH,M,GCHM
16 | --batch=option_set_all
17 |
18 | # Different data type combinations
19 |
20 | --dt=f32:bf16,bf16
21 | --dir=FWD_D,BWD_DW
22 | --ss_dt=bf16
23 | --flags=C,H,CH
24 | --batch=option_set_all
25 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/lnorm/test_lnorm_float16:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --inplace=true,false
4 | --dt=f16,f16:f32,f32:f16
5 |
6 | --dir=FWD_D,FWD_I
7 | --flags=,CH,G,GCH,M,GCHM
8 | --batch=option_set_all
9 |
10 | --dir=BWD_D
11 | --flags=,G,GM
12 | --batch=option_set_all
13 |
14 | --dir=BWD_DW
15 | --flags=CH,GCH,M,GCHM
16 | --batch=option_set_all
17 |
18 | # Different data type combinations
19 |
20 | --dt=f32:f16,f16
21 | --dir=FWD_D,BWD_DW
22 | --ss_dt=f16
23 | --flags=C,H,CH
24 | --batch=option_set_all
25 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/lnorm/test_lnorm_int8:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --inplace=false
4 | --dt=f32:s8,f32:u8,bf16:s8,bf16:u8, \
5 | s8:f32,u8:f32,s8:bf16,u8:bf16, \
6 | s8:s8,u8:s8,s8:u8,u8:u8
7 |
8 | --dir=FWD_I
9 | --attr-scales=,src:common:128,dst:common:0.125,src:common:64+dst:common:0.5
10 | --flags=,CH,G,GCH,M,GCHM
11 | --batch=option_set_all
12 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/lnorm/test_lnorm_smoke:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --match=.*lnorm_ci_0d.* # Use 0d problems only from shapes_ci
4 | --inplace=false
5 | --tag=axb
6 | --stat_tag=any
7 |
8 | --dir=FWD_D,BWD_DW
9 | --dt=f32,bf16,f16
10 | --flags=,G,CH,M,GCHM
11 | --batch=shapes_ci
12 |
13 | --dir=BWD_D
14 | --flags=,G,GM
15 | --batch=shapes_ci
16 |
17 | # Different data type combinations
18 | --dt=f32:s8,u8:f32
19 | --dir=FWD_I
20 | --attr-scales=,src:common:64+dst:common:0.5
21 | --flags=,CH
22 | --batch=shapes_ci
23 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/lrn/set_all:
--------------------------------------------------------------------------------
1 | --batch=shapes_2d
2 | --batch=shapes_3d
3 | --batch=shapes_topologies
4 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/lrn/shapes_0d:
--------------------------------------------------------------------------------
1 | # random problems
2 | mb3ic128_n"lrn_ci_0d:0"
3 | mb10ic63_n"lrn_ci_0d:1"
4 | mb17ic15_n"lrn_ci_0d:2"
5 | ic1_n"lrn_ci_0d:3"
6 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/lrn/shapes_2d:
--------------------------------------------------------------------------------
1 | # random problems
2 | ic15_ih16_n"lrn_ci_2d:channel_tail"
3 | ic16_ih5_ls7_n"lrn_ci_2d:non_default_local_size"
4 | ic16_ih5_beta1.0_n"lrn_ci_2d:non_default_beta"
5 | ic16_ih5_k14.2_n"lrn_ci_2d:non_default_k"
6 | ic16_ih10iw7_n"lrn_ci_2d:non_square_shape"
7 | ic12_ih15iw1_n"lrn_ci_2d:unit_width"
8 | ic31_ih1iw16_n"lrn_ci_2d:unit_heigth"
9 | ic64_ih1iw1_n"lrn_ci_2d:unit_heigth_and_width"
10 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/lrn/shapes_3d:
--------------------------------------------------------------------------------
1 | ic15_id2n"channels_only"
2 | ic15_id2ih16iw16n"channel_tail"
3 | ic16_id3ih5iw5_ls7n"non_default_local_size"
4 | ic16_id2ih5iw5_beta1.0n"non_default_beta"
5 | ic16_id4ih5iw5_k14.2n"non_default_k"
6 | ic16_id2ih10iw7n"non_square_shape"
7 | ic16_id3ih10iw7n"non_square_shape"
8 | ic12_id2ih15iw1n"unit_width"
9 | ic31_id1ih1iw16_ls4n"unit_heigth"
10 | ic31_id2ih1iw16_ls4n"unit_heigth"
11 | ic31_id3ih1iw16n"unit_heigth"
12 | ic64_id1ih1iw1n"unit_heigth_and_width"
13 | ic64_id2ih1iw1n"unit_heigth_and_width"
14 | ic64_id3ih1iw1n"unit_heigth_and_width"
15 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/lrn/shapes_ci:
--------------------------------------------------------------------------------
1 | --batch=shapes_0d
2 | --batch=shapes_2d
3 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/lrn/shapes_topologies:
--------------------------------------------------------------------------------
1 | # model problems
2 | mb256ic96_ih55n"alexnet:norm1"
3 | mb256ic256_ih27n"alexnet:norm2"
4 | mb96ic64_ih57n"googlenet:pool1/norm1"
5 | mb96ic192_ih57n"googlenet:conv2/norm2"
6 | mb50ic96ih112_alpha0.0005_k2n"fastrcnn:norm1"
7 | mb50ic96ih112_ls3_alpha0.00005n"fastrcnn_zf:norm1"
8 |
9 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/lrn/test_lrn_all:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --mb=2
4 | --dt=f32
5 | --alg=ACROSS,WITHIN
6 | --dir=FWD_D,FWD_I,BWD_D
7 |
8 | --tag=abx
9 | --batch=shapes_0d
10 |
11 | --tag=abx,axb,aBx8b,aBx16b
12 | --batch=set_all
13 |
14 | # bf16
15 | --batch=test_lrn_bfloat16
16 |
17 | # f16
18 | --batch=test_lrn_float16
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/lrn/test_lrn_bfloat16:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --mb=2
4 | --dt=bf16
5 | --alg=ACROSS,WITHIN
6 | --dir=FWD_D,FWD_I,BWD_D
7 |
8 | --tag=abx
9 | --batch=shapes_0d
10 |
11 | --tag=abx,axb,aBx8b,aBx16b
12 | --batch=set_all
13 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/lrn/test_lrn_ci:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --mb=2
4 | --dt=f32,bf16,f16
5 | --alg=ACROSS,WITHIN
6 | --dir=FWD_D,FWD_I,BWD_D
7 | --tag=abx,axb
8 | --batch=shapes_ci
9 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/lrn/test_lrn_float16:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --mb=2
4 | --dt=f16
5 | --alg=ACROSS,WITHIN
6 | --dir=FWD_D,FWD_I,BWD_D
7 |
8 | --tag=abx
9 | --batch=shapes_0d
10 |
11 | --tag=abx,axb
12 | --batch=set_all
13 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/lrn/test_lrn_gpu:
--------------------------------------------------------------------------------
1 | # f32, bf16
2 | --reset
3 |
4 | --dt=f32,bf16
5 | --alg=ACROSS,WITHIN
6 | --dir=FWD_D,FWD_I,BWD_D
7 | --tag=abx
8 | --batch=set_all
9 |
10 | # f16
11 | --reset
12 |
13 | --dt=f16
14 | --alg=ACROSS,WITHIN
15 | --dir=FWD_I
16 | --tag=abx
17 | --batch=set_all
18 |
19 | # Test CI in Nightly
20 | --reset
21 | --batch=test_lrn_ci
22 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/lrn/test_lrn_smoke:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --match=.*lrn_ci_2d.* # Use 2d problems only from shapes_ci
4 | --mb=2
5 | --dt=f32,bf16,f16
6 | --alg=ACROSS,WITHIN
7 | --dir=FWD_D,FWD_I,BWD_D
8 | --tag=axb
9 | --batch=shapes_ci
10 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/harness_matmul_dropout:
--------------------------------------------------------------------------------
1 | --reset
2 | --dt=f32,bf16
3 | --attr-fpmath=,bf16
4 | --check-ref-impl=false
5 | --attr-dropout=0.5:12345678
6 |
7 | --stag=ab --dtag=ab
8 | --batch=shapes_2d
9 |
10 | --stag=abc --dtag=abc
11 | --batch=shapes_3d
12 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/harness_matmul_regression_float16:
--------------------------------------------------------------------------------
1 |
2 | # test shapes with b_buffer
3 | --reset
4 | --dt=f16:f16:f16 --stag=ab --wtag=ba --dtag=ab --bia-dt=f16 327x256:256x256
5 |
6 | # Test that cases when M == 1 are handled correctly.
7 | --reset
8 | --stag=ba,ab --wtag=ab --dtag=ab --dt=f16 1x2:2x256
9 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/harness_matmul_runtime_f32:
--------------------------------------------------------------------------------
1 | # f32
2 | --reset
3 | --skip-impl=ref
4 |
5 | --dt=f32
6 | --stag=ab,ba --wtag=ab,ba --dtag=ab
7 | --bia-dt=undef,f32 --bia_mask=2
8 |
9 | --runtime_dims_masks=0
10 | --attr-scales=src:common:0.25+wei:common:0.5+dst:common:4
11 | --attr-post-ops=,sum,relu
12 | --batch=shapes_2d
13 |
14 | --runtime_dims_masks=3:3
15 | --attr-scales=src:common:0.25+wei:common:0.5+dst:common:4
16 | --attr-post-ops=,sum+add:s8,mul:f32:per_oc,mul:f32:per_tensor
17 | --batch=shapes_2d
18 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/perf_matmul_inference_batched:
--------------------------------------------------------------------------------
1 | --batch=shapes_bert
2 | --batch=shapes_bert_large
3 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_2d:
--------------------------------------------------------------------------------
1 | # random shapes for correctness testing
2 | 1x1:1x1
3 | 10x30:30x1
4 | 1x30:30x20
5 | 10x1:1x20
6 | 1x300:300x1
7 | 1x1:1x200
8 | 100x1:1x1
9 | 10x30:30x20
10 | 2x30:30x47
11 | 10x30:30x16
12 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_2d_ci:
--------------------------------------------------------------------------------
1 | # 2d shapes for correctness testing in CI
2 | --batch=shapes_converted_ip_inf_lb_dlrm
3 | --batch=shapes_converted_ip_inf_lb_ncf
4 | --batch=shapes_converted_ip_inf_lb_rnn_t
5 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_3d:
--------------------------------------------------------------------------------
1 | # random batched shapes for correctness testing
2 | 2x10x30:2x30x1
3 | 3x30x1:3x1x20
4 | 1x30x30:1x30x20
5 | 3x10x30:3x30x16
6 |
7 | # batch broadcast shapes
8 | 7x32x16:1x16x8
9 | 1x128x8:2x8x16
10 | 2x16x73:1x73x8
11 | 1x26x17:5x17x65
12 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_4bit:
--------------------------------------------------------------------------------
1 | 24x32:32x64
2 | 25x32:32x16
3 | 96x96:96x64
4 | 14x96:96x32
5 | 1x30:30x20
6 | 10x30:30x20
7 | 2048x1024:1024x512_n"DLRM:5*1"
8 | 2048x256:256x128_n"NCF:1*1"
9 | 2048x128:128x64_n"NCF:2*1"
10 | 896x240:240x4096_n"RNN-T:Encoder_cell1_Input*2"
11 | 896x1024:1024x4096_n"RNN-T:Encoder_cell1_Hidden*11"
12 | 896x320:320x1280_n"RNN-T:Prediction_Input*12"
13 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_4d:
--------------------------------------------------------------------------------
1 | 18x16x54x64:18x16x64x54
2 | 11x16x45x45:11x16x45x64
3 | 21x16x41x41:21x16x41x64
4 | 16x16x49x49:16x16x49x64
5 | 14x16x54x54:14x16x54x64
6 | 5x16x38x38:5x16x38x64
7 | 24x16x32x32:24x16x32x64
8 | 13x16x45x64:13x16x64x45
9 | 17x16x41x64:17x16x64x41
10 | 21x16x49x64:21x16x64x49
11 |
12 | # Broadcast shapes
13 | 2x16x384x384:2x1x384x64
14 | 1x1x35x64:13x16x64x35
15 | 1x16x38x64:5x1x64x38
16 | 14x16x54x64:1x1x64x54n"B_full_bcast"
17 | 14x6x1x253:1x1x253x1n"dot_prod_w_B_full_bcast"
18 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_bert:
--------------------------------------------------------------------------------
1 | # multihead self-attention layer
2 | # mb = 1, num_heads = 12, hidden_size = 1024, t_x = t_y = 128
3 | 12x128x64:12x64x128_n"encoder:QK_matmul:12"
4 | 12x128x128:12x128x64_n"encoder:WV_matmul:12"
5 |
6 | # mb = 128, num_heads = 12, hidden_size = 768, t_x = t_y = 128,
7 | 1536x128x64:1536x64x128_n"encoder:QK_matmul:12"
8 | 1536x128x128:1536x128x64_n"encoder:WV_matmul:12"
9 |
10 | # mb = 128, num_heads = 16, hidden_size = 1024, t_x = t_y = 128,
11 | #2048x128x64:2048x64x128_n"encoder:QK_matmul:24"
12 | #2048x128x128:2048x128x64_n"encoder:WV_matmul:24"
13 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_bert_large:
--------------------------------------------------------------------------------
1 | # multihead self-attention layer
2 | # mb = 1, num_heads = 12, hidden_size = 768, t_x = t_y = 384
3 | 12x384x64:12x64x384_n"encoder:QK_matmul:12"
4 | 12x384x384:12x384x64_n"encoder:WV_matmul:12"
5 |
6 | # mb = 128, num_heads = 12, hidden_size = 768, t_x = t_y = 384
7 | 1536x384x64:1536x64x384_n"encoder:QK_matmul:12"
8 | 1536x384x384:1536x384x64_n"encoder:WV_matmul:12"
9 |
10 | # mb = 128, num_heads = 16, hidden_size = 1024, t_x = t_y = 384
11 | #2048x384x64:2048x64x384_n"encoder:QK_matmul:24"
12 | #2048x384x384:2048x384x64_n"encoder:WV_matmul:24"
13 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_inf_lb_alexnet:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product forward
2 | # ip mb=1024
3 | 1024x9216:9216x4096n"Alexnet:ip1*1"
4 | 1024x4096:4096x4096n"Alexnet:ip2*1"
5 | 1024x4096:4096x1000n"Alexnet:ip3*1"
6 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_inf_lb_dlrm:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product forward
2 | # ip mb=2048
3 | 2048x13:13x512n"DLRM:0*1"
4 | 2048x512:512x256n"DLRM:1*2"
5 | 2048x256:256x128n"DLRM:2*1"
6 | 2048x479:479x1024n"DLRM:3*1"
7 | 2048x1024:1024x1024n"DLRM:4*1"
8 | 2048x1024:1024x512n"DLRM:5*1"
9 | 2048x256:256x1n"DLRM:7*1"
10 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_inf_lb_gmnt:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product forward
2 | # ip mb=64
3 | 64x512:512x512n"GNMT:0*1"
4 | 64x1024:1024x1024n"GNMT:1*1"
5 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_inf_lb_googlenet:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product forward
2 | # ip mb=128
3 | 128x2048:2048x1024n"googlenet_v1:ip1*1"
4 | 128x1024:1024x1000n"googlenet_v1:ip2*1"
5 |
6 | # ip mb=224
7 | 224x2048:2048x1000n"inceptionv3:ip1*1"
8 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_inf_lb_maskrcnn:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product forward
2 | # ip mb=1000
3 | 1000x12544:12544x1024n"masknet:ip1*1"
4 | 1000x1024:1024x1024n"masknet:ip2*1"
5 | 1000x1024:1024x324n"masknet:ip3*1"
6 | 1000x1024:1024x81n"masknet:ip4*1"
7 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_inf_lb_ncf:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product forward
2 | # ip mb=2048
3 | # Used in smoke validation, don't change the name
4 | 2048x256:256x256n"NCF:0*1"
5 | 2048x256:256x128n"NCF:1*1"
6 | 2048x128:128x64n"NCF:2*1"
7 | 2048x128:128x1n"NCF:3*1"
8 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_inf_lb_resnet:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product forward
2 | # ip mb=112
3 | 112x2048:2048x1000n"resnet:ip1*1"
4 |
5 | # ip mb=64
6 | 64x2048:2048x1000n"resnet_sparse:ip1*1"
7 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_inf_lb_rnn_t:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product forward
2 | # ip mb=896 (16 * num_cores)
3 |
4 | 896x240:240x4096n"RNN-T:Encoder_cell1_Input*2"
5 | 896x1024:1024x4096n"RNN-T:Encoder_cell1_Hidden*11"
6 | 896x2048:2048x4096n"RNN-T:Encoder_cell3_Input*1"
7 | 896x320:320x1280n"RNN-T:Prediction_Input*12"
8 | 896x1344:1344x512n"RNN-T:JointNet_Linear1*3"
9 | 896x512:512x29n"RNN-T:JointNet_Linear2*3"
10 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_inf_lb_vgg16:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product forward
2 | # ip mb=64
3 | 64x25088:25088x4096n"VGG16:ip1*1"
4 | 64x4096:4096x4096n"VGG16:ip2*1"
5 | 64x4096:4096x81n"VGG16:ip3*1"
6 | 64x4096:4096x324n"VGG16:ip4*1"
7 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_inf_lb_wd:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product forward
2 | # ip mb=512
3 | 512x845:845x1024n"WnD-512:0*1"
4 | 512x1024:1024x512n"WnD-512:1*1"
5 | 512x512:512x256n"WnD-512:2*1"
6 |
7 | # ip mb=1024
8 | #1024x845:845x1024n"WnD-1024:0*1"
9 | #1024x1024:1024x512n"WnD-1024:1*1"
10 | #1024x512:512x256n"WnD-1024:2*1"
11 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_tr_alexnet_bwd_d:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product shapes for
2 | # training backward wrt data mb=1024
3 |
4 | 1024x4096:4096x9216n"Alexnet_train:BWD_D,ip1*1"
5 | 1024x4096:4096x4096n"Alexnet_train:BWD_D,ip2*1"
6 | 1024x1000:1000x4096n"Alexnet_train:BWD_D,ip3*1"
7 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_tr_alexnet_bwd_w:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product shapes for
2 | # training backward wrt weights mb=1024
3 |
4 | 9216x1024:1024x4096n"Alexnet_train:BWD_W,ip1*1"
5 | 4096x1024:1024x4096n"Alexnet_train:BWD_W,ip2*1"
6 | 4096x1024:1024x1000n"Alexnet_train:BWD_W,ip3*1"
7 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_tr_alexnet_fwd:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product shapes for
2 | # training forward mb=1024
3 |
4 | 1024x9216:9216x4096n"Alexnet_train:FWD,ip1*1"
5 | 1024x4096:4096x4096n"Alexnet_train:FWD,ip2*1"
6 | 1024x4096:4096x1000n"Alexnet_train:FWD,ip3*1"
7 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_tr_dlrm_bwd_d:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product shapes for
2 | # training backward wrt data mb=2048
3 |
4 | 2048x512:512x13n"DLRM_train:BWD_D,0*1"
5 | 2048x256:256x512n"DLRM_train:BWD_D,1*2"
6 | 2048x128:128x256n"DLRM_train:BWD_D,2*1"
7 | 2048x1024:1024x479n"DLRM_train:BWD_D,3*1"
8 | 2048x1024:1024x1024n"DLRM_train:BWD_D,4*1"
9 | 2048x512:512x1024n"DLRM_train:BWD_D,5*1"
10 | 2048x1:1x256n"DLRM_train:BWD_D,7*1"
11 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_tr_dlrm_bwd_w:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product shapes for
2 | # training backward wrt weights mb=2048
3 |
4 | 13x2048:2048x512n"DLRM_train:BWD_W,0*1"
5 | 512x2048:2048x256n"DLRM_train:BWD_W,1*2"
6 | 256x2048:2048x128n"DLRM_train:BWD_W,2*1"
7 | 479x2048:2048x1024n"DLRM_train:BWD_W,3*1"
8 | 1024x2048:2048x1024n"DLRM_train:BWD_W,4*1"
9 | 1024x2048:2048x512n"DLRM_train:BWD_W,5*1"
10 | 256x2048:2048x1n"DLRM_train:BWD_W,7*1"
11 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_tr_dlrm_fwd:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product shapes for
2 | # training forward mb=2048
3 |
4 | 2048x13:13x512n"DLRM_train:FWD,0*1"
5 | 2048x512:512x256n"DLRM_train:FWD,1*2"
6 | 2048x256:256x128n"DLRM_train:FWD,2*1"
7 | 2048x479:479x1024n"DLRM_train:FWD,3*1"
8 | 2048x1024:1024x1024n"DLRM_train:FWD,4*1"
9 | 2048x1024:1024x512n"DLRM_train:FWD,5*1"
10 | 2048x256:256x1n"DLRM_train:FWD,7*1"
11 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_tr_gmnt_bwd_d:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product shapes for
2 | # training backward wrt data mb=128
3 |
4 | 128x512:512x512n"GNMT_train:BWD_D,0*1"
5 | 128x1024:1024x1024n"GNMT_train:BWD_D,1*1"
6 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_tr_gmnt_bwd_w:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product shapes for
2 | # training backward wrt weights mb=128
3 |
4 | 512x128:128x512n"GNMT_train:BWD_W,0*1"
5 | 1024x128:128x1024n"GNMT_train:BWD_W,1*1"
6 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_tr_gmnt_fwd:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product shapes for
2 | # training forward mb=128
3 |
4 | 128x512:512x512n"GNMT_train:FWD,0*1"
5 | 128x1024:1024x1024n"GNMT_train:FWD,1*1"
6 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_tr_googlenet_bwd_d:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product shapes for
2 | # training backward wrt data
3 |
4 | # googlenet_v1 mb=128
5 | 128x1024:1024x2048n"googlenet_v1_train:BWD_D,ip1*1"
6 | 128x1000:1000x1024n"googlenet_v1_train:BWD_D,ip2*1"
7 |
8 | # inceptionv3 mb=224
9 | 224x1000:1000x2048n"inceptionv3_train:BWD_D,ip1*1"
10 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_tr_googlenet_bwd_w:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product shapes for
2 | # training backward wrt weights
3 |
4 | # googlenet_v1 mb=128
5 | 2048x128:128x1024n"googlenet_v1_train:BWD_W,ip1*1"
6 | 1024x128:128x1000n"googlenet_v1_train:BWD_W,ip2*1"
7 |
8 | # inceptionv3 mb=224
9 | 2048x224:224x1000n"inceptionv3_train:BWD_W,ip1*1"
10 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_tr_googlenet_fwd:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product shapes for
2 | # training forward
3 |
4 | # googlenet_v1 mb=128
5 | 128x2048:2048x1024n"googlenet_v1_train:FWD,ip1*1"
6 | 128x1024:1024x1000n"googlenet_v1_train:FWD,ip2*1"
7 |
8 | # inceptionv3 mb=224
9 | 224x2048:2048x1000n"inceptionv3_train:FWD,ip1*1"
10 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_tr_maskrcnn_bwd_d:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product shapes for
2 | # training backward wrt data mb=512
3 |
4 | 512x1024:1024x12544n"masknet_train:BWD_D,ip1*1"
5 | 512x1024:1024x1024n"masknet_train:BWD_D,ip2*1"
6 | 512x324:324x1024n"masknet_train:BWD_D,ip3*1"
7 | 512x81:81x1024n"masknet_train:BWD_D,ip4*1"
8 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_tr_maskrcnn_bwd_w:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product shapes for
2 | # training backward wrt weights mb=512
3 |
4 | 12544x512:512x1024n"masknet_train:BWD_W,ip1*1"
5 | 1024x512:512x1024n"masknet_train:BWD_W,ip2*1"
6 | 1024x512:512x324n"masknet_train:BWD_W,ip3*1"
7 | 1024x512:512x81n"masknet_train:BWD_W,ip4*1"
8 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_tr_maskrcnn_fwd:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product shapes for
2 | # training forward mb=512
3 |
4 | 512x12544:12544x1024n"masknet_train:FWD,ip1*1"
5 | 512x1024:1024x1024n"masknet_train:FWD,ip2*1"
6 | 512x1024:1024x324n"masknet_train:FWD,ip3*1"
7 | 512x1024:1024x81n"masknet_train:FWD,ip4*1"
8 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_tr_ncf_bwd_d:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product shapes for
2 | # training backward wrt data mb=2048
3 |
4 | 2048x256:256x256n"NCF_train:BWD_D,0*1"
5 | 2048x128:128x256n"NCF_train:BWD_D,1*1"
6 | 2048x64:64x128n"NCF_train:BWD_D,2*1"
7 | 2048x1:1x128n"NCF_train:BWD_D,3*1"
8 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_tr_ncf_bwd_w:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product shapes for
2 | # training backward wrt weights mb=2048
3 |
4 | 256x2048:2048x256n"NCF_train:BWD_W,0*1"
5 | 256x2048:2048x128n"NCF_train:BWD_W,1*1"
6 | 128x2048:2048x64n"NCF_train:BWD_W,2*1"
7 | 128x2048:2048x1n"NCF_train:BWD_W,3*1"
8 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_tr_ncf_fwd:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product shapes for
2 | # training forward mb=2048
3 |
4 | 2048x256:256x256n"NCF_train:FWD,0*1"
5 | 2048x256:256x128n"NCF_train:FWD,1*1"
6 | 2048x128:128x64n"NCF_train:FWD,2*1"
7 | 2048x128:128x1n"NCF_train:FWD,3*1"
8 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_tr_resnet_bwd_d:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product shapes for
2 | # training backward wrt data
3 |
4 | # resnet mb=112
5 | 112x1000:1000x2048n"resnet_train:BWD_D,ip1*1"
6 |
7 | # resnet_sparse mb=64
8 | 64x1000:1000x2048n"resnet_sparse_train:BWD_D,ip1*1"
9 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_tr_resnet_bwd_w:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product shapes for
2 | # training backward wrt weights
3 |
4 | # resnet mb=112
5 | 2048x112:112x1000n"resnet_train:BWD_W,ip1*1"
6 |
7 | # resnet_sparse mb=64
8 | 2048x64:64x1000n"resnet_sparse_train:BWD_W,ip1*1"
9 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_tr_resnet_fwd:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product shapes for
2 | # training forward
3 |
4 | # resnet mb=112
5 | 112x2048:2048x1000n"resnet_train:FWD,ip1*1"
6 |
7 | # resnet_sparse mb=64
8 | 64x2048:2048x1000n"resnet_sparse_train:FWD,ip1*1"
9 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_tr_rnn_t_bwd_d:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product shapes for
2 | # training backward wrt data mb=896 (16 * num_cores)
3 |
4 | 896x4096:4096x240n"RNN-T:BWD_D,Encoder_cell1_Input*2"
5 | 896x4096:4096x1024n"RNN-T:BWD_D,Encoder_cell1_Hidden*11"
6 | 896x4096:4096x2048n"RNN-T:BWD_D,Encoder_cell3_Input*1"
7 | 896x1280:1280x320n"RNN-T:BWD_D,Prediction_Input*12"
8 | 896x512:512x1344n"RNN-T:BWD_D,JointNet_Linear1*3"
9 | 896x29:29x512n"RNN-T:BWD_D,JointNet_Linear2*3"
10 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_tr_rnn_t_bwd_w:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product shapes for
2 | # training backward wrt weights mb=896 (16 * num_cores)
3 |
4 | 240x896:896x4096n"RNN-T:BWD_W,Encoder_cell1_Input*2"
5 | 1024x896:896x4096n"RNN-T:BWD_W,Encoder_cell1_Hidden*11"
6 | 2048x896:896x4096n"RNN-T:BWD_W,Encoder_cell3_Input*1"
7 | 320x896:896x1280n"RNN-T:BWD_W,Prediction_Input*12"
8 | 1344x896:896x512n"RNN-T:BWD_W,JointNet_Linear1*3"
9 | 512x896:896x29n"RNN-T:BWD_W,JointNet_Linear2*3"
10 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_tr_rnn_t_fwd:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product shapes for
2 | # training forward mb=896 (16 * num_cores)
3 |
4 | 896x240:240x4096n"RNN-T:FWD,Encoder_cell1_Input*2"
5 | 896x1024:1024x4096n"RNN-T:FWD,Encoder_cell1_Hidden*11"
6 | 896x2048:2048x4096n"RNN-T:FWD,Encoder_cell3_Input*1"
7 | 896x320:320x1280n"RNN-T:FWD,Prediction_Input*12"
8 | 896x1344:1344x512n"RNN-T:FWD,JointNet_Linear1*3"
9 | 896x512:512x29n"RNN-T:FWD,JointNet_Linear2*3"
10 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_tr_vgg16_bwd_d:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product shapes for
2 | # training backward wrt data mb=64
3 |
4 | 64x4096:4096x25088n"VGG16_train:BWD_D,ip1*1"
5 | 64x4096:4096x4096n"VGG16_train:BWD_D,ip2*1"
6 | 64x81:81x4096n"VGG16_train:BWD_D,ip3*1"
7 | 64x324:324x4096n"VGG16_train:BWD_D,ip4*1"
8 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_tr_vgg16_bwd_w:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product shapes for
2 | # training backward wrt weights mb=64
3 |
4 | 25088x64:64x4096n"VGG16_train:BWD_W,ip1*1"
5 | 4096x64:64x4096n"VGG16_train:BWD_W,ip2*1"
6 | 4096x64:64x81n"VGG16_train:BWD_W,ip3*1"
7 | 4096x64:64x324n"VGG16_train:BWD_W,ip4*1"
8 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_tr_vgg16_fwd:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product shapes for
2 | # training forward mb=64
3 |
4 | 64x25088:25088x4096n"VGG16_train:FWD,ip1*1"
5 | 64x4096:4096x4096n"VGG16_train:FWD,ip2*1"
6 | 64x4096:4096x81n"VGG16_train:FWD,ip3*1"
7 | 64x4096:4096x324n"VGG16_train:FWD,ip4*1"
8 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_tr_wd_bwd_d:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product shapes for
2 | # training backward wrt data
3 |
4 | # mb=40
5 | 40x1024:1024x845n"WnD-40_train:BWD_D,0*1"
6 | 40x512:512x1024n"WnD-40_train:BWD_D,1*1"
7 | 40x256:256x512n"WnD-40_train:BWD_D,2*1"
8 |
9 | # mb=256
10 | #256x1024:1024x845n"WnD-256_train:BWD_D,0*1"
11 | #256x512:512x1024n"WnD-256_train:BWD_D,1*1"
12 | #256x256:256x512n"WnD-256_train:BWD_D,2*1"
13 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_tr_wd_bwd_w:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product shapes for
2 | # training backward wrt weights
3 |
4 | # mb=40
5 | 845x40:40x1024n"WnD-40_train:BWD_W,0*1"
6 | 1024x40:40x512n"WnD-40_train:BWD_W,1*1"
7 | 512x40:40x256n"WnD-40_train:BWD_W,2*1"
8 |
9 | # mb=256
10 | #845x256:256x1024n"WnD-256_train:BWD_W,0*1"
11 | #1024x256:256x512n"WnD-256_train:BWD_W,1*1"
12 | #512x256:256x256n"WnD-256_train:BWD_W,2*1"
13 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_converted_ip_tr_wd_fwd:
--------------------------------------------------------------------------------
1 | # These problems are ported from corresponding inner product shapes for
2 | # training forward
3 |
4 | # mb=40
5 | 40x845:845x1024n"WnD-40_train:FWD,0*1"
6 | 40x1024:1024x512n"WnD-40_train:FWD,1*1"
7 | 40x512:512x256n"WnD-40_train:FWD,2*1"
8 |
9 | # mb=256
10 | #256x845:845x1024n"WnD-256_train:FWD,0*1"
11 | #256x1024:1024x512n"WnD-256_train:FWD,1*1"
12 | #256x512:512x256n"WnD-256_train:FWD,2*1"
13 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_multidim:
--------------------------------------------------------------------------------
1 | # Broadcast shapes > 4D
2 | 22x3x1x8x14:22x1x4x14x8
3 | 13x1x1x8x16:1x13x8x16x8
4 | 2x3x1x1x6x2x3:1x1x4x5x6x3x4
5 | 2x1x7x3x1x6x9:2x6x7x3x4x9x6
6 | 3x5x2x9x4x7x3:1x5x2x1x4x3x9
7 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_sparse:
--------------------------------------------------------------------------------
1 | 1x1000005:1000005x6
2 | 1x1000010:1000010x16
3 | 1x1000025:1000025x17
4 | 1x1000045:1000045x32
5 | 1x1000075:1000075x33
6 | 1x1000100:1000100x64
7 | 1x1000120:1000120x65
8 | 1x1000300:1000300x72
9 | 1x1000500:1000500x96
10 | 1x1000600:1000600x100
11 | 1x1000700:1000700x128
12 | 1x1000800:1000800x131
13 |
14 | 4x1000005:1000005x6
15 | 4x1000010:1000010x16
16 | 4x1000025:1000025x17
17 | 4x1000045:1000045x32
18 | 4x1000075:1000075x33
19 | 4x1000100:1000100x64
20 | 4x1000120:1000120x65
21 | 4x1000300:1000300x72
22 | 4x1000500:1000500x96
23 | 4x1000600:1000600x100
24 | 4x1000700:1000700x128
25 | 4x1000800:1000800x131
26 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_sparse_packed:
--------------------------------------------------------------------------------
1 | 1x64:64x128
2 | 2x64:64x128
3 | 2x128:128x128
4 | 2x300:300x128
5 | 2x300:300x129
6 | 256x256:256x256
7 | 2x1024:1024x128
8 | 2x1030:1030x128
9 | 2x1030:1030x200
10 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/shapes_transformer:
--------------------------------------------------------------------------------
1 | # multihead self-attention layer
2 | # mb = 1, num_heads = 16, hidden_size = 1024, t_x = t_y = 40
3 | 16x40x64:16x64x40
4 | 16x40x40:16x40x64
5 | # mb = 128, num_heads = 16, hidden_size = 1024, t_x = t_y = 40
6 | 2048x40x64:2048x64x40
7 | 2048x40x40:2048x40x64
8 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/test_matmul_bf32_bf16:
--------------------------------------------------------------------------------
1 | # bf32
2 | --reset
3 | --skip-impl=ref,x64:gemm
4 | --dt=f32 --attr-fpmath=bf16
5 | --stag=ab,ba --wtag=ab,ba --dtag=ab
6 |
7 | # test any + blocked
8 | --stag=any --wtag=any,BA16a64b,BA16a48b,BA16a32b,BA16a16b --dtag=any
9 | --batch=shapes_2d
10 |
11 | # 3d
12 | --reset
13 | --skip-impl=ref,x64:gemm
14 | --dt=f32 --attr-fpmath=bf16
15 | --stag=abc,acb --wtag=abc,acb --dtag=abc
16 | --batch=shapes_3d
17 | 2x20x30:2x30x4
18 | 2x20x30:1x30x4
19 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/test_matmul_bfloat16_ymm:
--------------------------------------------------------------------------------
1 | # bf16
2 |
3 | # global benchdnn knob, will not be reset again
4 | --cpu-isa-hints=prefer_ymm
5 | --batch=test_matmul_bfloat16
6 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/test_matmul_llm_gpu:
--------------------------------------------------------------------------------
1 | --reset
2 | --batch=option_set_fwks_llm_gpu
3 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/test_matmul_sparse:
--------------------------------------------------------------------------------
1 | --reset
2 | --dt=f16:f16:f16,f32:f32:f32
3 | --dtag=ab
4 | --encoding=csr+0.9::,:csr+0.9:
5 | --batch=shapes_sparse
6 |
7 | --reset
8 | --dt=f16:f16:f16,f32:f32:f32
9 | --wtag=ab,ba
10 | --dtag=ab
11 | --encoding=coo+0.9::,:coo+0.9:
12 | --batch=shapes_sparse
13 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/test_matmul_sparse_ci:
--------------------------------------------------------------------------------
1 | --reset
2 | --dt=f16:f16:f16,f32:f32:f32
3 | --dtag=ab
4 | --encoding=csr+0.99::,:csr+0.99:
5 | --batch=shapes_sparse
6 |
7 | --reset
8 | --dt=f16:f16:f16,f32:f32:f32
9 | --dtag=ab
10 | --encoding=coo+0.99::,:coo+0.99:
11 | --batch=shapes_sparse
12 |
13 | --dt=u8:s8:s32,s8:s8:s32,u8:s8:f32,s8:s8:f32
14 | --encoding=:packed+0.99:,:packed+0.5:,:packed+0.0:,:packed+1.0:
15 | --batch=shapes_sparse_packed
16 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/matmul/test_matmul_sparse_gpu:
--------------------------------------------------------------------------------
1 | --reset
2 | --batch=test_matmul_sparse
3 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/pool/harness_pool_regression:
--------------------------------------------------------------------------------
1 | # This shape dispatches to jit:ir and has
2 | # (1) no extra zero-padding, so no zero-out statement, and
3 | # (2) a number of channels that requires a 3-register (mod 4) accumulation
4 | # buffer (ic=208 has the same issue)
5 | --reset --dir=FWD_I --alg=pooling_max --dt=f16:f16 --tag=aBcd16b mb1ic80_ih160oh160kh3sh1dh0ph1_iw160ow160kw3sw1dw0pw1
6 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/pool/harness_pool_smoke_ref:
--------------------------------------------------------------------------------
1 | --reset
2 | --skip-impl=ir # Intentionally test ocl impl coverage
3 | --check-ref-impl=
4 | --match=.*pool_ci_2d.* # Use 2d problems only from shapes_basic
5 | --mb=2
6 | --tag=axb
7 | --alg=max,avg_np,avg_p
8 |
9 | # Training
10 | --dt=f32,bf16,f16
11 | --dir=FWD_D,BWD_D
12 | --batch=shapes_basic
13 |
14 | # Inference
15 | --dir=FWD_I
16 | --tag=axb
17 | --dt=f16,s8,u8
18 | --attr-post-ops=,add:f32:per_oc
19 | --batch=shapes_basic
20 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/pool/harness_pooling_different_dt:
--------------------------------------------------------------------------------
1 | --reset
2 | --mb=2
3 | --dt=s8:u8,u8:s8,u8:f32,f32:u8,s8:f32,f32:s8,u8:f16,f16:u8,s8:f16,f16:s8
4 | --dir=FWD_I
5 | --tag=axb
6 | --alg=max,avg_np,avg_p
7 | --attr-post-ops=,add:f32:per_oc
8 | --batch=shapes_1d
9 | --batch=shapes_2d_small
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/pool/perf_pool_gpu:
--------------------------------------------------------------------------------
1 | # mb=1, inference only
2 | --reset
3 | --mb=1
4 |
5 | --dt=f32,s8
6 | --dir=FWD_I
7 | --alg=max,avg_np,avg_p
8 | --tag=axb,aBx16b
9 |
10 | --batch=set_topologies_gpu
11 |
12 | # mb>1, inference only
13 | --reset
14 | --mb=16,32,64
15 |
16 | --dt=f32,s8
17 | --dir=FWD_I
18 | --alg=max,avg_np,avg_p
19 | --tag=axb,ABx16a16b
20 |
21 | --batch=set_topologies_gpu
22 |
23 | # mb>1, training only
24 | --reset
25 | --mb=16,32,64
26 |
27 | --dt=f32
28 | --dir=FWD_D,BWD_D
29 | --alg=max,avg_np,avg_p
30 | --tag=axb,ABx16a16b
31 |
32 | --batch=set_topologies_gpu
33 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/pool/set_all:
--------------------------------------------------------------------------------
1 | --batch=shapes_1d
2 | --batch=shapes_2d
3 | --batch=shapes_3d
4 |
5 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/pool/set_all_small:
--------------------------------------------------------------------------------
1 | --batch=shapes_1d
2 | --batch=shapes_2d_small
3 | --batch=shapes_3d_small
4 |
5 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/pool/set_topologies:
--------------------------------------------------------------------------------
1 | --batch=shapes_3d_unet
2 | --batch=shapes_alexnet
3 | --batch=shapes_googlenet_v1
4 | --batch=shapes_googlenet_v3
5 | --batch=shapes_i3d_resnet50_v1
6 | --batch=shapes_resnet_50
7 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/pool/set_topologies_gpu:
--------------------------------------------------------------------------------
1 | --batch=shapes_alexnet
2 | --batch=shapes_googlenet_v1
3 | --batch=shapes_googlenet_v3
4 | --batch=shapes_resnet_50
5 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/pool/shapes_3d_unet:
--------------------------------------------------------------------------------
1 | # 3d unet
2 |
3 | mb1ic64_id64od32_kd2sd2n"3d_unet:max_pool1"
4 | mb1ic128_id28od14_kd2sd2n"3d_unet:max_pool2"
5 | mb1ic256_id12od6_kd2sd2n"3d_unet:max_pool3"
6 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/pool/shapes_alexnet:
--------------------------------------------------------------------------------
1 | # alexnet
2 |
3 | mb256ic96_ih55oh27_kh3sh2n"alexnet:max_pool1"
4 | mb256ic256_ih27oh13_kh3sh2n"alexnet:max_pool2"
5 | mb256ic256_ih13oh6_kh3sh2n"alexnet:max_pool5"
6 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/pool/shapes_global_pooling:
--------------------------------------------------------------------------------
1 | mb2ic5_iw11ow1_kw11sw11pw0_ih7oh1_kh7sh7ph0
2 | mb2ic5_iw11ow1_kw11sw11pw0_ih1oh1_kh1sh1ph0
3 | mb2ic5_iw11ow1_kw11sw11pw0_ih1oh1_kh1sh1ph0_iw15ow1_kw15sw15pw0
4 | mb4ic32_iw32ow1_kw32sw32pw0_ih32oh1_kh32sh32ph0
5 | mb4ic16_iw8ow1_kw8sw8pw0_ih16oh1_kh16sh16ph0
6 | mb16ic16_iw8ow1_kw8sw8pw0_ih4oh1_kh4sh4ph0
7 | mb32ic16_iw8ow1_kw8sw8pw0_ih4oh1_kh4sh4ph0
8 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/pool/shapes_i3d_resnet50_v1:
--------------------------------------------------------------------------------
1 | # i3d resnet50 v1
2 |
3 | mb1ic64_id16od8kd1sd2pd0_ih112oh56kh3sh2ph1_iw112ow56kw3sw2pw1_n"i3d_resnet50_v1:max_pool1"
4 | mb1ic256_id8od4kd2sd2pd0_ih56oh56kh1sh1ph0_iw56ow56kw1sw1pw0_n"i3d_resnet50_v1:max_pool2"
5 | mb1ic2048_id4od1kd4sd1pd0_ih7oh1kh7sh1ph0_iw7ow1kw7sw1pw0_n"i3d_resnet50_v1:max_pool3"
6 |
7 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/pool/shapes_large_pool:
--------------------------------------------------------------------------------
1 | # Large iw
2 | mb1ic32iw134217732kw7sw5
3 | mb1ic1iw4294967311kw7sw5
4 |
5 | # Large mb
6 | mb4294967311ic1iw1pw1kw3
7 |
8 | # Large ic
9 | mb1ic4294967311iw1pw1kw3
10 | mb1ic4294967311iw1kw1
11 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/pool/shapes_resnet_50:
--------------------------------------------------------------------------------
1 | # resnet_50
2 |
3 | mb50ic64_ih112oh56_kh3sh2n"resnet_50:max_pool1"
4 | # mb50ic256_ih56oh28_khXshXn"resnet_50:res2c_max_pool"
5 | # mb50ic512_ih28oh14_khXshXn"resnet_50:res3d_max_pool"
6 | # mb50ic1024_ih14oh7_khXshXn"resnet_50:res4f_max_pool"
7 | mb50ic2048_ih7oh1_kh7sh1n"resnet_50:ave_pool5"
8 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/pool/test_pool_ci:
--------------------------------------------------------------------------------
1 | --reset
2 | --mb=2
3 |
4 | --alg=max,avg_np,avg_p
5 | # Training
6 | --dt=f32,bf16,f16
7 | --dir=FWD_D,BWD_D
8 | --tag=abx,axb
9 | --batch=shapes_basic
10 |
11 | # Inference
12 | --dir=FWD_I
13 | --tag=axb
14 |
15 | ## All inference configs
16 | --dt=f32,bf16,f16,s32,s8,u8, \
17 | s8:u8,u8:s8,s8:f32,f32:s8,u8:f32,f32:u8,s8:f16,f16:s8,u8:f16,f16:u8
18 | --batch=shapes_basic
19 |
20 | ## Attributes
21 | --dt=f32,bf16,f16,s32,s8,u8
22 | --attr-post-ops=add:f32:per_oc,linear:0.5:-1
23 | --batch=shapes_basic
24 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/pool/test_pool_float16:
--------------------------------------------------------------------------------
1 | # f16
2 | --reset
3 | --mb=2
4 |
5 | --dt=f16
6 | --alg=max,avg_np,avg_p
7 | # Training
8 | --tag=abx,axb
9 |
10 | --dir=FWD_D,BWD_D
11 | --batch=set_all
12 | --batch=set_topologies
13 |
14 | --dir=FWD_D
15 | --attr-post-ops=add:f32:per_oc,linear:0.5:-1
16 | --batch=set_all_small
17 |
18 | # Inference
19 | --dir=FWD_I
20 | --tag=axb
21 | --batch=set_all
22 |
23 | --attr-post-ops=add:f16,linear:0.5:-1
24 | --batch=set_all_small
25 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/pool/test_pool_fp8:
--------------------------------------------------------------------------------
1 | # fp8
2 | --reset
3 | --mb=2
4 |
5 | --dt=f8_e5m2,f8_e4m3
6 | --alg=max,avg_np,avg_p
7 | # Training
8 | --tag=abx,axb,aBx16b
9 |
10 | --dir=FWD_D
11 | --batch=set_all
12 | --batch=set_topologies
13 |
14 | --dir=FWD_D
15 | --attr-post-ops=add:f32:per_oc,linear:0.5:-1
16 | --batch=set_all_small
17 |
18 | # Inference
19 | --dir=FWD_I
20 | --tag=axb
21 | --batch=set_all
22 |
23 | --attr-post-ops=add:f16,linear:0.5:-1
24 | --batch=set_all_small
25 |
26 | --attr-post-ops=add:f8_e5m2,linear:0.5:-1
27 | --batch=set_all_small
28 |
29 | --attr-post-ops=add:f8_e4m3,linear:0.5:-1
30 | --batch=set_all_small
31 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/pool/test_pool_large_gpu:
--------------------------------------------------------------------------------
1 | # Implicitly test FWD_D via BWD_D
2 | --dir=FWD_I,BWD_D
3 | --dt=bf16:bf16
4 | --alg=max,avg_p
5 | --tag=axb
6 |
7 | --impl=jit
8 | --batch=shapes_large_pool
9 |
10 | # Test both gen_pooling and xe_global pooling in the same pass
11 | --impl=xe
12 | --batch=shapes_large_pool
13 |
14 | --impl=ref
15 | --batch=shapes_large_pool
16 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/pool/test_pool_smoke:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --match=.*pool_ci_2d.* # Use 2d problems only from shapes_basic
4 | --mb=2
5 | --tag=axb
6 | --alg=max,avg_np,avg_p
7 |
8 | # Training
9 | --dt=f32,bf16,f16
10 | --dir=FWD_D,BWD_D
11 | --batch=shapes_basic
12 |
13 | # Inference
14 | --dir=FWD_I
15 | --tag=axb
16 | --dt=f16,s8,u8
17 | --attr-post-ops=,add:f32:per_oc
18 | --batch=shapes_basic
19 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/prelu/option_set_all:
--------------------------------------------------------------------------------
1 | # tag:any
2 | --stag=abx:any, \
3 | axb:any, \
4 | aBx8b:any, \
5 | aBx16b:any
6 | --batch=shapes_all
7 |
8 | # tag:tag
9 | --stag=abx:abx, \
10 | axb:axb, \
11 | aBx8b:aBx8b, \
12 | aBx16b:aBx16b
13 | --batch=shapes_all
14 |
15 | # tag1:tag2
16 | --stag=abx:axb, \
17 | axb:aBx8b, \
18 | aBx16b:abx
19 | --batch=shapes_all
20 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/prelu/test_prelu_all:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --dir=FWD_D,BWD_DW
4 | --sdt=f32,s32,s8,u8
5 | --batch=option_set_all
6 |
7 | --batch=test_prelu_bfloat16
8 |
9 | --batch=test_prelu_float16
10 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/prelu/test_prelu_bfloat16:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --dir=FWD_D,BWD_DW
4 | --sdt=bf16:f32,bf16
5 | --batch=option_set_all
6 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/prelu/test_prelu_ci:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --stag=abx:any,axb:any,abx:abx,axb:axb
4 |
5 | --dir=FWD_D,BWD_DW
6 | --sdt=f32,bf16:f32,bf16,f16,f16:f32
7 | --batch=shapes_ci
8 |
9 | --dir=FWD_I
10 | --sdt=s8:s8,u8:u8,s8:bf16,u8:bf16,s8:f32,u8:f32
11 | --batch=shapes_ci
12 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/prelu/test_prelu_float16:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --dir=FWD_D,BWD_DW
4 | --sdt=f16:f32,f16
5 | --batch=option_set_all
6 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/prelu/test_prelu_smoke:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --match=.*prelu_ci_2d.* # Use 2d problems only from shapes_ci
4 | --stag=axb:any
5 |
6 | --dir=FWD_D,BWD_DW
7 | --sdt=f32,bf16,f16
8 | --batch=shapes_ci
9 |
10 | --dir=FWD_I
11 | --sdt=s8,u8
12 | --batch=shapes_ci
13 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/reduction/harness_reduction_bf16:
--------------------------------------------------------------------------------
1 | # bf16
2 | --reset
3 |
4 | --sdt=bf16 --ddt=bf16,f32
5 | --attr-post-ops=,sum+linear:2:1+add:f32,add:f32:per_oc
6 | --batch=option_set_all_algs
7 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/reduction/harness_reduction_f16:
--------------------------------------------------------------------------------
1 | # f16
2 | --reset
3 |
4 | --sdt=f16 --ddt=f16,f32
5 | --attr-post-ops=,sum+linear:2:1+add:f32,add:f32:per_oc
6 | --batch=option_set_all_algs
7 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/reduction/harness_reduction_f32:
--------------------------------------------------------------------------------
1 | # f32
2 | --reset
3 |
4 | --sdt=f32 --ddt=f32
5 | --attr-post-ops=,sum+linear:2:1+add:f32,add:f32:per_oc
6 | --batch=option_set_all_algs
7 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/reduction/harness_reduction_i8:
--------------------------------------------------------------------------------
1 | # i8
2 | --reset
3 |
4 | --attr-post-ops=,sum+linear:2:1+add:f32,add:f32:per_oc
5 | --sdt=u8 --ddt=u8,s32,f32
6 | --batch=option_set_all_algs_int8
7 |
8 | --sdt=s8 --ddt=s8,s32,f32
9 | --batch=option_set_all_algs_int8
10 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/reduction/option_set_all_algs:
--------------------------------------------------------------------------------
1 | # Algorithm coverage based on p and eps validity
2 | --p=1,2 --eps=0.5
3 | --alg=norm_lp_max,norm_lp_sum,norm_lp_power_p_max,norm_lp_power_p_sum
4 | --batch=option_set_all
5 |
6 | --p= --eps=
7 | --alg=sum,mul,max,min,mean
8 | --batch=option_set_all
9 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/reduction/option_set_all_algs_ci:
--------------------------------------------------------------------------------
1 | # Algorithm coverage based on p and eps validity
2 | --p=1,2 --eps=0.5
3 | --alg=norm_lp_max,norm_lp_sum,norm_lp_power_p_max,norm_lp_power_p_sum
4 | --batch=shapes_ci
5 |
6 | --p= --eps=
7 | --alg=sum,mul,max,min,mean
8 | --batch=shapes_ci
9 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/reduction/option_set_all_algs_int8:
--------------------------------------------------------------------------------
1 | # i8
2 | --alg=sum,mul,max,min,mean
3 | --batch=option_set_all
4 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/reduction/option_set_all_algs_int8_ci:
--------------------------------------------------------------------------------
1 | # i8
2 | --alg=sum,mul,max,min,mean
3 | --batch=shapes_ci
4 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/reduction/perf_reduction_gpu:
--------------------------------------------------------------------------------
1 | # Test layers of some key & extened GPU DL Frameworks
2 | --reset
3 | --batch=option_set_fwks_key_gpu
4 |
5 | --reset
6 | --batch=option_set_fwks_ext_gpu
7 |
8 | # Test nested cases
9 | --reset
10 | --batch=shapes_nested_gpu
11 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/reduction/shapes_ci:
--------------------------------------------------------------------------------
1 | 5x3x2x9:1x3x2x9
2 | 32x17x2x3:32x17x1x1
3 | 32x17x2x3:1x17x1x1
4 | 15x12x3x5:15x1x1x1
5 | 15x12x3x5:1x1x1x1
6 | 12x12:1x12
7 | 10x16x32:10x1x32
8 | 1x17x64:1x1x64
9 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/reduction/shapes_gpu_all:
--------------------------------------------------------------------------------
1 | # ND single-dim reduction
2 | 10000:1
3 | 8000x32:1x32
4 | 32x4096x16:32x1x16
5 | 2x52x128x36:2x1x128x36
6 | 17x17x15x15x35:1x17x15x15x35
7 | 11x23x16x16x7x2:11x1x16x16x7x2
8 |
9 | # 2-dim reduction
10 | 16x3x3x30:16x1x1x30
11 | 16x128x4x64:16x128x1x1
12 | 128x53x17:128x1x1
13 |
14 | # 3-dim reduction
15 | 4x16x3x4:1x1x1x4
16 | 4x16x3x4:1x1x3x1
17 | 4x16x3x4:1x16x1x1
18 | 4x16x3x4:4x1x1x1
19 |
20 | # Full reduction
21 | 4192x17:1x1
22 | 36x640x4:1x1x1
23 | 16x16x32x16:1x1x1x1
24 |
25 | # Split reduction
26 | 16x16x32x32:1x16x1x32
27 | 16x16x32x32:16x1x32x1
28 | 5x15x7x17x9:1x15x1x17x1
29 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/reduction/test_reduction_all:
--------------------------------------------------------------------------------
1 | # all
2 | --reset
3 |
4 | --batch=harness_reduction_f32
5 | --batch=harness_reduction_i8
6 | --batch=test_reduction_bfloat16
7 | --batch=test_reduction_float16
8 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/reduction/test_reduction_bfloat16:
--------------------------------------------------------------------------------
1 | # bf16
2 | --reset
3 |
4 | --batch=harness_reduction_bf16
5 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/reduction/test_reduction_ci:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --stag=abx,axb --dtag=abx,axb,any
4 | --attr-post-ops=,sum+linear:2:1+add:f32
5 |
6 | --sdt=f32 --ddt=f32
7 | --batch=option_set_all_algs_ci
8 |
9 | --sdt=bf16 --ddt=bf16,f32
10 | --batch=option_set_all_algs_ci
11 |
12 | --sdt=f16 --ddt=f16,f32
13 | --batch=option_set_all_algs_ci
14 |
15 | --sdt=s8 --ddt=s8,s32,f32
16 | --batch=option_set_all_algs_int8_ci
17 |
18 | --sdt=u8 --ddt=u8,s32,f32
19 | --batch=option_set_all_algs_int8_ci
20 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/reduction/test_reduction_float16:
--------------------------------------------------------------------------------
1 | # f16
2 | --reset
3 |
4 | --batch=harness_reduction_f16
5 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/reduction/test_reduction_smoke:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --stag=axb --dtag=any
4 | --attr-post-ops=,sum+linear:2:1+add:f32
5 |
6 | --sdt=f32 --ddt=f32
7 | --batch=option_set_all_algs_ci
8 |
9 | --sdt=bf16 --ddt=bf16
10 | --batch=option_set_all_algs_ci
11 |
12 | --sdt=f16 --ddt=f16
13 | --batch=option_set_all_algs_ci
14 |
15 | --sdt=s8 --ddt=s8
16 | --batch=option_set_all_algs_int8_ci
17 |
18 | --sdt=u8 --ddt=u8
19 | --batch=option_set_all_algs_int8_ci
20 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/reorder/harness_reorder_large:
--------------------------------------------------------------------------------
1 | # test if jit kernels properly handle corner cases:
2 | # * large stride problems
3 | # * huge dimensions (UINT_MAX + 1)
4 | --reset
5 | --skip-impl=ref,simple # run only jit impl, won't iterate
6 | --sdt=f32
7 | --ddt=f32
8 | --stag=abx
9 | --dtag=aBx8b
10 | 2x16x19200x19200
11 | 1x4294967296x1
12 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/reorder/harness_reorder_regression:
--------------------------------------------------------------------------------
1 | # Blocked format with tail processing in compensation
2 | --reset
3 | --sdt=s8 --ddt=s8 --oflag=zp_comp:3
4 | --stag=aBxC16b4c --dtag=xcab
5 | 2x2x32x1x3 1x15x32x1 2x17x32x5
6 |
7 | # test if jit kernel applies zero-point
8 | --reset
9 | --skip-impl=ref,simple # ! test jit version only
10 | --sdt=u8 --ddt=f32
11 | --stag=abdc --dtag=abcd
12 | --attr-zero-points=src0:common:1
13 | 1x32x128x33
14 |
15 | # Test bf16 with aBcde4b format
16 | --reset
17 | --skip-impl=simple #skip non-jit version
18 | --sdt=bf16 --ddt=bf16
19 | --stag=aBcde4b --dtag=aBcde4b
20 | 2x24x19x19x19
21 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/reorder/harness_reorder_saturation:
--------------------------------------------------------------------------------
1 | # basic checks for saturation
2 | --reset
3 |
4 | --sdt=f32,s32,s8,u8
5 | --ddt=f32,s32,s8,u8
6 |
7 | --attr-scales=src:common:4294967295
8 | --stag=abx,axb,aBx4b,aBx8b,aBx16b
9 | --dtag=abx,axb,aBx4b,aBx8b,aBx16b
10 | 1x17x9x5 2x64x3x3
11 |
12 | # checks for int overflow
13 | --reset
14 |
15 | --sdt=s32
16 | --ddt=f32,s8
17 |
18 | --attr-scales=src:common:4294967295
19 | --attr-zero-points=src:common:1
20 | --stag=abx,axb,aBx4b,aBx8b,aBx16b
21 | --dtag=abx,axb,aBx4b,aBx8b,aBx16b
22 | 1x17x9x5
23 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/reorder/test_reorder_smoke:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --sdt=f32,bf16,f16,s32,s8,u8
4 | --ddt=f32,bf16,f16,s32,s8,u8
5 | --attr-scales=,src:per_dim_1+dst:per_dim_1
6 | --attr-zero-points=,src:common:-1+dst:common:2
7 | --attr-post-ops=,sum:0.5
8 | --runtime-dim-mask=0,63
9 | --stag=abx
10 | --dtag=axb
11 | --oflag=
12 | 2x16x3x4 1x17x5x3
13 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/resampling/set_all:
--------------------------------------------------------------------------------
1 | --batch=shapes_1d
2 | --batch=shapes_2d
3 | --batch=shapes_3d
4 | --batch=shapes_maskrcnn
5 |
6 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/resampling/shapes_1d:
--------------------------------------------------------------------------------
1 | # random problems
2 |
3 | # upsampling
4 | mb16ic2_iw2_ow6
5 | mb16ic64_iw32_ow64
6 | mb1ic32_iw151_ow300
7 | mb4ic17_iw17_ow20
8 |
9 | # downsampling
10 | mb1ic8_iw14_ow7
11 | mb1ic3_iw20_ow17
12 | mb1ic1_iw21_ow13
13 | mb1ic32_iw32_ow6
14 | mb2ic5_iw42_ow14
15 | mb1ic23_iw525_ow5
16 |
17 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/resampling/shapes_2d:
--------------------------------------------------------------------------------
1 | # random problems
2 |
3 | # upsampling
4 | mb1ic8_ih6oh6_iw24ow24
5 | mb1ic8_ih3oh6_iw7ow21
6 | mb2ic16_ih5oh3_iw10ow11
7 | mb12ic8_ih5oh3_iw7ow11
8 |
9 | # downsampling
10 | mb1ic8_ih14oh14_iw7ow7
11 | mb1ic8_ih14oh6_iw7ow3
12 | mb2ic16_ih15oh73_iw10ow11
13 | mb12ic8_ih15oh3_iw5ow11
14 |
15 | # mixed
16 | mb16ic18_ih14oh6_iw7ow12
17 | mb16ic8_ih4oh63_iw9ow13
18 |
19 | # tails for blocked format
20 | mb4ic17_ih6oh7_iw12ow14
21 | mb4ic23_ih60oh60_iw30ow75
22 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/resampling/shapes_3d:
--------------------------------------------------------------------------------
1 | # random problems
2 |
3 | # upsampling
4 | ic8_id2ih4iw6_od4oh8ow12
5 | ic9_id3ih6iw4_od4oh8ow9
6 |
7 | # downsampling
8 | ic16_id6ih6iw6_od3oh3ow3
9 | ic19_id6ih6iw6_od4oh4ow4
10 |
11 | # mixed
12 | mb4ic16_id6ih6iw6_od12oh6ow3
13 | mb32ic32_id31ih50iw31_od16oh77ow16
14 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/resampling/shapes_maskrcnn:
--------------------------------------------------------------------------------
1 | mb1ic256ih34iw25oh68ow50n"maskrcnn1"
2 | mb1ic256ih68iw50oh136ow100n"maskrcnn2"
3 | mb1ic256ih136iw100oh272ow200n"maskrcnn3"
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/resampling/test_resampling_bfloat16:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | # bf16
4 | --mb=2
5 | --sdt=bf16
6 | --ddt=bf16
7 | --dir=FWD_D,BWD_D
8 | --alg=nearest,linear
9 | --tag=abx,axb,aBx8b,aBx16b
10 | --batch=set_all
11 |
12 | # post-ops
13 | --dir=FWD_D
14 | --sdt=bf16
15 | --ddt=bf16
16 | --attr-post-ops=add:bf16
17 | --batch=shapes_ci
18 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/resampling/test_resampling_ci:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --mb=2
4 | --tag=abx,axb
5 | --alg=nearest,linear
6 |
7 | --dir=FWD_D
8 | --attr-post-ops=,sum+add:f32
9 |
10 | --sdt=f32 --ddt=f32,s8
11 | --batch=shapes_ci
12 |
13 | --sdt=bf16 --ddt=bf16
14 | --batch=shapes_ci
15 |
16 | --sdt=f16 --ddt=f16
17 | --batch=shapes_ci
18 |
19 | # backward
20 | --dir=BWD_D
21 | --attr-post-ops=
22 |
23 | --sdt=f32 --ddt=f32
24 | --batch=shapes_ci
25 |
26 | --sdt=bf16 --ddt=bf16
27 | --batch=shapes_ci
28 |
29 | --sdt=f16 --ddt=f16
30 | --batch=shapes_ci
31 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/resampling/test_resampling_float16:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | # f16
4 | --mb=2
5 | --sdt=f16
6 | --ddt=f16
7 | --dir=FWD_D,BWD_D
8 | --alg=nearest,linear
9 | --tag=abx,axb
10 | --batch=set_all
11 |
12 | # post ops
13 | --dir=FWD_D
14 | --sdt=f16
15 | --ddt=f16
16 | --attr-post-ops=add:f16
17 | --batch=shapes_ci
18 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/resampling/test_resampling_smoke:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --match=.*resampling_ci_2d.* # Use 2d problems only from shapes_ci
4 | --mb=2
5 | --tag=axb
6 | --alg=nearest,linear
7 |
8 | --dir=FWD_D
9 | --attr-post-ops=,sum+add:f32
10 |
11 | --sdt=f32 --ddt=f32
12 | --batch=shapes_ci
13 |
14 | --sdt=bf16 --ddt=bf16
15 | --batch=shapes_ci
16 |
17 | --sdt=f16 --ddt=f16
18 | --batch=shapes_ci
19 |
20 | # backward
21 | --dir=BWD_D
22 | --attr-post-ops=
23 |
24 | --sdt=f32 --ddt=f32
25 | --batch=shapes_ci
26 |
27 | --sdt=bf16 --ddt=bf16
28 | --batch=shapes_ci
29 |
30 | --sdt=f16 --ddt=f16
31 | --batch=shapes_ci
32 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/harness_augru_bf32:
--------------------------------------------------------------------------------
1 | # bf32
2 | --reset
3 |
4 | --trivial-strides=true,false
5 | --alg=VANILLA_AUGRU,LBR_AUGRU
6 | --activation=UNDEF
7 | --attr-fpmath=bf16
8 | --prop=FWD_I
9 |
10 | # small problems
11 | --direction=left2right
12 | --batch=option_set_small
13 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/harness_augru_bfloat16:
--------------------------------------------------------------------------------
1 | # bf16
2 | --reset
3 |
4 | --trivial-strides=true,false
5 | --alg=VANILLA_AUGRU,LBR_AUGRU
6 | --activation=UNDEF
7 | --cfg=bf16f32,bf16
8 | --prop=FWD_I,BWD_DW
9 |
10 | # small problems
11 | --direction=left2right
12 | --batch=option_set_small
13 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/harness_augru_float16:
--------------------------------------------------------------------------------
1 | # f16
2 | --reset
3 |
4 | --trivial-strides=true,false
5 | --alg=VANILLA_AUGRU,LBR_AUGRU
6 | --activation=UNDEF
7 | --cfg=f16f32,f16
8 | --prop=FWD_I,BWD_DW
9 |
10 | # small problems
11 | --direction=left2right
12 | --batch=option_set_small
13 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/harness_gru_bf32:
--------------------------------------------------------------------------------
1 | # bf32
2 | --reset
3 |
4 | --trivial-strides=true,false
5 | --alg=VANILLA_GRU,LBR_GRU
6 | --activation=UNDEF
7 | --attr-fpmath=bf16
8 | --prop=FWD_I
9 |
10 | # small problems
11 | --direction=left2right,right2left,concat,sum
12 | --batch=option_set_small
13 |
14 | # large problems
15 | --direction=left2right
16 | --batch=option_set_large
17 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/harness_gru_bfloat16:
--------------------------------------------------------------------------------
1 | # bf16
2 | --reset
3 |
4 | --trivial-strides=true,false
5 | --alg=VANILLA_GRU,LBR_GRU
6 | --activation=UNDEF
7 | --cfg=bf16f32,bf16
8 | --prop=FWD_I,BWD_DW
9 |
10 | # small problems
11 | --direction=left2right,right2left,concat,sum
12 | --batch=option_set_small
13 |
14 | # large problems
15 | --direction=left2right
16 | --batch=option_set_large
17 |
18 | --prop=BWD_DW
19 | --flags=O
20 |
21 | # small problems
22 | --direction=left2right,right2left,concat,sum
23 | --batch=option_set_small
24 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/harness_gru_f32:
--------------------------------------------------------------------------------
1 | # f32
2 | --reset
3 |
4 | --trivial-strides=true,false
5 | --alg=VANILLA_GRU,LBR_GRU
6 | --activation=UNDEF
7 | --cfg=f32
8 | --prop=FWD_I,BWD_DW
9 |
10 | # small problems
11 | --direction=left2right,right2left,concat,sum
12 | --batch=option_set_small
13 |
14 | # large problems
15 | --direction=left2right
16 | --batch=option_set_large
17 |
18 | --prop=BWD_DW
19 | --flags=O
20 |
21 | # small problems
22 | --direction=left2right,right2left,concat,sum
23 | --batch=option_set_small
24 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/harness_gru_float16:
--------------------------------------------------------------------------------
1 | # f16
2 | --reset
3 |
4 | --trivial-strides=true,false
5 | --alg=VANILLA_GRU,LBR_GRU
6 | --activation=UNDEF
7 | --cfg=f16f32,f16
8 | --prop=FWD_I,BWD_DW
9 |
10 | # small problems
11 | --direction=left2right,right2left,concat,sum
12 | --batch=option_set_small
13 |
14 | # large problems
15 | --direction=left2right
16 | --batch=option_set_large
17 |
18 | --prop=BWD_DW
19 | --flags=O
20 |
21 | # small problems
22 | --direction=left2right,right2left,concat,sum
23 | --batch=option_set_small
24 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/harness_gru_int8:
--------------------------------------------------------------------------------
1 | # int8
2 | --reset
3 |
4 | --trivial-strides=true
5 | --prop=FWD_I
6 | --alg=VANILLA_GRU
7 | --activation=UNDEF
8 |
9 | # small problems
10 | --cfg=u8u8u8u8,u8u8u8f32,f32u8f32u8,f32u8f32f32
11 | --direction=left2right,right2left,concat,sum
12 | --scaling=common,per_oc
13 | --batch=option_set_small
14 |
15 | # large problems
16 | --cfg=u8u8u8u8
17 | --direction=left2right
18 | --scaling=per_oc
19 | --batch=option_set_large
20 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/harness_gru_regression:
--------------------------------------------------------------------------------
1 | # int8 SIC != SLC
2 | --reset
3 | --trivial-strides=true --prop=FWD_I --alg=VANILLA_GRU --activation=UNDEF
4 | --direction=left2right --cfg=u8u8u8f32 l1t47mb100sic128slc256dhc128dic128
5 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/harness_lstm_bf32:
--------------------------------------------------------------------------------
1 | # bf32
2 | --reset
3 |
4 | --trivial-strides=true,false
5 | --prop=FWD_I
6 | --alg=VANILLA_LSTM
7 | --activation=UNDEF
8 | --attr-fpmath=bf16
9 | --with-peephole=false,true
10 | --with-projection=false,true
11 |
12 | # small problems
13 | --direction=left2right,right2left,concat,sum
14 | --batch=option_set_small
15 | --batch=option_set_lstmp_small
16 |
17 | # large problems
18 | --direction=left2right
19 | --batch=option_set_large
20 | --batch=option_set_lstmp_large
21 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/harness_rnn_bf32:
--------------------------------------------------------------------------------
1 | # bf32
2 | --reset
3 |
4 | --trivial-strides=true,false
5 | --alg=VANILLA_RNN
6 | --attr-fpmath=bf16
7 | --prop=FWD_I
8 |
9 | # small test case to check GEMM and non-GEMM ops accuracy
10 | --direction=left2right,right2left,concat,sum
11 | --activation=RELU,TANH,LOGISTIC
12 | --batch=option_set_small
13 |
14 | # large cases - test linear activations
15 | --direction=left2right
16 | --activation=RELU
17 | --batch=option_set_large
18 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/option_set_gnmt_decoder:
--------------------------------------------------------------------------------
1 | --direction=left2right
2 | l1t1sic512slc768dhc512n"GNMT:decoder_0"
3 | l1t1sic512slc1024dhc512n"GNMT:decoder_1"
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/option_set_large:
--------------------------------------------------------------------------------
1 | # option set to run all reasonable large shapes
2 |
3 | --skip-nonlinear=true
4 | --tag=tnc:any:tnc,ntc:any:ntc
5 | --batch=shapes_large
6 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/option_set_lstmp_large:
--------------------------------------------------------------------------------
1 | # option set to run all reasonable large lstmp shapes
2 |
3 | --skip-nonlinear=true
4 | --batch=shapes_lstmp_large
5 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/option_set_lstmp_small:
--------------------------------------------------------------------------------
1 | # option set to run all reasonable small lstmp shapes
2 |
3 | --skip-nonlinear=false
4 | --batch=shapes_lstmp_small
5 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/option_set_perf_inference_lb:
--------------------------------------------------------------------------------
1 | # inference_lb -- inference with large batch size
2 |
3 | --alg=VANILLA_LSTM
4 | --activation=UNDEF
5 |
6 | --mb=640
7 | --batch=option_set_gnmt_decoder
8 |
9 | --mb=64
10 | --batch=option_set_gnmt_encoder
11 |
12 | --mb=64
13 | --alg=VANILLA_RNN
14 | --direction=left2right
15 | --activation=TANH
16 | --batch=shapes_deepspeech_2
17 |
18 | --mb=64
19 | --alg=LBR_GRU
20 | --direction=left2right
21 | --activation=UNDEF
22 | --batch=shapes_deepspeech_2
23 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/option_set_perf_inference_sb:
--------------------------------------------------------------------------------
1 | # inference_sb -- inference with small batch size
2 |
3 | --alg=VANILLA_LSTM
4 | --activation=UNDEF
5 |
6 | --mb=10
7 | --batch=option_set_gnmt_decoder
8 |
9 | --mb=1
10 | --batch=option_set_gnmt_encoder
11 |
12 | --mb=1
13 | --alg=VANILLA_RNN
14 | --direction=left2right
15 | --activation=TANH
16 | --batch=shapes_deepspeech_2
17 |
18 | --mb=1
19 | --alg=LBR_GRU
20 | --direction=left2right
21 | --activation=UNDEF
22 | --batch=shapes_deepspeech_2
23 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/option_set_perf_training:
--------------------------------------------------------------------------------
1 | --mb=128
2 |
3 | --alg=VANILLA_LSTM
4 | --activation=UNDEF
5 | --batch=option_set_gnmt_decoder
6 | --batch=option_set_gnmt_encoder
7 |
8 | --alg=VANILLA_RNN
9 | --direction=left2right
10 | --activation=TANH
11 | --batch=shapes_deepspeech_2
12 |
13 | --alg=LBR_GRU
14 | --direction=left2right
15 | --activation=UNDEF
16 | --batch=shapes_deepspeech_2
17 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/option_set_rnnt:
--------------------------------------------------------------------------------
1 | # RNN-T LSTM shapes
2 |
3 | --alg=VANILLA_LSTM
4 | --activation=UNDEF
5 | --direction=left2right
6 |
7 | --batch=shapes_rnn_t
8 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/option_set_small:
--------------------------------------------------------------------------------
1 | # option set to run all reasonable small shapes
2 |
3 | --skip-nonlinear=false
4 | --tag=tnc:any:tnc,ntc:any:ntc
5 | --batch=shapes_small
6 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/perf_rnn_cpu:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | # Forward
4 |
5 | --prop=FWD_I
6 |
7 | --cfg=f32
8 | --batch=option_set_perf_inference_lb
9 | --batch=option_set_perf_inference_sb
10 | --batch=option_set_perf_training
11 |
12 | --cfg=u8u8u8u8
13 | --trivial-strides=true
14 | --scaling=per_oc
15 | --batch=option_set_perf_inference_lb
16 | --batch=option_set_perf_inference_sb
17 | --batch=option_set_perf_training
18 |
19 | # Backward
20 |
21 | --prop=BWD_DW
22 | --cfg=f32
23 | --trivial-strides=
24 | --scaling=
25 |
26 | --batch=option_set_perf_inference_lb
27 | --batch=option_set_perf_inference_sb
28 | --batch=option_set_perf_training
29 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/perf_rnn_inference_lb:
--------------------------------------------------------------------------------
1 | # inference_lb -- inference with large batch size
2 | --reset
3 |
4 | --mb=640
5 | --alg=VANILLA_LSTM
6 | --activation=TANH
7 | --batch=option_set_gnmt_decoder
8 |
9 | --mb=64
10 | --alg=VANILLA_LSTM
11 | --activation=TANH
12 | --batch=option_set_gnmt_encoder
13 |
14 | --mb=64
15 | --alg=VANILLA_RNN,LBR_GRU
16 | --direction=left2right
17 | --batch=shapes_deepspeech_2
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/perf_rnn_inference_sb:
--------------------------------------------------------------------------------
1 | # inference_sb -- inference with small batch size
2 | --reset
3 |
4 | --mb=10
5 | --alg=VANILLA_LSTM
6 | --activation=TANH
7 | --batch=option_set_gnmt_decoder
8 |
9 | --mb=1
10 | --alg=VANILLA_LSTM
11 | --activation=TANH
12 | --batch=option_set_gnmt_encoder
13 |
14 | --mb=1
15 | --alg=VANILLA_RNN,LBR_GRU
16 | --direction=left2right
17 | --batch=shapes_deepspeech_2
18 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/perf_rnn_knx:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --prop=FWD_I,BWD_DW
4 |
5 | --cfg=f32
6 | --batch=option_set_perf_inference_lb
7 | --batch=option_set_perf_inference_sb
8 | --batch=option_set_perf_training
9 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/perf_rnn_training:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --mb=128
4 | --alg=VANILLA_LSTM
5 | --activation=TANH
6 | --batch=option_set_gnmt_decoder
7 | --batch=option_set_gnmt_encoder
8 |
9 | --alg=VANILLA_RNN,LBR_GRU
10 | --direction=left2right
11 | --batch=shapes_deepspeech_2
12 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/perf_rnn_xe:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | # Forward
4 |
5 | --prop=FWD_I
6 | --cfg=f16,f32
7 |
8 | --batch=option_set_perf_inference_lb
9 | --batch=option_set_perf_inference_sb
10 | --batch=option_set_perf_training
11 |
12 | # Backward
13 |
14 | --prop=BWD_DW
15 | --cfg=f32
16 |
17 | --batch=option_set_perf_inference_lb
18 | --batch=option_set_perf_inference_sb
19 | --batch=option_set_perf_training
20 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/perf_rnn_xe_hp:
--------------------------------------------------------------------------------
1 | --batch=perf_rnn_xe_lp
2 |
3 | --reset
4 |
5 | # Forward, bf16
6 |
7 | --prop=FWD_I
8 | --cfg=bf16f32
9 |
10 | --batch=option_set_perf_inference_lb
11 | --batch=option_set_perf_inference_sb
12 | --batch=option_set_perf_training
13 |
14 | # Backward
15 |
16 | --prop=BWD_DW
17 | --cfg=bf16f32
18 |
19 | --batch=option_set_perf_inference_lb
20 | --batch=option_set_perf_inference_sb
21 | --batch=option_set_perf_training
22 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/perf_rnn_xe_lp:
--------------------------------------------------------------------------------
1 | --batch=perf_rnn_xe
2 |
3 | --reset
4 |
5 | # Forward, int8
6 | --prop=FWD_I
7 | --cfg=u8u8u8u8
8 | --scaling=per_oc
9 | --trivial-strides=true
10 |
11 | --batch=option_set_perf_inference_lb
12 | --batch=option_set_perf_inference_sb
13 | --batch=option_set_perf_training
14 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/shapes_deepspeech_2:
--------------------------------------------------------------------------------
1 | l1t50sic1760n"deepspeech2:0"
2 | l1t100sic1760n"deepspeech2:1"
3 | l1t200sic1760n"deepspeech2:2"
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/shapes_inference:
--------------------------------------------------------------------------------
1 | l1t30mb1sic512n"GNMT_enc-inference"
2 | l7t30mb1sic1024n"GNMT_enc-inference"
3 | l8t1mb1sic2048slc1024dhc1024n"GNMT_dec-inference"
4 | l1t1mb1sic2048slc1024dhc1024n"GNMT_dec-inference"
5 | l1t1mb640sic2048slc1024dhc1024n"GNMT_dec-inference"
6 | l1t50mb1sic1760n"deepspeech2-inference"
7 | l1t100mb1sic1760n"deepspeech2-inference"
8 | l1t200mb1sic1760n"deepspeech2-inference"
9 | l1t50mb1sic500n"pytorch_testcase-inference"
10 | l1t629mb1sic128n"paddlepaddle_testcase-inference"
11 | l1t10mb1sic128slc512dhc128n"exp-0"
12 | l10t1mb1sic512slc128dhc128n"exp-1"
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/shapes_large:
--------------------------------------------------------------------------------
1 | # large shapes
2 |
3 | l1t1mb63_sic64_n"uniform"
4 | l1t1mb34_sic65_n"uniform:tail"
5 | l1t1mb19_sic64_slc128_n"non-uniform:slc_neq_sic"
6 | l1t1mb12_sic65_dhc130_n"non-uniform:slc_neq_dhc_tail"
7 | l1t1mb6_sic64_slc128_dhc256_n"non-uniform:slc_neq_sic_neq_dhc"
8 | l1t1mb4_sic65_slc130_dhc260_n"non-uniform:slc_neq_sic_neq_dhc_tail"
9 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/shapes_large_gru:
--------------------------------------------------------------------------------
1 | l1t1mb65_sic64_n"uniform"
2 | l1t1mb17_sic128_n"uniform"
3 | l1t1mb100_sic65_n"uniform:tail"
4 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/shapes_lstmp_large:
--------------------------------------------------------------------------------
1 | # large shapes for lstm w/ projection when dhc != dic
2 |
3 | l1t1mb31_sic64_dic128_n"non-uniform:dhc_neq_dic"
4 | l1t1mb32_sic65_dic130_n"non-uniform:dhc_neq_dic_tail"
5 | l1t1mb12_sic64_slc128_dic128_n"non-uniform:slc_neq_sic_and_dhc_neq_dic"
6 | l1t1mb10_sic65_dhc130_dic260_n"non-uniform:slc_neq_dhc_neq_dic_tail"
7 | l1t1mb3_sic64_slc128_dhc256_dic320_n"non-uniform:slc_neq_sic_neq_dhc_neq_dic"
8 | l1t1mb4_sic65_slc130_dhc260_dic325_n"non-uniform:slc_neq_sic_neq_dhc_neq_dic_tail"
9 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/shapes_lstmp_small:
--------------------------------------------------------------------------------
1 | # small shapes for lstm w/ projection when dhc != dic
2 |
3 | l1t1mb9_sic16_dic32_n"non-uniform:dhc_neq_dic"
4 | l1t1mb7_sic17_dic34_n"non-uniform:dhc_neq_dic_tail"
5 | l1t1mb3_sic16_slc32_dic32_n"non-uniform:slc_neq_sic_and_dhc_neq_dic"
6 | l1t1mb4_sic17_dhc34_dic68_n"non-uniform:slc_neq_dhc_neq_dic_tail"
7 | l1t1mb2_sic16_slc32_dhc64_dic80_n"non-uniform:slc_neq_sic_neq_dhc_neq_dic"
8 | l1t1mb3_sic17_slc34_dhc68_dic85_n"non-uniform:slc_neq_sic_neq_dhc_neq_dic_tail"
9 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/shapes_rnn_t:
--------------------------------------------------------------------------------
1 | # RNN-T shapes with fixed timestamp = 256
2 | l1t256mb896sic1024slc240dhc1024dic1024n"RNN-T:Encoder_1"
3 | l1t256mb896sic1024slc1024dhc1024dic1024n"RNN-T:Encoder_2"
4 | l1t128mb896sic1024slc2048dhc1024dic1024n"RNN-T:Encoder_3"
5 | l1t128mb896sic1024slc1024dhc1024dic1024n"RNN-T:Encoder_4*2"
6 | l1t1mb896sic320slc320dhc320dic320n"RNN-T:Prediction*768"
7 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/shapes_small:
--------------------------------------------------------------------------------
1 | # small shapes
2 |
3 | l8t3mb12_sic16_n"uniform"
4 | l4t3mb20_sic36_n"uniform:unroll_tail"
5 | l1t2mb6_sic16_slc32_n"non-uniform:slc_neq_sic"
6 | l1t1mb7_sic17_dhc34_n"non-uniform:slc_neq_dhc_tail"
7 | l1t1mb3_sic16_slc32_dhc64_n"non-uniform:slc_neq_sic_neq_dhc"
8 | l1t1mb4_sic17_slc34_dhc68_n"non-uniform:slc_neq_sic_neq_dhc_tail"
9 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/shapes_small_gru:
--------------------------------------------------------------------------------
1 | l14t10mb12_sic16_n"uniform"
2 | l10t14mb10_sic17_n"uniform:tail"
3 | l5t7mb7_sic32_n"uniform:unroll"
4 | l1t6mb8_sic36_n"uniform:unroll_tail"
5 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/shapes_training:
--------------------------------------------------------------------------------
1 | l1t1mb128sic512n"GNMT_enc-training"
2 | l2t2mb128sic1024n"GNMT_enc-training"
3 | l8t1mb128sic2048slc1024dhc1024n"GNMT_dec-training"
4 | l1t1mb128sic2048slc1024dhc1024n"GNMT_dec-training"
5 | l1t50mb32sic1760n"deepspeech2-training"
6 | l1t100mb32sic1760n"deepspeech2-training"
7 | l1t200mb32sic1760n"deepspeech2-training"
8 | l1t50mb64sic500n"pytorch_testcase-training"
9 | l1t629mb128sic128n"paddlepaddle_testcase-training"
10 | l1t952mb128sic128n"paddlepaddle_testcase-training"
11 | l1t10mb32sic128slc512dhc128n"exp-0"
12 | l10t1mb32sic512slc128dhc128n"exp-1"
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/test_augru_all:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --batch=test_augru_ci
4 |
5 | --batch=test_augru_bfloat16
6 |
7 | --batch=test_augru_bf32_bfloat16
8 |
9 | --batch=test_augru_float16
10 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/test_augru_bf32_bfloat16:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --batch=harness_augru_bf32
4 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/test_augru_bfloat16:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --batch=harness_augru_bfloat16
4 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/test_augru_ci:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --alg=LBR_AUGRU,VANILLA_AUGRU
4 | --activation=UNDEF
5 | --direction=left2right
6 | --skip-nonlinear=false
7 |
8 | --trivial-strides=true,false
9 | --prop=FWD_I,BWD_DW
10 | --cfg=f32,bf16f32,bf16,f16
11 | --batch=shapes_small
12 |
13 | --trivial-strides=true,false
14 | --prop=FWD_I
15 | --cfg=f32
16 | --attr-fpmath=bf16
17 | --batch=shapes_small
18 |
19 |
20 | # flags
21 | --trivial-strides=true,false
22 | --prop=BWD_DW
23 | --cfg=f32
24 | --flags=O
25 | --batch=shapes_small
26 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/test_augru_float16:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --batch=harness_augru_float16
4 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/test_gru_all:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --batch=harness_gru_f32
4 |
5 | --batch=test_gru_int8
6 |
7 | --batch=test_gru_bfloat16
8 |
9 | --batch=test_gru_bf32_bfloat16
10 |
11 | --batch=test_gru_float16
12 |
13 | --batch=harness_gru_regression
14 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/test_gru_bf32_bfloat16:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --batch=harness_gru_bf32
4 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/test_gru_bfloat16:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --batch=harness_gru_bfloat16
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/test_gru_float16:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --batch=harness_gru_float16
4 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/test_gru_int8:
--------------------------------------------------------------------------------
1 | # int8
2 | --reset
3 |
4 | --batch=harness_gru_int8
5 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/test_lstm_all:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --batch=test_lstm_f32
4 |
5 | --batch=test_lstm_int8
6 |
7 | --batch=test_lstm_bfloat16
8 |
9 | --batch=test_lstm_bf32_bfloat16
10 |
11 | --batch=test_lstm_float16
12 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/test_lstm_bf32_bfloat16:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --batch=harness_lstm_bf32
4 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/test_lstm_bfloat16:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --batch=harness_lstm_bfloat16
4 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/test_lstm_bfloat16_ymm:
--------------------------------------------------------------------------------
1 | # global benchdnn knob, will not be reset again
2 | --cpu-isa-hints=prefer_ymm
3 |
4 | --reset
5 | --batch=test_lstm_bfloat16
6 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/test_lstm_f32:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --batch=harness_lstm_f32
4 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/test_lstm_float16:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --batch=harness_lstm_float16
4 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/test_lstm_int8:
--------------------------------------------------------------------------------
1 | # int8
2 | --reset
3 |
4 | --batch=harness_lstm_int8
5 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/test_rnn_all:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --batch=harness_rnn_f32
4 |
5 | --batch=test_rnn_bfloat16
6 |
7 | --batch=test_rnn_bf32_bfloat16
8 |
9 | --batch=test_rnn_float16
10 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/test_rnn_bf32_bfloat16:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --batch=harness_rnn_bf32
4 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/test_rnn_bfloat16:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --batch=harness_rnn_bfloat16
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/rnn/test_rnn_float16:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --batch=harness_rnn_float16
4 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/self/test_self_ci:
--------------------------------------------------------------------------------
1 | # Use empty input file to align with other drivers
2 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/self/test_self_f32:
--------------------------------------------------------------------------------
1 | # Use empty input file to align with other drivers
2 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/self/test_self_smoke:
--------------------------------------------------------------------------------
1 | # Use empty input file to align with other drivers
2 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/shuffle/option_set_all:
--------------------------------------------------------------------------------
1 | --group=2,4
2 |
3 | --axis=1 1x12x56x56 1x24x56x56 1x36x56x56 1x68x56x56
4 | --axis=1,2 1x68x56x56 1x136x56x56 1x272x56x56
5 | --axis=1,3,4 1x272x2x56x56
6 |
7 | --group=3
8 |
9 | --axis=1 1x3x224x224 1x24x56x56 1x48x28x28 1x96x14x14 1x192x7x7
10 | --axis=1,2 1x36x225x225 1x72x57x57 1x144x27x27
11 | --axis=1,3,4 1x282x2x57x57
12 |
13 | --group=8
14 |
15 | --axis=1,2 1x24x56x56 1x32x56x56 1x72x56x56
16 | --axis=1,3,4 1x272x2x56x56
17 |
18 | --group=16
19 |
20 | --axis=1,2 1x16x64x64 1x48x64x64 1x128x64x64
21 | --axis=1,3,4 1x272x2x64x64
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/shuffle/option_set_min:
--------------------------------------------------------------------------------
1 | --axis=1,2
2 |
3 | --group=2,4 1x12x56x56 1x24x56x56 1x272x56x56
4 |
5 | --group=3 1x36x225x225 1x72x57x57 1x144x27x27
6 |
7 | --group=8 1x24x56x56 1x32x56x56 1x72x56x56
8 |
9 | --group=16 1x16x64x64 1x48x64x64 1x128x64x64
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/shuffle/perf_shuffle_cpu:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --dir=FWD_D
4 | --dt=f32,bf16
5 | --tag=abx,axb,aBx4b,aBx8b,aBx16b
6 | --batch=option_set_perf
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/shuffle/test_shuffle_all:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --dir=FWD_D
4 | --dt=f32,s32,s8,u8
5 | --tag=abx,axb,aBx4b,aBx8b,aBx16b
6 | --batch=option_set_all
7 |
8 | --dir=BWD_D
9 | --dt=f32
10 | --batch=option_set_min
11 |
12 | # bf16
13 | --batch=test_shuffle_bfloat16
14 |
15 | # f16
16 | --batch=test_shuffle_float16
17 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/shuffle/test_shuffle_bfloat16:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --dt=bf16
4 |
5 | --dir=FWD_D
6 | --tag=abx,axb,aBx16b
7 | --batch=option_set_all
8 |
9 | --dir=BWD_D
10 | --batch=option_set_min
11 |
12 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/shuffle/test_shuffle_ci:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --group=4
4 |
5 | --dir=FWD_D,BWD_D
6 | --dt=f32,bf16,f16,s32,s8,u8
7 | --tag=abx,axb
8 | --axis=1,2
9 | 2x12x32x17 3x16x36x9
10 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/shuffle/test_shuffle_float16:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --dt=f16
4 |
5 | --dir=FWD_D
6 | --tag=abx,axb
7 | --batch=option_set_all
8 |
9 | --dir=BWD_D
10 | --batch=option_set_min
11 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/shuffle/test_shuffle_gpu:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --group=4
4 | --dir=FWD_D,BWD_D
5 |
6 | --dt=f32,f16,bf16,s8,u8
7 |
8 | --tag=abx,aBx16b
9 | --axis=1,2 1x68x56x56 1x272x56x56
10 | --axis=1,3,4 1x272x2x56x56
11 |
12 | --tag=ABx16a16b
13 | --axis=1,2 32x64x56x56
14 | --axis=1,3,4 32x64x2x56x56
15 |
16 | # blocked with tail
17 | --dt=f32
18 | --tag=aBx16b --axis=1 1x12x56x56 1x36x56x56
19 |
20 | # double block
21 | --allow-enum-tags-only=0
22 | --tag=BA8b4a2b
23 | --group=6
24 | --axis=0,1
25 | 48x48
26 |
27 | # Test CI in Nightly
28 | --reset
29 | --batch=test_shuffle_ci
30 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/shuffle/test_shuffle_smoke:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --dir=FWD_D,BWD_D
4 | --dt=f32,bf16,f16,s8,u8
5 | --tag=axb
6 | --group=4
7 | --axis=1
8 | 2x12x32x17 3x16x36x9
9 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/softmax/harness_softmax_regression:
--------------------------------------------------------------------------------
1 | # "vectorized" and "small" reusable kernels (stride 1, aligned) with scale
2 | --reset
3 | --dir=FWD_I
4 | --sdt=u8
5 | --ddt=s8
6 | --attr-scales=dst:common:0.125,src:common:2
7 | --axis=3 1x4x192x256 1x4x16x32
8 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/softmax/set_0d:
--------------------------------------------------------------------------------
1 | # 2d dataset
2 |
3 | --batch=shapes_0d
4 | --batch=shapes_nlp
5 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/softmax/shapes_0d:
--------------------------------------------------------------------------------
1 | # 2d dataset
2 |
3 | 96x1000
4 | 256x10
5 | 32x100
6 | 2x113
7 | 128x365
8 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/softmax/shapes_2d:
--------------------------------------------------------------------------------
1 | # 4d dataset
2 |
3 | # spatial is 1
4 | 96x1000x1x1
5 | 256x10x1x1
6 |
7 | # regular 4d
8 | 2x19x128x256
9 | 2x16x128x128
10 | 1x8x1024x16
11 | 1x2x64x64
12 | 448x16x28x28
13 | 64x1011x1x1
14 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/softmax/shapes_3d:
--------------------------------------------------------------------------------
1 | # 5d dataset
2 |
3 | # spatial is 1
4 | 96x1024x1x1x1
5 | 256x10x1x1x1
6 |
7 | # regular 5d
8 | 3x17x9x37x19
9 | 2x16x128x2x4
10 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/softmax/shapes_ci:
--------------------------------------------------------------------------------
1 | # basic shapes
2 |
3 | 16x16_n"softmax_ci_0d:0"
4 | 255x10_n"softmax_ci_0d:1"
5 | 2x19x17x13_n"softmax_ci_2d:0" # Used in smoke validation, don't change the name
6 | 1x16x2x12_n"softmax_ci_2d:1" # Used in smoke validation, don't change the name
7 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/softmax/shapes_large:
--------------------------------------------------------------------------------
1 | 8192x64
2 | 16384x64
3 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/softmax/shapes_large_axis:
--------------------------------------------------------------------------------
1 | 8x3x8320
2 | 4x3x5600
3 | 1x1x4097
4 | 2x3x9999
5 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/softmax/shapes_nlp:
--------------------------------------------------------------------------------
1 | # nlp
2 |
3 | 240x20
4 | 2560x20
5 | 2688x21
6 | 3712x29
7 | 4480x35
8 | 4736x37
9 | 4864x38
10 | 4992x39
11 | 5120x40
12 | 5248x41
13 | 5376x42
14 | 5504x43
15 | 5632x44
16 | 6144x48
17 | 6272x49
18 | 6400x50
19 | 6784x53
20 | 7296x57
21 | 7424x58
22 | 7808x61
23 | 8192x64
24 | 8448x66
25 | 8576x67
26 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/softmax/test_softmax_acl:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | # only FWD_I is supported with ACL
4 | --dir=FWD_I
5 |
6 | # do not run ref
7 | --skip-impl=ref
8 |
9 | --alg=SOFTMAX,LOGSOFTMAX
10 |
11 | --sdt=f32
12 | --ddt=f32
13 | --dtag=any
14 |
15 | --axis=0,1
16 | --batch=shapes_ci
17 | --batch=shapes_nlp
18 |
19 | --stag=abx
20 | --axis=0,1
21 | --batch=set_0d
22 | --axis=1,3
23 | --batch=shapes_2d
24 | --axis=3,4
25 | --batch=shapes_3d
26 |
27 | --stag=axb
28 | --axis=0,1
29 | --batch=shapes_2d
30 | --batch=shapes_3d
31 |
32 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/softmax/test_softmax_smoke:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --match=.*softmax_ci_2d.* # Use 2d problems only from shapes_ci
4 | --inplace=false
5 | --stag=axb
6 | --dtag=any
7 | --alg=SOFTMAX,LOGSOFTMAX
8 | --axis=1
9 |
10 | --dir=FWD_D,BWD_D
11 | --sdt=f32,bf16,f16
12 | --ddt=f32,bf16,f16
13 | --batch=shapes_ci
14 |
15 | --dir=FWD_I
16 | --sdt=s8,u8
17 | --ddt=s8,u8
18 | --attr-scales=,src:common:64+dst:common:0.5
19 | --batch=shapes_ci
20 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/sum/option_set_fwks_ext_gpu:
--------------------------------------------------------------------------------
1 | --reset --allow-enum-tags-only=0 --sdt=f32:f32 --ddt=f32 --stag=abcd:abcd --dtag=abcd 1113x8x30x64_n"1bca9bd0e78574fe0ab41bbc68b795f9"
2 | --reset --allow-enum-tags-only=0 --sdt=f32:f32 --ddt=f32 --stag=abcd:abcd --dtag=abcd 64x8x50x64_n"8def238b6b453fb393856464ba8a0d4d"
3 | --reset --allow-enum-tags-only=0 --sdt=f32:f32:f32 --ddt=f32 --stag=abc:abc:abc --dtag=abc 64x50x512_n"7d5660f01c9be41fba9d6135abf33d2d"
4 | --reset --allow-enum-tags-only=0 --sdt=f32:f32:f32:f32 --ddt=f32 --stag=ab:ab:ab:ab --dtag=ab 64x512_n"0bab0aacc5f8028faa61019716a8ef50"
5 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/sum/test_sum_all:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --inplace=true,false
4 | --scales=0.25,1,4
5 | --ddt=f32,s32,s8,u8
6 | --sdt=f32:f32,f32:s32,f32:s8,f32:u8,s32:s8,s32:u8,s8:u8
7 | --dtag=undef,abx,axb,aBx8b,aBx16b
8 | --stag=abx:abx 3x3x16x4
9 | --stag=axb:axb 4x4x2x16 5x5x1x15
10 | --stag=aBx8b:aBx8b 2x8x3x10 1x9x4x7
11 | --stag=aBx16b:aBx16b 1x16x5x11 2x15x6x3
12 |
13 | --ddt=f32,s32
14 | --sdt=f32:s32:s8
15 | --stag=aBx8b:abx:axb,axb:axb:axb
16 | --scales=1.25:3:0.5 16x2x6x4x3
17 |
18 | # bf16
19 | --batch=test_sum_bfloat16
20 |
21 | # f16
22 | --batch=test_sum_float16
23 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/sum/test_sum_bfloat16:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --scales=0.25,1,4
4 | --ddt=f32,bf16
5 | --sdt=bf16:bf16
6 | --dtag=undef,abx,axb,aBx16b
7 | --stag=abx:abx 3x3x16x4
8 | --stag=axb:axb 4x4x2x16 5x5x1x15
9 | --stag=aBx16b:aBx16b 1x16x5x11 2x15x6x3
10 |
11 | --sdt=bf16:bf16:bf16
12 | --stag=aBx16b:abx:axb,axb:axb:axb
13 | --scales=2:0.25:0.5 16x2x6x4x3
14 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/sum/test_sum_ci:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --inplace=true,false
4 | --ddt=f32,bf16,f16,s32,s8,u8
5 | --sdt=f32:f32,f32:bf16,f32:s8,bf16:bf16,f16:f16,s32:s8,u8:u8
6 | --dtag=undef,any,abx,axb
7 | --stag=abx:abx,axb:axb
8 | --scales=0.25:2
9 | 3x17x5x7 4x16x8x10
10 |
11 | --ddt=f32,s8
12 | --sdt=f32:u8:s8
13 | --stag=abx:abx:abx,axb:axb:axb
14 | --scales=0.25:2:0.5
15 | 2x17x5x7x3 4x16x8x10x2
16 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/sum/test_sum_float16:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --scales=0.25,1,4
4 | --ddt=f32,f16
5 | --sdt=f16:f16
6 | --dtag=undef,abx,axb
7 | --stag=abx:abx 3x3x16x4
8 | --stag=axb:axb 4x4x2x16 5x5x1x15
9 |
10 | --sdt=f16:f16:f16
11 | --stag=axb:axb:axb
12 | --scales=2:0.25:0.5 16x2x6x4x3
13 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/sum/test_sum_smoke:
--------------------------------------------------------------------------------
1 | --reset
2 |
3 | --inplace=false
4 | --ddt=f32,bf16,f16,s8,u8
5 | --sdt=f32:f32,bf16:bf16,f16:f16,s8:s8,u8:u8
6 | --dtag=undef
7 | --stag=axb
8 | --scales=0.25
9 | 3x17x5x7 4x16x8x10
10 |
11 | --ddt=f32
12 | --sdt=f32:f32:f32
13 | --stag=axb:axb:axb
14 | 2x17x5x7x3 4x16x8x10x2
15 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/zeropad/option_set_fwks_ext_gpu:
--------------------------------------------------------------------------------
1 | #
2 | --reset --allow-enum-tags-only=0 --dt=f16 --tag=aBcd16b 1x4x1080x1920n"text-image-super-resolution-0001_onnx.inf.fp16.ov.b1*1"
3 | #
4 | --reset --allow-enum-tags-only=0 --dt=f16 --tag=ABcd32a16b 32x4x1080x1920n"text-image-super-resolution-0001_onnx.inf.fp16.ov.b32*1"
5 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/zeropad/option_set_fwks_key_gpu:
--------------------------------------------------------------------------------
1 | #
2 | --reset --allow-enum-tags-only=0 --dt=f32 --tag=ABcde32a16b 32x6x13x14x9n"3dgan.tr.fp32.tf.mb256*8"
3 | --reset --allow-enum-tags-only=0 --dt=f32 --tag=ABcde32a16b 256x8x25x25x25n"3dgan.tr.fp32.tf.mb256*8"
4 | --reset --allow-enum-tags-only=0 --dt=f32 --tag=ABcde16a16b 256x8x25x25x25n"3dgan.tr.fp32.tf.mb256*8"
5 | --reset --allow-enum-tags-only=0 --dt=f32 --tag=ABcde16a16b 256x8x23x23x23n"3dgan.tr.fp32.tf.mb256*8"
6 | --reset --allow-enum-tags-only=0 --dt=f32 --tag=ABcde32a16b 256x6x13x14x9n"3dgan.tr.fp32.tf.mb256*4"
7 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/zeropad/set_dim1_block_3d:
--------------------------------------------------------------------------------
1 | --tag=Abc16a,Abc4a
2 | --batch=shapes_dim1_block_3d
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/zeropad/set_dim1dim2_block_2d:
--------------------------------------------------------------------------------
1 | --tag=AB48a16b,AB48a32b,BA4b8a8b4a
2 |
3 | --batch=shapes_dim1dim2_block_2d
4 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/zeropad/set_dim1dim2_block_3d:
--------------------------------------------------------------------------------
1 | # The follwing tags are omitted as they should be tested in the 4d dim2 dim3
2 | # blocked cases
3 | # --tag=ABc16a16b,ABc4a4b,ABc16b16a,ABc4b16a4b,ABc2b8a4b,
4 | # ABc16b16a4b,ABc16b16a2b,ABc4b4a,ABc8a16b2a,ABc8a8b,
5 | # ABc8a4b,ABc8b16a2b,ABc8b8a
6 |
7 | --tag=ABc32a32b,ABc16b32a,ABc16b64a,ABc4b32a4b,ABc4b64a4b,ABc8b32a2b,ABc8b64a2b,ABc4a8b8a4b
8 | --batch=shapes_dim1dim2_block_3d
9 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/zeropad/set_dim2_block_3d:
--------------------------------------------------------------------------------
1 | --tag=aBc16b,aBc32b,aBc4b,aBc8b
2 | --batch=shapes_dim2_block_3d
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/zeropad/set_dim2dim3_block_4d:
--------------------------------------------------------------------------------
1 | --tag=aBCd16b16c,aBCd16c16b,aBCd2c4b2c,aBCd4b8c2b,aBCd4c16b4c,aBCd2c8b4c,aBCd16c16b4c,aBCd16c16b2c,aBCd4c4b,aBCd4b4c,aBCd4c8b2c,aBCd8b16c2b,aBCd8b8c,aBCd8b4c,aBCd8c16b2c,aBCd8c8b,aBCd2b4c2b
2 | --batch=shapes_dim2dim3_block_4d
3 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/zeropad/shapes_dim1_block_3d:
--------------------------------------------------------------------------------
1 | 1x1x1
2 | 1x1x33
3 | 1x33x1
4 | 1x33x33
5 | 3x1x1
6 | 3x1x33
7 | 3x33x1
8 | 3x33x33
9 | 7x1x1
10 | 7x1x33
11 | 7x33x1
12 | 7x33x33
13 | 9x1x1
14 | 9x1x33
15 | 9x33x1
16 | 9x33x33
17 | 19x1x1
18 | 19x1x33
19 | 19x33x1
20 | 19x33x33
21 | 43x1x1
22 | 43x1x33
23 | 43x33x1
24 | 43x33x33
25 | 71x1x1
26 | 71x1x33
27 | 71x33x1
28 | 71x33x33
29 | 128x1x1
30 | 128x1x33
31 | 128x33x1
32 | 128x33x33
33 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/zeropad/shapes_dim2_block_3d:
--------------------------------------------------------------------------------
1 | 1x1x1
2 | 1x1x33
3 | 1x3x1
4 | 1x3x33
5 | 1x7x1
6 | 1x7x33
7 | 1x9x1
8 | 1x9x33
9 | 1x19x1
10 | 1x19x33
11 | 1x43x1
12 | 1x43x33
13 | 1x71x1
14 | 1x71x33
15 | 1x128x1
16 | 1x128x33
17 | 33x1x1
18 | 33x1x33
19 | 33x3x1
20 | 33x3x33
21 | 33x7x1
22 | 33x7x33
23 | 33x9x1
24 | 33x9x33
25 | 33x19x1
26 | 33x19x33
27 | 33x43x1
28 | 33x43x33
29 | 33x71x1
30 | 33x71x33
31 | 33x128x1
32 | 33x128x33
33 |
--------------------------------------------------------------------------------
/tests/benchdnn/inputs/zeropad/test_zeropad_ci:
--------------------------------------------------------------------------------
1 | --dt=s8,f16,f32,f64
2 | --batch=set_dim1_block_3d
3 | --batch=set_dim2_block_3d
4 | --batch=set_dim1dim2_block_3d
5 | --batch=set_dim2dim3_block_4d
6 |
--------------------------------------------------------------------------------
/third_party/.clang-tidy:
--------------------------------------------------------------------------------
1 | Checks: '-*,misc-definitions-in-headers'
2 | CheckOptions:
3 | - { key: HeaderFileExtensions, value: "x" }
4 |
--------------------------------------------------------------------------------
/third_party/ittnotify/README.md:
--------------------------------------------------------------------------------
1 | This code is from [Intel(R) Instrumentation and Tracing Technology (ITT) and
2 | Just-In-Time (JIT) API](https://github.com/intel/ittapi)
3 |
4 | tag: 3.22.5
5 |
--------------------------------------------------------------------------------
/third_party/spdlog/README.md:
--------------------------------------------------------------------------------
1 | This code is from [spdlog](https://github.com/gabime/spdlog).
2 |
3 | tag: 1.15.1
4 |
--------------------------------------------------------------------------------
/third_party/spdlog/details/windows_include.h:
--------------------------------------------------------------------------------
1 | #pragma once
2 |
3 | #ifndef NOMINMAX
4 | #define NOMINMAX // prevent windows redefining min/max
5 | #endif
6 |
7 | #ifndef WIN32_LEAN_AND_MEAN
8 | #define WIN32_LEAN_AND_MEAN
9 | #endif
10 |
11 | #include <windows.h>
12 |
--------------------------------------------------------------------------------
/third_party/spdlog/fmt/bundled/core.h:
--------------------------------------------------------------------------------
1 | // This file is only provided for compatibility and may be removed in future
2 | // versions. Use fmt/base.h if you don't need fmt::format and fmt/format.h
3 | // otherwise.
4 |
5 | #include "format.h"
6 |
--------------------------------------------------------------------------------
/third_party/spdlog/formatter.h:
--------------------------------------------------------------------------------
1 | // Copyright(c) 2015-present, Gabi Melman & spdlog contributors.
2 | // Distributed under the MIT License (http://opensource.org/licenses/MIT)
3 |
4 | #pragma once
5 |
6 | #include <spdlog/details/log_msg.h>
7 | #include <spdlog/fmt/fmt.h>
8 |
9 | namespace spdlog {
10 |
11 | class formatter {
12 | public:
13 | virtual ~formatter() = default;
14 | virtual void format(const details::log_msg &msg, memory_buf_t &dest) = 0;
15 | virtual std::unique_ptr<formatter> clone() const = 0;
16 | };
17 | } // namespace spdlog
18 |
--------------------------------------------------------------------------------
/third_party/spdlog/version.h:
--------------------------------------------------------------------------------
1 | // Copyright(c) 2015-present, Gabi Melman & spdlog contributors.
2 | // Distributed under the MIT License (http://opensource.org/licenses/MIT)
3 |
4 | #pragma once
5 |
6 | #define SPDLOG_VER_MAJOR 1
7 | #define SPDLOG_VER_MINOR 15
8 | #define SPDLOG_VER_PATCH 1
9 |
10 | #define SPDLOG_TO_VERSION(major, minor, patch) (major * 10000 + minor * 100 + patch)
11 | #define SPDLOG_VERSION SPDLOG_TO_VERSION(SPDLOG_VER_MAJOR, SPDLOG_VER_MINOR, SPDLOG_VER_PATCH)
12 |
--------------------------------------------------------------------------------
/third_party/xbyak_aarch64/xbyak_aarch64/xbyak_aarch64_version.h:
--------------------------------------------------------------------------------
1 | static const int majorVersion = 1;
2 | static const int minorVersion = 1;
3 | static const int patchVersion = 2;
4 | static int getVersion() { return (majorVersion << 16) + (minorVersion << 8) + patchVersion; }
5 | static const char *getVersionString() { return "1.1.2"; }
6 |
--------------------------------------------------------------------------------