From ee975de77b547554c99788e30a1779ebc03d6c8c Mon Sep 17 00:00:00 2001 From: Yufeng Li Date: Wed, 15 Dec 2021 15:45:04 -0800 Subject: [PATCH] reorganize quantization files (#10023) * reorganize quantization files --- .../contrib_ops/cpu/quantization/dynamic_quantize_matmul.cc | 2 +- .../contrib_ops/cpu/{ => quantization}/matmul_integer16.cc | 2 +- .../contrib_ops/cpu/{ => quantization}/matmul_integer16.h | 0 .../cpu/{bert => quantization}/qembed_layer_norm.cc | 2 +- .../cpu/{bert => quantization}/qembed_layer_norm.h | 2 +- .../contrib_ops/cpu/{ => quantization}/qlinear_activations.cc | 2 +- .../contrib_ops/cpu/{ => quantization}/qlinear_activations.h | 0 .../contrib_ops/cpu/{ => quantization}/qlinear_binary_op.cc | 0 .../contrib_ops/cpu/{ => quantization}/qlinear_binary_op.h | 0 .../contrib_ops/cpu/{ => quantization}/qlinear_concat.cc | 3 ++- .../contrib_ops/cpu/{ => quantization}/qlinear_concat.h | 0 .../cpu/{ => quantization}/qlinear_global_average_pool.cc | 0 .../cpu/{ => quantization}/qlinear_global_average_pool.h | 0 .../cpu/{ => quantization}/qlinear_lookup_table.cc | 0 .../contrib_ops/cpu/{ => quantization}/qlinear_lookup_table.h | 0 .../contrib_ops/cpu/{ => quantization}/qlinear_pool.cc | 4 ++-- onnxruntime/contrib_ops/cpu/{ => quantization}/qlinear_pool.h | 0 onnxruntime/contrib_ops/cpu/quantization/quant_gemm.cc | 2 +- onnxruntime/contrib_ops/cpu/quantization/quantize_ops.cc | 2 +- .../core/providers/cpu/{nn => quantization}/conv_integer.cc | 0 .../cpu/{tensor => quantization}/dynamicquantizelinear.cc | 0 .../cpu/{tensor => quantization}/dynamicquantizelinear.h | 0 .../providers/cpu/{math => quantization}/matmul_integer.cc | 0 .../cpu/{math => quantization}/matmul_integer_base.h | 0 .../core/providers/cpu/{nn => quantization}/qlinearconv.cc | 0 .../providers/cpu/{tensor => quantization}/quantize_linear.cc | 2 +- .../providers/cpu/{tensor => quantization}/quantize_linear.h | 0 .../cpu/{math => quantization}/quantize_linear_matmul.cc | 0 .../cpu/{math => quantization}/quantize_linear_matmul.h | 0 .../test/providers/cpu/math/quantize_linear_matmul_test.cc | 2 +- 30 files changed, 13 insertions(+), 12 deletions(-) rename onnxruntime/contrib_ops/cpu/{ => quantization}/matmul_integer16.cc (97%) rename onnxruntime/contrib_ops/cpu/{ => quantization}/matmul_integer16.h (100%) rename onnxruntime/contrib_ops/cpu/{bert => quantization}/qembed_layer_norm.cc (99%) rename onnxruntime/contrib_ops/cpu/{bert => quantization}/qembed_layer_norm.h (91%) rename onnxruntime/contrib_ops/cpu/{ => quantization}/qlinear_activations.cc (100%) rename onnxruntime/contrib_ops/cpu/{ => quantization}/qlinear_activations.h (100%) rename onnxruntime/contrib_ops/cpu/{ => quantization}/qlinear_binary_op.cc (100%) rename onnxruntime/contrib_ops/cpu/{ => quantization}/qlinear_binary_op.h (100%) rename onnxruntime/contrib_ops/cpu/{ => quantization}/qlinear_concat.cc (99%) rename onnxruntime/contrib_ops/cpu/{ => quantization}/qlinear_concat.h (100%) rename onnxruntime/contrib_ops/cpu/{ => quantization}/qlinear_global_average_pool.cc (100%) rename onnxruntime/contrib_ops/cpu/{ => quantization}/qlinear_global_average_pool.h (100%) rename onnxruntime/contrib_ops/cpu/{ => quantization}/qlinear_lookup_table.cc (100%) rename onnxruntime/contrib_ops/cpu/{ => quantization}/qlinear_lookup_table.h (100%) rename onnxruntime/contrib_ops/cpu/{ => quantization}/qlinear_pool.cc (99%) rename onnxruntime/contrib_ops/cpu/{ => quantization}/qlinear_pool.h (100%) rename onnxruntime/core/providers/cpu/{nn => quantization}/conv_integer.cc (100%) rename onnxruntime/core/providers/cpu/{tensor => quantization}/dynamicquantizelinear.cc (100%) rename onnxruntime/core/providers/cpu/{tensor => quantization}/dynamicquantizelinear.h (100%) rename onnxruntime/core/providers/cpu/{math => quantization}/matmul_integer.cc (100%) rename onnxruntime/core/providers/cpu/{math => quantization}/matmul_integer_base.h (100%) rename onnxruntime/core/providers/cpu/{nn => quantization}/qlinearconv.cc (100%) rename onnxruntime/core/providers/cpu/{tensor => quantization}/quantize_linear.cc (99%) rename onnxruntime/core/providers/cpu/{tensor => quantization}/quantize_linear.h (100%) rename onnxruntime/core/providers/cpu/{math => quantization}/quantize_linear_matmul.cc (100%) rename onnxruntime/core/providers/cpu/{math => quantization}/quantize_linear_matmul.h (100%) diff --git a/onnxruntime/contrib_ops/cpu/quantization/dynamic_quantize_matmul.cc b/onnxruntime/contrib_ops/cpu/quantization/dynamic_quantize_matmul.cc index a9d817b5e8..b0d1c3aec8 100644 --- a/onnxruntime/contrib_ops/cpu/quantization/dynamic_quantize_matmul.cc +++ b/onnxruntime/contrib_ops/cpu/quantization/dynamic_quantize_matmul.cc @@ -5,7 +5,7 @@ #include "core/mlas/inc/mlas.h" #include "core/providers/cpu/math/element_wise_ops.h" #include "core/providers/cpu/math/matmul_helper.h" -#include "core/providers/cpu/math/matmul_integer_base.h" +#include "core/providers/cpu/quantization/matmul_integer_base.h" #include "core/util/math_cpuonly.h" #include "core/util/qmath.h" diff --git a/onnxruntime/contrib_ops/cpu/matmul_integer16.cc b/onnxruntime/contrib_ops/cpu/quantization/matmul_integer16.cc similarity index 97% rename from onnxruntime/contrib_ops/cpu/matmul_integer16.cc rename to onnxruntime/contrib_ops/cpu/quantization/matmul_integer16.cc index d94914cf41..64a307530b 100644 --- a/onnxruntime/contrib_ops/cpu/matmul_integer16.cc +++ b/onnxruntime/contrib_ops/cpu/quantization/matmul_integer16.cc @@ -1,7 +1,7 @@ // Copyright (c) Microsoft Corporation. All rights reserved. // Licensed under the MIT License. -#include "contrib_ops/cpu/matmul_integer16.h" +#include "matmul_integer16.h" #include "core/providers/cpu/math/matmul_helper.h" namespace onnxruntime { diff --git a/onnxruntime/contrib_ops/cpu/matmul_integer16.h b/onnxruntime/contrib_ops/cpu/quantization/matmul_integer16.h similarity index 100% rename from onnxruntime/contrib_ops/cpu/matmul_integer16.h rename to onnxruntime/contrib_ops/cpu/quantization/matmul_integer16.h diff --git a/onnxruntime/contrib_ops/cpu/bert/qembed_layer_norm.cc b/onnxruntime/contrib_ops/cpu/quantization/qembed_layer_norm.cc similarity index 99% rename from onnxruntime/contrib_ops/cpu/bert/qembed_layer_norm.cc rename to onnxruntime/contrib_ops/cpu/quantization/qembed_layer_norm.cc index fee4d5a169..2f684362e1 100644 --- a/onnxruntime/contrib_ops/cpu/bert/qembed_layer_norm.cc +++ b/onnxruntime/contrib_ops/cpu/quantization/qembed_layer_norm.cc @@ -5,7 +5,7 @@ #include -#include "embed_layer_norm_helper.h" +#include "contrib_ops/cpu/bert/embed_layer_norm_helper.h" #include "core/framework/op_kernel.h" #include "core/providers/common.h" #include "core/quantization/quantization.h" diff --git a/onnxruntime/contrib_ops/cpu/bert/qembed_layer_norm.h b/onnxruntime/contrib_ops/cpu/quantization/qembed_layer_norm.h similarity index 91% rename from onnxruntime/contrib_ops/cpu/bert/qembed_layer_norm.h rename to onnxruntime/contrib_ops/cpu/quantization/qembed_layer_norm.h index 1788c2e2c6..baab2cac7c 100644 --- a/onnxruntime/contrib_ops/cpu/bert/qembed_layer_norm.h +++ b/onnxruntime/contrib_ops/cpu/quantization/qembed_layer_norm.h @@ -3,7 +3,7 @@ #pragma once -#include "embed_layer_norm.h" +#include "contrib_ops/cpu/bert/embed_layer_norm.h" #include "core/common/common.h" #include "core/framework/op_kernel.h" diff --git a/onnxruntime/contrib_ops/cpu/qlinear_activations.cc b/onnxruntime/contrib_ops/cpu/quantization/qlinear_activations.cc similarity index 100% rename from onnxruntime/contrib_ops/cpu/qlinear_activations.cc rename to onnxruntime/contrib_ops/cpu/quantization/qlinear_activations.cc index b1e9778373..3bb663c77f 100644 --- a/onnxruntime/contrib_ops/cpu/qlinear_activations.cc +++ b/onnxruntime/contrib_ops/cpu/quantization/qlinear_activations.cc @@ -2,8 +2,8 @@ // Licensed under the MIT License. #include "qlinear_activations.h" - #include "qlinear_lookup_table.h" + #include "core/mlas/inc/mlas.h" #include "core/platform/threadpool.h" diff --git a/onnxruntime/contrib_ops/cpu/qlinear_activations.h b/onnxruntime/contrib_ops/cpu/quantization/qlinear_activations.h similarity index 100% rename from onnxruntime/contrib_ops/cpu/qlinear_activations.h rename to onnxruntime/contrib_ops/cpu/quantization/qlinear_activations.h diff --git a/onnxruntime/contrib_ops/cpu/qlinear_binary_op.cc b/onnxruntime/contrib_ops/cpu/quantization/qlinear_binary_op.cc similarity index 100% rename from onnxruntime/contrib_ops/cpu/qlinear_binary_op.cc rename to onnxruntime/contrib_ops/cpu/quantization/qlinear_binary_op.cc diff --git a/onnxruntime/contrib_ops/cpu/qlinear_binary_op.h b/onnxruntime/contrib_ops/cpu/quantization/qlinear_binary_op.h similarity index 100% rename from onnxruntime/contrib_ops/cpu/qlinear_binary_op.h rename to onnxruntime/contrib_ops/cpu/quantization/qlinear_binary_op.h diff --git a/onnxruntime/contrib_ops/cpu/qlinear_concat.cc b/onnxruntime/contrib_ops/cpu/quantization/qlinear_concat.cc similarity index 99% rename from onnxruntime/contrib_ops/cpu/qlinear_concat.cc rename to onnxruntime/contrib_ops/cpu/quantization/qlinear_concat.cc index 2adeb4d697..76288daa8f 100644 --- a/onnxruntime/contrib_ops/cpu/qlinear_concat.cc +++ b/onnxruntime/contrib_ops/cpu/quantization/qlinear_concat.cc @@ -2,7 +2,8 @@ // Licensed under the MIT License. #include "qlinear_concat.h" -#include "contrib_ops/cpu/qlinear_lookup_table.h" +#include "qlinear_lookup_table.h" + #include "core/providers/common.h" #include "core/mlas/inc/mlas.h" #include "core/platform/threadpool.h" diff --git a/onnxruntime/contrib_ops/cpu/qlinear_concat.h b/onnxruntime/contrib_ops/cpu/quantization/qlinear_concat.h similarity index 100% rename from onnxruntime/contrib_ops/cpu/qlinear_concat.h rename to onnxruntime/contrib_ops/cpu/quantization/qlinear_concat.h diff --git a/onnxruntime/contrib_ops/cpu/qlinear_global_average_pool.cc b/onnxruntime/contrib_ops/cpu/quantization/qlinear_global_average_pool.cc similarity index 100% rename from onnxruntime/contrib_ops/cpu/qlinear_global_average_pool.cc rename to onnxruntime/contrib_ops/cpu/quantization/qlinear_global_average_pool.cc diff --git a/onnxruntime/contrib_ops/cpu/qlinear_global_average_pool.h b/onnxruntime/contrib_ops/cpu/quantization/qlinear_global_average_pool.h similarity index 100% rename from onnxruntime/contrib_ops/cpu/qlinear_global_average_pool.h rename to onnxruntime/contrib_ops/cpu/quantization/qlinear_global_average_pool.h diff --git a/onnxruntime/contrib_ops/cpu/qlinear_lookup_table.cc b/onnxruntime/contrib_ops/cpu/quantization/qlinear_lookup_table.cc similarity index 100% rename from onnxruntime/contrib_ops/cpu/qlinear_lookup_table.cc rename to onnxruntime/contrib_ops/cpu/quantization/qlinear_lookup_table.cc diff --git a/onnxruntime/contrib_ops/cpu/qlinear_lookup_table.h b/onnxruntime/contrib_ops/cpu/quantization/qlinear_lookup_table.h similarity index 100% rename from onnxruntime/contrib_ops/cpu/qlinear_lookup_table.h rename to onnxruntime/contrib_ops/cpu/quantization/qlinear_lookup_table.h diff --git a/onnxruntime/contrib_ops/cpu/qlinear_pool.cc b/onnxruntime/contrib_ops/cpu/quantization/qlinear_pool.cc similarity index 99% rename from onnxruntime/contrib_ops/cpu/qlinear_pool.cc rename to onnxruntime/contrib_ops/cpu/quantization/qlinear_pool.cc index 6953206a70..e97b654994 100644 --- a/onnxruntime/contrib_ops/cpu/qlinear_pool.cc +++ b/onnxruntime/contrib_ops/cpu/quantization/qlinear_pool.cc @@ -1,10 +1,10 @@ // Copyright (c) Microsoft Corporation. All rights reserved. // Licensed under the MIT License. +#include "qlinear_global_average_pool.h" +#include "qlinear_lookup_table.h" #include "qlinear_pool.h" -#include "contrib_ops/cpu/qlinear_lookup_table.h" -#include "contrib_ops/cpu/qlinear_global_average_pool.h" #include "core/common/safeint.h" #include "core/util/math_cpuonly.h" #include "core/providers/common.h" diff --git a/onnxruntime/contrib_ops/cpu/qlinear_pool.h b/onnxruntime/contrib_ops/cpu/quantization/qlinear_pool.h similarity index 100% rename from onnxruntime/contrib_ops/cpu/qlinear_pool.h rename to onnxruntime/contrib_ops/cpu/quantization/qlinear_pool.h diff --git a/onnxruntime/contrib_ops/cpu/quantization/quant_gemm.cc b/onnxruntime/contrib_ops/cpu/quantization/quant_gemm.cc index 523d80f441..9bc7b8e769 100644 --- a/onnxruntime/contrib_ops/cpu/quantization/quant_gemm.cc +++ b/onnxruntime/contrib_ops/cpu/quantization/quant_gemm.cc @@ -4,7 +4,7 @@ #include "core/common/safeint.h" #include "core/providers/cpu/math/gemm_base.h" #include "core/providers/cpu/math/gemm_helper.h" -#include "core/providers/cpu/math/matmul_integer_base.h" +#include "core/providers/cpu/quantization/matmul_integer_base.h" #include "core/quantization/quantization.h" #include "core/util/math_cpuonly.h" diff --git a/onnxruntime/contrib_ops/cpu/quantization/quantize_ops.cc b/onnxruntime/contrib_ops/cpu/quantization/quantize_ops.cc index 46587733c4..9d2931e9c9 100644 --- a/onnxruntime/contrib_ops/cpu/quantization/quantize_ops.cc +++ b/onnxruntime/contrib_ops/cpu/quantization/quantize_ops.cc @@ -1,7 +1,7 @@ // Copyright (c) Microsoft Corporation. All rights reserved. // Licensed under the MIT License. -#include "core/providers/cpu/tensor/quantize_linear.h" +#include "core/providers/cpu/quantization/quantize_linear.h" #include "core/providers/common.h" namespace onnxruntime { diff --git a/onnxruntime/core/providers/cpu/nn/conv_integer.cc b/onnxruntime/core/providers/cpu/quantization/conv_integer.cc similarity index 100% rename from onnxruntime/core/providers/cpu/nn/conv_integer.cc rename to onnxruntime/core/providers/cpu/quantization/conv_integer.cc diff --git a/onnxruntime/core/providers/cpu/tensor/dynamicquantizelinear.cc b/onnxruntime/core/providers/cpu/quantization/dynamicquantizelinear.cc similarity index 100% rename from onnxruntime/core/providers/cpu/tensor/dynamicquantizelinear.cc rename to onnxruntime/core/providers/cpu/quantization/dynamicquantizelinear.cc diff --git a/onnxruntime/core/providers/cpu/tensor/dynamicquantizelinear.h b/onnxruntime/core/providers/cpu/quantization/dynamicquantizelinear.h similarity index 100% rename from onnxruntime/core/providers/cpu/tensor/dynamicquantizelinear.h rename to onnxruntime/core/providers/cpu/quantization/dynamicquantizelinear.h diff --git a/onnxruntime/core/providers/cpu/math/matmul_integer.cc b/onnxruntime/core/providers/cpu/quantization/matmul_integer.cc similarity index 100% rename from onnxruntime/core/providers/cpu/math/matmul_integer.cc rename to onnxruntime/core/providers/cpu/quantization/matmul_integer.cc diff --git a/onnxruntime/core/providers/cpu/math/matmul_integer_base.h b/onnxruntime/core/providers/cpu/quantization/matmul_integer_base.h similarity index 100% rename from onnxruntime/core/providers/cpu/math/matmul_integer_base.h rename to onnxruntime/core/providers/cpu/quantization/matmul_integer_base.h diff --git a/onnxruntime/core/providers/cpu/nn/qlinearconv.cc b/onnxruntime/core/providers/cpu/quantization/qlinearconv.cc similarity index 100% rename from onnxruntime/core/providers/cpu/nn/qlinearconv.cc rename to onnxruntime/core/providers/cpu/quantization/qlinearconv.cc diff --git a/onnxruntime/core/providers/cpu/tensor/quantize_linear.cc b/onnxruntime/core/providers/cpu/quantization/quantize_linear.cc similarity index 99% rename from onnxruntime/core/providers/cpu/tensor/quantize_linear.cc rename to onnxruntime/core/providers/cpu/quantization/quantize_linear.cc index 76b61cb772..595b424973 100644 --- a/onnxruntime/core/providers/cpu/tensor/quantize_linear.cc +++ b/onnxruntime/core/providers/cpu/quantization/quantize_linear.cc @@ -1,7 +1,7 @@ // Copyright (c) Microsoft Corporation. All rights reserved. // Licensed under the MIT License. -#include "core/providers/cpu/tensor/quantize_linear.h" +#include "core/providers/cpu/quantization/quantize_linear.h" #include "core/providers/common.h" #include "core/mlas/inc/mlas.h" #include "core/util/qmath.h" diff --git a/onnxruntime/core/providers/cpu/tensor/quantize_linear.h b/onnxruntime/core/providers/cpu/quantization/quantize_linear.h similarity index 100% rename from onnxruntime/core/providers/cpu/tensor/quantize_linear.h rename to onnxruntime/core/providers/cpu/quantization/quantize_linear.h diff --git a/onnxruntime/core/providers/cpu/math/quantize_linear_matmul.cc b/onnxruntime/core/providers/cpu/quantization/quantize_linear_matmul.cc similarity index 100% rename from onnxruntime/core/providers/cpu/math/quantize_linear_matmul.cc rename to onnxruntime/core/providers/cpu/quantization/quantize_linear_matmul.cc diff --git a/onnxruntime/core/providers/cpu/math/quantize_linear_matmul.h b/onnxruntime/core/providers/cpu/quantization/quantize_linear_matmul.h similarity index 100% rename from onnxruntime/core/providers/cpu/math/quantize_linear_matmul.h rename to onnxruntime/core/providers/cpu/quantization/quantize_linear_matmul.h diff --git a/onnxruntime/test/providers/cpu/math/quantize_linear_matmul_test.cc b/onnxruntime/test/providers/cpu/math/quantize_linear_matmul_test.cc index 090feb09ae..aa5e103874 100644 --- a/onnxruntime/test/providers/cpu/math/quantize_linear_matmul_test.cc +++ b/onnxruntime/test/providers/cpu/math/quantize_linear_matmul_test.cc @@ -1,7 +1,7 @@ // Copyright (c) Microsoft Corporation. All rights reserved. // Licensed under the MIT License. -#include "core/providers/cpu/math/quantize_linear_matmul.h" +#include "core/providers/cpu/quantization/quantize_linear_matmul.h" #include "core/mlas/inc/mlas.h" #include "gtest/gtest.h"