Refactor: Move fundamental data type headers to core/common (#26330)

Changming Sun · web-flow · commit e0569fd9e2fe · 2025-10-17T10:59:30.000-07:00
Move `endian.h`, `float16.h`, and `float8.h` from `core/framework/` to `core/common/` to avoid circular dependencies and improve architectural layering. ## Motivation These headers define fundamental data types that are used across multiple low-level libraries: - `onnxruntime_common` (foundation layer) - `onnxruntime_mlas` (math library, depends on common) - `onnxruntime_util` (utilities, depends on common) - `onnxruntime_graph` (graph IR, depends on common) Previously, these types were in `core/framework/`, which is part of the `onnxruntime_framework` library that sits at a higher architectural level. This created circular dependency issues since mlas uses the "float16.h" . ## Changes ### File Moves (3 files): - `include/onnxruntime/core/framework/endian.h` → `include/onnxruntime/core/common/endian.h` - `include/onnxruntime/core/framework/float16.h` → `include/onnxruntime/core/common/float16.h` - `include/onnxruntime/core/framework/float8.h` → `include/onnxruntime/core/common/float8.h` ### Include Path Updates (53 files): Updated all references from: - `core/framework/endian.h` → `core/common/endian.h` - `core/framework/float16.h` → `core/common/float16.h` - `core/framework/float8.h` → `core/common/float8.h` Affected components: - Contrib ops (CPU, CUDA, ROCm) - Core framework and utilities - Providers (CPU, CUDA, CANN, QNN, OpenVINO, MIGraphX) - Tests - Training code ## Architectural Benefits This change establishes clearer architectural boundaries: ``` Level 0 (Foundation): onnxruntime_common (includes endian, float16, float8) onnxruntime_mlas → depends on common Level 1 (Core): onnxruntime_util → depends on common onnxruntime_graph → depends on common Level 2 (Framework): onnxruntime_framework → depends on common ``` By placing fundamental types in `common`, we ensure: 1. No circular dependencies between library targets 2. Lower-level libraries can access these types without pulling in framework 3. Clear separation between fundamental types (common) and framework-specific types like int4, float4 (framework) This PR is split from #26317 as suggested by the reviewer.
diff --git a/include/onnxruntime/core/common/endian.h b/include/onnxruntime/core/common/endian.h
diff --git a/include/onnxruntime/core/common/float16.h b/include/onnxruntime/core/common/float16.h
@@ -4,9 +4,9 @@
 
 #include <math.h>
 
-#include "endian.h"
+#include "core/common/endian.h"
 #if defined(CUDA_VERSION) && CUDA_VERSION >= 11000
-#include "cuda_bf16.h"
+#include "cuda_bf16.h"  // from CUDA SDK
 #endif
 
 #if !defined(__CUDACC__) && !defined(__HIPCC__)
diff --git a/include/onnxruntime/core/common/float8.h b/include/onnxruntime/core/common/float8.h
@@ -16,7 +16,7 @@
 
 #if !defined(DISABLE_FLOAT8_TYPES)
 
-#include "endian.h"
+#include "core/common/endian.h"
 
 #if defined(__CUDACC__)
 // Needed for CUDA_VERSION check below
diff --git a/include/onnxruntime/core/framework/data_types.h b/include/onnxruntime/core/framework/data_types.h
@@ -12,9 +12,9 @@
 #include <gsl/gsl>
 #include "core/common/common.h"
 #include "core/common/exceptions.h"
-#include "core/framework/endian.h"
-#include "core/framework/float8.h"
-#include "core/framework/float16.h"
+#include "core/common/endian.h"
+#include "core/common/float8.h"
+#include "core/common/float16.h"
 #include "core/framework/int4.h"
 #include "core/framework/float4.h"
 #include "core/graph/onnx_protobuf.h"
diff --git a/include/onnxruntime/core/framework/to_tensor_proto_element_type.h b/include/onnxruntime/core/framework/to_tensor_proto_element_type.h
@@ -11,8 +11,8 @@
 #endif
 
 #include "core/framework/float4.h"
-#include "core/framework/float8.h"
-#include "core/framework/float16.h"
+#include "core/common/float8.h"
+#include "core/common/float16.h"
 #include "core/framework/int4.h"
 
 namespace onnxruntime {
diff --git a/onnxruntime/contrib_ops/cpu/moe/moe_cpu.cc b/onnxruntime/contrib_ops/cpu/moe/moe_cpu.cc
@@ -9,7 +9,7 @@
 #include "core/providers/cpu/math/gemm_helper.h"
 #include "core/util/math_cpuonly.h"
 #include "core/mlas/inc/mlas.h"
-#include "core/framework/float16.h"
+#include "core/common/float16.h"
 #include "core/framework/allocator.h"
 #include "core/platform/threadpool.h"
 #include "core/common/narrow.h"
diff --git a/onnxruntime/contrib_ops/cpu/moe/moe_quantization_cpu.cc b/onnxruntime/contrib_ops/cpu/moe/moe_quantization_cpu.cc
@@ -3,7 +3,7 @@
 
 #include "contrib_ops/cpu/moe/moe_quantization_cpu.h"
 #include "core/framework/allocator.h"
-#include "core/framework/float16.h"
+#include "core/common/float16.h"
 #include "core/mlas/inc/mlas.h"
 #include "core/mlas/inc/mlas_q4.h"
 #include "core/platform/threadpool.h"
diff --git a/onnxruntime/contrib_ops/cpu/quantization/dequantize_blockwise_bnb4.h b/onnxruntime/contrib_ops/cpu/quantization/dequantize_blockwise_bnb4.h
@@ -8,7 +8,7 @@
 #include <vector>
 
 #include "core/common/safeint.h"
-#include "core/framework/float16.h"
+#include "core/common/float16.h"
 #include "core/platform/threadpool.h"
 #include <iostream>
 
diff --git a/onnxruntime/contrib_ops/cpu/quantization/gather_block_quantized.cc b/onnxruntime/contrib_ops/cpu/quantization/gather_block_quantized.cc
@@ -7,7 +7,7 @@
 #include "core/common/common.h"
 #include "core/common/narrow.h"
 #include "core/common/safeint.h"
-#include "core/framework/float16.h"
+#include "core/common/float16.h"
 #include "core/framework/int4.h"
 #include "core/framework/op_kernel.h"
 #include "core/platform/threadpool.h"
diff --git a/onnxruntime/contrib_ops/cpu/quantization/matmul_nbits_impl.cc b/onnxruntime/contrib_ops/cpu/quantization/matmul_nbits_impl.cc
@@ -9,7 +9,7 @@
 #include <type_traits>
 
 #include "core/common/common.h"
-#include "core/framework/float16.h"
+#include "core/common/float16.h"
 #include "core/providers/common.h"
 #include "core/platform/threadpool.h"
 
diff --git a/onnxruntime/contrib_ops/cpu/utils/console_dumper.h b/onnxruntime/contrib_ops/cpu/utils/console_dumper.h
@@ -5,7 +5,7 @@
 #include <string>
 #include <iostream>
 #include "core/framework/ort_value.h"
-#include "core/framework/float16.h"
+#include "core/common/float16.h"
 #include "contrib_ops/cpu/utils/debug_macros.h"
 
 namespace onnxruntime {
diff --git a/onnxruntime/contrib_ops/cuda/quantization/matmul_nbits.cc b/onnxruntime/contrib_ops/cuda/quantization/matmul_nbits.cc
@@ -6,7 +6,7 @@
 #include <cstdint>
 
 #include "core/common/status.h"
-#include "core/framework/float16.h"
+#include "core/common/float16.h"
 #include "core/providers/cpu/math/matmul_helper.h"
 #include "core/providers/cuda/cuda_type_conversion.h"
 #include "contrib_ops/cuda/utils/dump_cuda_tensor.h"
diff --git a/onnxruntime/contrib_ops/rocm/bert/gemm_fast_gelu_impl.h b/onnxruntime/contrib_ops/rocm/bert/gemm_fast_gelu_impl.h
@@ -5,7 +5,7 @@
 
 #include "contrib_ops/rocm/bert/gemm_fast_gelu_common.h"
 #include "core/common/status.h"
-#include "core/framework/float16.h"
+#include "core/common/float16.h"
 
 namespace onnxruntime {
 namespace contrib {
diff --git a/onnxruntime/contrib_ops/rocm/math/gemm_float8.cu b/onnxruntime/contrib_ops/rocm/math/gemm_float8.cu
@@ -2,7 +2,7 @@
 // Licensed under the MIT License.
 
 #include "core/common/common.h"
-#include "core/framework/float16.h"
+#include "core/common/float16.h"
 #include "core/providers/rocm/rocm_kernel.h"
 #include "contrib_ops/rocm/math/gemm_float8_ck.cuh"
 
diff --git a/onnxruntime/contrib_ops/rocm/math/gemm_float8_ck.cuh b/onnxruntime/contrib_ops/rocm/math/gemm_float8_ck.cuh
@@ -19,7 +19,7 @@
 #endif
 
 #if !defined(DISABLE_FLOAT8_TYPES)
-#include "core/framework/float8.h"
+#include "core/common/float8.h"
 #endif
 #include "core/providers/rocm/tunable/gemm_common.h"
 
diff --git a/onnxruntime/core/framework/element_type_lists.h b/onnxruntime/core/framework/element_type_lists.h
@@ -9,8 +9,8 @@
 #include "boost/mp11.hpp"
 
 #include "core/common/type_list.h"
-#include "core/framework/float8.h"
-#include "core/framework/float16.h"
+#include "core/common/float8.h"
+#include "core/common/float16.h"
 #include "core/framework/int4.h"
 #include "core/framework/float4.h"
 
diff --git a/onnxruntime/core/framework/endian_utils.cc b/onnxruntime/core/framework/endian_utils.cc
@@ -6,7 +6,7 @@
 #include <cassert>
 #include <cstring>
 
-#include "core/framework/endian.h"
+#include "core/common/endian.h"
 
 namespace onnxruntime {
 namespace utils {
diff --git a/onnxruntime/core/framework/murmurhash3.cc b/onnxruntime/core/framework/murmurhash3.cc
@@ -15,7 +15,7 @@
 
 /* Modifications Copyright (c) Microsoft. */
 
-#include "core/framework/endian.h"
+#include "core/common/endian.h"
 
 #include "core/util/force_inline.h"
 
diff --git a/onnxruntime/core/graph/contrib_ops/onnx_function_util.cc b/onnxruntime/core/graph/contrib_ops/onnx_function_util.cc
@@ -1,7 +1,7 @@
 #include "core/graph/contrib_ops/onnx_function_util.h"
 #include "core/util/math.h"
-#include "core/framework/float8.h"
-#include "core/framework/float16.h"
+#include "core/common/float8.h"
+#include "core/common/float16.h"
 
 namespace ONNX_NAMESPACE {
 
diff --git a/onnxruntime/core/mlas/lib/mlasi.h b/onnxruntime/core/mlas/lib/mlasi.h
@@ -162,7 +162,7 @@ MLAS_FORCEINLINE void
 #include "core/common/cpuid_info.h"
 using MLAS_CPUIDINFO = onnxruntime::CPUIDInfo;
 
-#include "core/framework/float16.h"
+#include "core/common/float16.h"
 
 #else  // BUILD_MLAS_NO_ONNXRUNTIME
 
diff --git a/onnxruntime/core/providers/cann/cann_common.h b/onnxruntime/core/providers/cann/cann_common.h
@@ -6,7 +6,7 @@
 
 #include "core/providers/shared_library/provider_api.h"
 #include "core/providers/cann/cann_call.h"
-#include "core/framework/float16.h"
+#include "core/common/float16.h"
 
 namespace onnxruntime {
 namespace cann {
diff --git a/onnxruntime/core/providers/cpu/fp16/fp16_activations.h b/onnxruntime/core/providers/cpu/fp16/fp16_activations.h
@@ -4,7 +4,7 @@
 #pragma once
 
 #include "core/mlas/inc/mlas.h"
-#include "core/framework/float16.h"
+#include "core/common/float16.h"
 #include "core/providers/cpu/activation/activations.h"
 
 #ifdef MLAS_F16VEC_INTRINSICS_SUPPORTED
diff --git a/onnxruntime/core/providers/cpu/fp16/fp16_conv.cc b/onnxruntime/core/providers/cpu/fp16/fp16_conv.cc
@@ -10,7 +10,7 @@
 #ifdef MLAS_F16VEC_INTRINSICS_SUPPORTED
 
 #include "core/common/safeint.h"
-#include "core/framework/float16.h"
+#include "core/common/float16.h"
 #include "core/framework/op_kernel.h"
 #include "core/providers/cpu/nn/conv_attributes.h"
 
diff --git a/onnxruntime/core/providers/cpu/quantization/quantize_linear.cc b/onnxruntime/core/providers/cpu/quantization/quantize_linear.cc
@@ -4,8 +4,8 @@
 #include <type_traits>
 #include <core/common/safeint.h>
 #include "core/framework/element_type_lists.h"
-#include "core/framework/float8.h"
-#include "core/framework/float16.h"
+#include "core/common/float8.h"
+#include "core/common/float16.h"
 #include "core/framework/int4.h"
 #include "core/framework/op_kernel.h"
 #include "core/providers/common.h"
diff --git a/onnxruntime/core/providers/cuda/atomic/common.cuh b/onnxruntime/core/providers/cuda/atomic/common.cuh
@@ -20,7 +20,7 @@
 #include "cuda.h"
 #include "cuda_fp16.h"
 #include "cuda_runtime.h"
-#include "core/framework/float16.h"
+#include "core/common/float16.h"
 
 namespace onnxruntime {
 namespace cuda {
diff --git a/onnxruntime/core/providers/cuda/cuda_common.h b/onnxruntime/core/providers/cuda/cuda_common.h
@@ -27,8 +27,8 @@
 
 #include "core/providers/shared_library/provider_api.h"
 #include "core/common/status.h"
-#include "core/framework/float8.h"
-#include "core/framework/float16.h"
+#include "core/common/float8.h"
+#include "core/common/float16.h"
 #include "core/framework/float4.h"
 #include "core/providers/cuda/cuda_pch.h"
 #include "core/providers/cuda/shared_inc/cuda_call.h"
diff --git a/onnxruntime/core/providers/cuda/cuda_type_conversion.h b/onnxruntime/core/providers/cuda/cuda_type_conversion.h
@@ -26,8 +26,8 @@
 #include <type_traits>
 #include <cstdint>
 #include "core/framework/int4.h"
-#include "core/framework/float8.h"
-#include "core/framework/float16.h"
+#include "core/common/float8.h"
+#include "core/common/float16.h"
 #include "core/framework/float4.h"
 
 namespace onnxruntime {
diff --git a/onnxruntime/core/providers/cuda/math/unary_elementwise_ops_impl.h b/onnxruntime/core/providers/cuda/math/unary_elementwise_ops_impl.h
@@ -5,7 +5,7 @@
 
 #include <stdint.h>
 #include "core/providers/cuda/shared_inc/cuda_utils.h"
-#include "core/framework/float8.h"
+#include "core/common/float8.h"
 #include <cuda_runtime.h>
 
 namespace onnxruntime {
diff --git a/onnxruntime/core/providers/cuda/shared_inc/accumulation_type.h b/onnxruntime/core/providers/cuda/shared_inc/accumulation_type.h
@@ -4,7 +4,7 @@
 #pragma once
 
 #include <cuda_fp16.h>
-#include "core/framework/float16.h"
+#include "core/common/float16.h"
 
 namespace onnxruntime {
 namespace cuda {
diff --git a/onnxruntime/core/providers/cuda/shared_inc/cuda_utils.h b/onnxruntime/core/providers/cuda/shared_inc/cuda_utils.h
@@ -13,7 +13,7 @@
 #include <limits>
 
 #include <gsl/gsl>
-#include "core/framework/float16.h"
+#include "core/common/float16.h"
 #include "core/providers/cuda/shared_inc/fast_divmod.h"
 
 namespace onnxruntime {
diff --git a/onnxruntime/core/providers/migraphx/migraphx_allocator.cc b/onnxruntime/core/providers/migraphx/migraphx_allocator.cc
@@ -5,7 +5,7 @@
 #include "core/providers/migraphx/migraphx_call.h"
 #include "core/providers/migraphx/migraphx_allocator.h"
 #include "core/common/status.h"
-#include "core/framework/float16.h"
+#include "core/common/float16.h"
 #include "core/providers/migraphx/gpu_data_transfer.h"
 
 namespace onnxruntime {
diff --git a/onnxruntime/core/providers/openvino/qdq_transformations/qdq_scales_fix.cpp b/onnxruntime/core/providers/openvino/qdq_transformations/qdq_scales_fix.cpp
@@ -4,7 +4,7 @@
 #include "qdq_scales_fix.h"
 #include "core/providers/openvino/ov_protobuf_utils.h"
 #include "core/framework/ort_value.h"
-#include "core/framework/float16.h"
+#include "core/common/float16.h"
 
 #include <fstream>
 #include <list>
diff --git a/onnxruntime/core/providers/qnn/ort_api.h b/onnxruntime/core/providers/qnn/ort_api.h
@@ -26,7 +26,7 @@
 #include "core/graph/onnx_protobuf.h"
 #include "core/platform/env.h"
 #include "core/framework/data_types.h"
-#include "core/framework/float16.h"
+#include "core/common/float16.h"
 #include "core/framework/run_options.h"
 #include "core/framework/execution_provider.h"
 #include "core/framework/model_metadef_id_generator.h"
diff --git a/onnxruntime/core/providers/shared_library/provider_api.h b/onnxruntime/core/providers/shared_library/provider_api.h
@@ -27,8 +27,8 @@
 #include "core/common/type_list.h"
 #include "core/common/logging/severity.h"
 #include "core/framework/allocator.h"
-#include "core/framework/float8.h"
-#include "core/framework/float16.h"
+#include "core/common/float8.h"
+#include "core/common/float16.h"
 #include "core/framework/int4.h"
 #include "core/framework/float4.h"
 #include "core/framework/tensor_shape.h"
diff --git a/onnxruntime/core/util/math_cpu.cc b/onnxruntime/core/util/math_cpu.cc
@@ -17,7 +17,7 @@
 
 #include "core/util/math_cpuonly.h"
 #include "core/util/math.h"
-#include "core/framework/float16.h"
+#include "core/common/float16.h"
 
 #include <algorithm>
 #include <type_traits>
diff --git a/onnxruntime/core/util/qmath.h b/onnxruntime/core/util/qmath.h
@@ -7,7 +7,7 @@
 #include "core/platform/threadpool.h"
 #include "core/common/narrow.h"
 #include "core/framework/element_type_lists.h"
-#include "core/framework/float8.h"
+#include "core/common/float8.h"
 #include "core/framework/int4.h"
 #include <cmath>
 #include <limits>
diff --git a/onnxruntime/lora/adapter_format_utils.cc b/onnxruntime/lora/adapter_format_utils.cc
@@ -6,7 +6,7 @@
 
 #include "core/framework/allocator.h"
 #include "core/common/common.h"
-#include "core/framework/endian.h"
+#include "core/common/endian.h"
 #include "core/framework/endian_utils.h"
 #include "core/common/span_utils.h"
 #include "core/framework/ortdevice.h"
diff --git a/onnxruntime/test/common/random_generator.h b/onnxruntime/test/common/random_generator.h
@@ -13,7 +13,7 @@
 #include "core/common/narrow.h"
 #include "core/common/optional.h"
 #include "core/common/type_utils.h"
-#include "core/framework/float16.h"
+#include "core/common/float16.h"
 #include "core/framework/int4.h"
 #include "test/util/include/test_random_seed.h"
 
diff --git a/onnxruntime/test/framework/data_types_test.cc b/onnxruntime/test/framework/data_types_test.cc
@@ -7,8 +7,8 @@
 #include "core/common/inlined_containers.h"
 #include "core/framework/data_types.h"
 #include "core/framework/data_types_internal.h"
-#include "core/framework/float16.h"
-#include "core/framework/float8.h"
+#include "core/common/float16.h"
+#include "core/common/float8.h"
 #include "core/graph/onnx_protobuf.h"
 #include "gtest/gtest.h"
 
diff --git a/onnxruntime/test/framework/endian_test.cc b/onnxruntime/test/framework/endian_test.cc
@@ -1,4 +1,4 @@
-#include "core/framework/endian.h"
+#include "core/common/endian.h"
 #include "core/framework/endian_utils.h"
 #include "core/graph/onnx_protobuf.h"         // For TensorProto
 #include "core/framework/tensorprotoutils.h"  // For ConvertRawDataInTensorProto
diff --git a/onnxruntime/test/framework/float_8_test.cc b/onnxruntime/test/framework/float_8_test.cc
@@ -5,7 +5,7 @@
 
 #include <vector>
 
-#include "core/framework/float8.h"
+#include "core/common/float8.h"
 #include "test/capturing_sink.h"
 #include "test/test_environment.h"
 #include "test/unittest_util/framework_test_utils.h"
diff --git a/onnxruntime/test/mlas/bench/bench_rope.cpp b/onnxruntime/test/mlas/bench/bench_rope.cpp
@@ -4,7 +4,7 @@
 #include "mlas.h"
 #include "benchmark/benchmark.h"
 #include "bench_util.h"
-#include "core/framework/float16.h"
+#include "core/common/float16.h"
 
 using namespace onnxruntime;
 
diff --git a/onnxruntime/test/mlas/bench/bench_util.h b/onnxruntime/test/mlas/bench/bench_util.h
@@ -8,7 +8,7 @@
 #include <functional>
 #include <random>
 
-#include "core/framework/float16.h"
+#include "core/common/float16.h"
 #include "core/mlas/inc/mlas.h"
 
 template <typename ElementType>
diff --git a/onnxruntime/test/onnx/tensorprotoutils.cc b/onnxruntime/test/onnx/tensorprotoutils.cc
diff --git a/onnxruntime/test/optimizer/test_optimizer_utils.h b/onnxruntime/test/optimizer/test_optimizer_utils.h
diff --git a/onnxruntime/test/providers/cuda/test_cases/blkq4_fp16_gemm_sm80_test.cc b/onnxruntime/test/providers/cuda/test_cases/blkq4_fp16_gemm_sm80_test.cc
diff --git a/onnxruntime/test/providers/openvino/openvino_ep_bfloat16_pass_test.cc b/onnxruntime/test/providers/openvino/openvino_ep_bfloat16_pass_test.cc
diff --git a/onnxruntime/test/providers/openvino/openvino_ep_context_test.cc b/onnxruntime/test/providers/openvino/openvino_ep_context_test.cc
diff --git a/onnxruntime/test/providers/qnn/batch_norm_test.cc b/onnxruntime/test/providers/qnn/batch_norm_test.cc
diff --git a/onnxruntime/test/providers/qnn/cast_test.cc b/onnxruntime/test/providers/qnn/cast_test.cc
diff --git a/onnxruntime/test/providers/qnn/qnn_test_utils.h b/onnxruntime/test/providers/qnn/qnn_test_utils.h
diff --git a/onnxruntime/test/unittest_util/conversion.h b/onnxruntime/test/unittest_util/conversion.h
diff --git a/onnxruntime/test/unittest_util/function_test_util.h b/onnxruntime/test/unittest_util/function_test_util.h
diff --git a/orttraining/orttraining/core/graph/gradient_builder_base.h b/orttraining/orttraining/core/graph/gradient_builder_base.h
