From 97123eff9496776d31cac2131ad65f07e9614f95 Mon Sep 17 00:00:00 2001 From: Pete Warden Date: Tue, 25 Jun 2019 15:02:38 -0700 Subject: [PATCH] Fix for SparkFun Edge build problem The lite/internal/common.h header is used by reference kernel implementations, which are used by the Micro build. This means it can't directly or indirectly include headers like pthreads that aren't available on embedded platforms. A recent change added an indirect dependency on another header which does, so this modularization ensures that only the required portion is included as a separate file, avoiding the dependency. PiperOrigin-RevId: 255055378 --- .../experimental/micro/tools/make/Makefile | 2 +- tensorflow/lite/kernels/internal/BUILD | 1 + tensorflow/lite/kernels/internal/common.h | 2 +- .../kernels/internal/optimized/cpu_check.h | 29 +----------- .../kernels/internal/optimized/neon_check.h | 46 +++++++++++++++++++ 5 files changed, 50 insertions(+), 30 deletions(-) create mode 100644 tensorflow/lite/kernels/internal/optimized/neon_check.h diff --git a/tensorflow/lite/experimental/micro/tools/make/Makefile b/tensorflow/lite/experimental/micro/tools/make/Makefile index 6e100987b7d..8f5c8842021 100644 --- a/tensorflow/lite/experimental/micro/tools/make/Makefile +++ b/tensorflow/lite/experimental/micro/tools/make/Makefile @@ -105,7 +105,7 @@ tensorflow/lite/kernels/kernel_util.h \ tensorflow/lite/kernels/op_macros.h \ tensorflow/lite/kernels/padding.h \ tensorflow/lite/kernels/internal/common.h \ -tensorflow/lite/kernels/internal/compatibility.h \ +tensorflow/lite/kernels/internal/optimized/neon_check.h \ tensorflow/lite/kernels/internal/reference/conv.h \ tensorflow/lite/kernels/internal/reference/depthwiseconv_float.h \ tensorflow/lite/kernels/internal/reference/depthwiseconv_uint8.h \ diff --git a/tensorflow/lite/kernels/internal/BUILD b/tensorflow/lite/kernels/internal/BUILD index 16d7f926793..44f2d023b73 100644 --- a/tensorflow/lite/kernels/internal/BUILD +++ b/tensorflow/lite/kernels/internal/BUILD @@ -807,6 +807,7 @@ cc_library( name = "cpu_check", hdrs = [ "optimized/cpu_check.h", + "optimized/neon_check.h", ], deps = [ "//tensorflow/lite/kernels:cpu_backend_context", diff --git a/tensorflow/lite/kernels/internal/common.h b/tensorflow/lite/kernels/internal/common.h index 8ba2d92a315..0c4fbc1e84e 100644 --- a/tensorflow/lite/kernels/internal/common.h +++ b/tensorflow/lite/kernels/internal/common.h @@ -22,7 +22,7 @@ limitations under the License. #endif #include "fixedpoint/fixedpoint.h" -#include "tensorflow/lite/kernels/internal/optimized/cpu_check.h" +#include "tensorflow/lite/kernels/internal/optimized/neon_check.h" #include "tensorflow/lite/kernels/internal/types.h" namespace tflite { diff --git a/tensorflow/lite/kernels/internal/optimized/cpu_check.h b/tensorflow/lite/kernels/internal/optimized/cpu_check.h index 3847d3a0634..2c6a682f3b2 100644 --- a/tensorflow/lite/kernels/internal/optimized/cpu_check.h +++ b/tensorflow/lite/kernels/internal/optimized/cpu_check.h @@ -16,22 +16,7 @@ limitations under the License. #define TENSORFLOW_LITE_KERNELS_INTERNAL_OPTIMIZED_CPU_CHECK_H_ #include "tensorflow/lite/kernels/cpu_backend_context.h" - -#if defined(__ARM_NEON__) || defined(__ARM_NEON) -#define USE_NEON -#include -#endif // __ARM_NEON - -#if defined __GNUC__ && defined __SSE4_1__ && !defined TF_LITE_DISABLE_X86_NEON -#define USE_NEON -#pragma GCC diagnostic push -#pragma GCC diagnostic ignored "-Wdeprecated-declarations" -#pragma GCC diagnostic ignored "-Wattributes" -#pragma GCC diagnostic ignored "-Wnarrowing" -#pragma GCC diagnostic ignored "-Wsequence-point" -#include "NEON_2_SSE.h" -#pragma GCC diagnostic pop -#endif // __SSE4_1__ +#include "tensorflow/lite/kernels/internal/optimized/neon_check.h" namespace tflite { @@ -49,16 +34,4 @@ inline void GetCpuFlags(CpuBackendContext* cpu_backend_context, } // namespace tflite -// NEON_OR_PORTABLE(SomeFunc, args) calls NeonSomeFunc(args) if USE_NEON is -// defined, PortableSomeFunc(args) otherwise. -#ifdef USE_NEON -// Always use Neon code -#define NEON_OR_PORTABLE(funcname, ...) Neon##funcname(__VA_ARGS__) - -#else -// No NEON available: Use Portable code -#define NEON_OR_PORTABLE(funcname, ...) Portable##funcname(__VA_ARGS__) - -#endif // defined(USE_NEON) - #endif // TENSORFLOW_LITE_KERNELS_INTERNAL_OPTIMIZED_CPU_CHECK_H_ diff --git a/tensorflow/lite/kernels/internal/optimized/neon_check.h b/tensorflow/lite/kernels/internal/optimized/neon_check.h new file mode 100644 index 00000000000..a72af90f52b --- /dev/null +++ b/tensorflow/lite/kernels/internal/optimized/neon_check.h @@ -0,0 +1,46 @@ +/* Copyright 2019 The TensorFlow Authors. All Rights Reserved. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. +==============================================================================*/ +#ifndef TENSORFLOW_LITE_KERNELS_INTERNAL_OPTIMIZED_NEON_CHECK_H_ +#define TENSORFLOW_LITE_KERNELS_INTERNAL_OPTIMIZED_NEON_CHECK_H_ + +#if defined(__ARM_NEON__) || defined(__ARM_NEON) +#define USE_NEON +#include +#endif + +#if defined __GNUC__ && defined __SSE4_1__ && !defined TF_LITE_DISABLE_X86_NEON +#define USE_NEON +#pragma GCC diagnostic push +#pragma GCC diagnostic ignored "-Wdeprecated-declarations" +#pragma GCC diagnostic ignored "-Wattributes" +#pragma GCC diagnostic ignored "-Wnarrowing" +#pragma GCC diagnostic ignored "-Wsequence-point" +#include "NEON_2_SSE.h" +#pragma GCC diagnostic pop +#endif + +// NEON_OR_PORTABLE(SomeFunc, args) calls NeonSomeFunc(args) if USE_NEON is +// defined, PortableSomeFunc(args) otherwise. +#ifdef USE_NEON +// Always use Neon code +#define NEON_OR_PORTABLE(funcname, ...) Neon##funcname(__VA_ARGS__) + +#else +// No NEON available: Use Portable code +#define NEON_OR_PORTABLE(funcname, ...) Portable##funcname(__VA_ARGS__) + +#endif // defined(USE_NEON) + +#endif // TENSORFLOW_LITE_KERNELS_INTERNAL_OPTIMIZED_NEON_CHECK_H_