diff options
Diffstat (limited to 'tensorflow/contrib/lite')
19 files changed, 154 insertions, 35 deletions
diff --git a/tensorflow/contrib/lite/Makefile b/tensorflow/contrib/lite/Makefile index b4504f246a..65fba52d46 100644 --- a/tensorflow/contrib/lite/Makefile +++ b/tensorflow/contrib/lite/Makefile @@ -90,7 +90,8 @@ $(wildcard tensorflow/contrib/lite/kernels/*.c) \ $(wildcard tensorflow/contrib/lite/kernels/internal/*.c) \ $(wildcard tensorflow/contrib/lite/kernels/internal/optimized/*.c) \ $(wildcard tensorflow/contrib/lite/kernels/internal/reference/*.c) \ -$(wildcard tensorflow/contrib/lite/downloads/farmhash/src/farmhash.cc) +$(wildcard tensorflow/contrib/lite/downloads/farmhash/src/farmhash.cc) \ +$(wildcard tensorflow/contrib/lite/downloads/fft2d/fftsg.c) # Remove any duplicates. CORE_CC_ALL_SRCS := $(sort $(CORE_CC_ALL_SRCS)) CORE_CC_EXCLUDE_SRCS := \ diff --git a/tensorflow/contrib/lite/download_dependencies.sh b/tensorflow/contrib/lite/download_dependencies.sh index a93ed201d6..436c3e1d4c 100755 --- a/tensorflow/contrib/lite/download_dependencies.sh +++ b/tensorflow/contrib/lite/download_dependencies.sh @@ -30,12 +30,15 @@ if [ ! -f $BZL_FILE_PATH ]; then fi EIGEN_URL="$(grep -o 'http.*bitbucket.org/eigen/eigen/get/.*tar\.gz' "${BZL_FILE_PATH}" | grep -v mirror.bazel | head -n1)" -GEMMLOWP_URL="$(grep -o 'https://mirror.bazel.build/github.com/google/gemmlowp/.*zip' "${BZL_FILE_PATH}" | head -n1)" +# TODO (yongtang): Replace the following with 'https://mirror.bazel.build/github.com/google/gemmlowp/.*zip' once +# the archive has been propagated in mirror.bazel.build. +GEMMLOWP_URL="$(grep -o 'https://github.com/google/gemmlowp/.*zip' "${BZL_FILE_PATH}" | head -n1)" GOOGLETEST_URL="https://github.com/google/googletest/archive/release-1.8.0.tar.gz" ABSL_URL="$(grep -o 'https://github.com/abseil/abseil-cpp/.*tar.gz' "${BZL_FILE_PATH}" | head -n1)" NEON_2_SSE_URL="https://github.com/intel/ARM_NEON_2_x86_SSE/archive/master.zip" FARMHASH_URL="https://mirror.bazel.build/github.com/google/farmhash/archive/816a4ae622e964763ca0862d9dbd19324a1eaf45.tar.gz" FLATBUFFERS_URL="https://github.com/google/flatbuffers/archive/master.zip" +FFT2D_URL="https://mirror.bazel.build/www.kurims.kyoto-u.ac.jp/~ooura/fft.tgz" # TODO(petewarden): Some new code in Eigen triggers a clang bug with iOS arm64, # so work around it by patching the source. @@ -91,6 +94,7 @@ download_and_extract "${ABSL_URL}" "${DOWNLOADS_DIR}/absl" download_and_extract "${NEON_2_SSE_URL}" "${DOWNLOADS_DIR}/neon_2_sse" download_and_extract "${FARMHASH_URL}" "${DOWNLOADS_DIR}/farmhash" download_and_extract "${FLATBUFFERS_URL}" "${DOWNLOADS_DIR}/flatbuffers" +download_and_extract "${FFT2D_URL}" "${DOWNLOADS_DIR}/fft2d" replace_by_sed 's#static uint32x4_t p4ui_CONJ_XOR = vld1q_u32( conj_XOR_DATA );#static uint32x4_t p4ui_CONJ_XOR; // = vld1q_u32( conj_XOR_DATA ); - Removed by script#' \ "${DOWNLOADS_DIR}/eigen/Eigen/src/Core/arch/NEON/Complex.h" diff --git a/tensorflow/contrib/lite/examples/ios/camera/tflite_camera_example.xcodeproj/project.pbxproj b/tensorflow/contrib/lite/examples/ios/camera/tflite_camera_example.xcodeproj/project.pbxproj index b0236e9c60..98d3b5bb8a 100644 --- a/tensorflow/contrib/lite/examples/ios/camera/tflite_camera_example.xcodeproj/project.pbxproj +++ b/tensorflow/contrib/lite/examples/ios/camera/tflite_camera_example.xcodeproj/project.pbxproj @@ -326,10 +326,6 @@ GCC_WARN_UNUSED_VARIABLE = YES; HEADER_SEARCH_PATHS = ( "$(inherited)", - ../../../../../../, - ../../../downloads/flatbuffers/include/, - ../../../downloads/eigen/, - ../../../downloads/, ); IPHONEOS_DEPLOYMENT_TARGET = 8.0; MTL_ENABLE_DEBUG_INFO = YES; @@ -373,10 +369,6 @@ GCC_WARN_UNUSED_VARIABLE = YES; HEADER_SEARCH_PATHS = ( "$(inherited)", - ../../../../../../, - ../../../downloads/flatbuffers/include/, - ../../../downloads/eigen/, - ../../../downloads/, ); IPHONEOS_DEPLOYMENT_TARGET = 8.0; MTL_ENABLE_DEBUG_INFO = NO; diff --git a/tensorflow/contrib/lite/g3doc/apis.md b/tensorflow/contrib/lite/g3doc/apis.md index fe208e47d1..50cc146a87 100644 --- a/tensorflow/contrib/lite/g3doc/apis.md +++ b/tensorflow/contrib/lite/g3doc/apis.md @@ -29,7 +29,7 @@ interpreter->AllocateTensors(); float* input = interpreter->typed_input_tensor<float>(0); // Fill `input`. interpreter->Invoke(); -float* output = interpreter->type_output_tensor<float>(0); +float* output = interpreter->typed_output_tensor<float>(0); ``` ### Data Alignment diff --git a/tensorflow/contrib/lite/java/demo/app/src/main/java/com/example/android/tflitecamerademo/Camera2BasicFragment.java b/tensorflow/contrib/lite/java/demo/app/src/main/java/com/example/android/tflitecamerademo/Camera2BasicFragment.java index 300786c3ca..18f6465188 100644 --- a/tensorflow/contrib/lite/java/demo/app/src/main/java/com/example/android/tflitecamerademo/Camera2BasicFragment.java +++ b/tensorflow/contrib/lite/java/demo/app/src/main/java/com/example/android/tflitecamerademo/Camera2BasicFragment.java @@ -54,6 +54,9 @@ import android.view.Surface; import android.view.TextureView; import android.view.View; import android.view.ViewGroup; +import android.widget.CompoundButton; +import android.widget.NumberPicker; +import android.widget.ToggleButton; import android.widget.TextView; import android.widget.Toast; import java.io.IOException; @@ -82,6 +85,8 @@ public class Camera2BasicFragment extends Fragment private boolean runClassifier = false; private boolean checkedPermissions = false; private TextView textView; + private ToggleButton toggle; + private NumberPicker np; private ImageClassifier classifier; /** Max preview width that is guaranteed by Camera2 API */ @@ -289,6 +294,24 @@ public class Camera2BasicFragment extends Fragment public void onViewCreated(final View view, Bundle savedInstanceState) { textureView = (AutoFitTextureView) view.findViewById(R.id.texture); textView = (TextView) view.findViewById(R.id.text); + toggle = (ToggleButton) view.findViewById(R.id.button); + + toggle.setOnCheckedChangeListener(new CompoundButton.OnCheckedChangeListener() { + public void onCheckedChanged(CompoundButton buttonView, boolean isChecked) { + classifier.setUseNNAPI(isChecked); + } + }); + + np = (NumberPicker) view.findViewById(R.id.np); + np.setMinValue(1); + np.setMaxValue(10); + np.setWrapSelectorWheel(true); + np.setOnValueChangedListener(new NumberPicker.OnValueChangeListener() { + @Override + public void onValueChange(NumberPicker picker, int oldVal, int newVal){ + classifier.setNumThreads(newVal); + } + }); } /** Load the model and labels. */ diff --git a/tensorflow/contrib/lite/java/demo/app/src/main/java/com/example/android/tflitecamerademo/ImageClassifier.java b/tensorflow/contrib/lite/java/demo/app/src/main/java/com/example/android/tflitecamerademo/ImageClassifier.java index c57bb348c5..d32c077910 100644 --- a/tensorflow/contrib/lite/java/demo/app/src/main/java/com/example/android/tflitecamerademo/ImageClassifier.java +++ b/tensorflow/contrib/lite/java/demo/app/src/main/java/com/example/android/tflitecamerademo/ImageClassifier.java @@ -142,6 +142,16 @@ public abstract class ImageClassifier { } } + public void setUseNNAPI(Boolean nnapi) { + if (tflite != null) + tflite.setUseNNAPI(nnapi); + } + + public void setNumThreads(int num_threads) { + if (tflite != null) + tflite.setNumThreads(num_threads); + } + /** Closes tflite to release resources. */ public void close() { tflite.close(); diff --git a/tensorflow/contrib/lite/java/demo/app/src/main/res/layout/fragment_camera2_basic.xml b/tensorflow/contrib/lite/java/demo/app/src/main/res/layout/fragment_camera2_basic.xml index 15305c436e..db557ad62f 100644 --- a/tensorflow/contrib/lite/java/demo/app/src/main/res/layout/fragment_camera2_basic.xml +++ b/tensorflow/contrib/lite/java/demo/app/src/main/res/layout/fragment_camera2_basic.xml @@ -22,24 +22,59 @@ android:layout_width="wrap_content" android:layout_height="wrap_content" android:layout_alignParentStart="true" + android:layout_alignParentLeft="true" android:layout_alignParentTop="true" /> <FrameLayout android:id="@+id/control" android:layout_width="match_parent" - android:layout_height="112dp" + android:layout_height="135dp" android:layout_alignParentBottom="true" android:layout_alignParentStart="true" + android:layout_alignParentLeft="true" + android:layout_alignParentEnd="true" + android:layout_alignParentRight="true" + android:layout_marginEnd="150dp" + android:layout_marginRight="150dp" android:background="@color/control_background"> - <TextView android:id="@+id/text" + <TextView + android:id="@+id/text" android:layout_width="wrap_content" android:layout_height="wrap_content" - android:paddingLeft="80dp" + android:paddingLeft="20dp" android:textColor="#FFF" android:textSize="20sp" android:textStyle="bold" /> </FrameLayout> + <RelativeLayout + android:id="@+id/control2" + android:layout_width="match_parent" + android:layout_height="135dp" + android:layout_alignParentLeft="true" + android:layout_alignParentStart="true" + android:layout_alignTop="@+id/control" + android:layout_marginLeft="300dp" + android:layout_marginStart="300dp" + android:background="@color/control_background"> + + <ToggleButton + android:id="@+id/button" + android:textOff="@string/tflite" + android:textOn="@string/nnapi" + android:layout_width="wrap_content" + android:layout_height="wrap_content" + android:layout_alignParentLeft="true" + android:layout_alignParentStart="true" /> + + <NumberPicker + android:id="@+id/np" + android:layout_width="wrap_content" + android:layout_height="wrap_content" + android:layout_below="@+id/button" + android:visibility="visible" /> + </RelativeLayout> + </RelativeLayout> diff --git a/tensorflow/contrib/lite/java/demo/app/src/main/res/values/strings.xml b/tensorflow/contrib/lite/java/demo/app/src/main/res/values/strings.xml index a08ec3eb62..29a033bcd4 100644 --- a/tensorflow/contrib/lite/java/demo/app/src/main/res/values/strings.xml +++ b/tensorflow/contrib/lite/java/demo/app/src/main/res/values/strings.xml @@ -21,4 +21,6 @@ <string name="toggle_turn_on">NN:On</string> <string name="toggle_turn_off">NN:Off</string> <string name="toggle">Use NNAPI</string> + <string name="tflite">tflite</string> + <string name="nnapi">NNAPI</string> </resources> diff --git a/tensorflow/contrib/lite/java/src/main/java/org/tensorflow/lite/Interpreter.java b/tensorflow/contrib/lite/java/src/main/java/org/tensorflow/lite/Interpreter.java index e915e65aa1..e84ee71129 100644 --- a/tensorflow/contrib/lite/java/src/main/java/org/tensorflow/lite/Interpreter.java +++ b/tensorflow/contrib/lite/java/src/main/java/org/tensorflow/lite/Interpreter.java @@ -215,6 +215,13 @@ public final class Interpreter implements AutoCloseable { } } + public void setNumThreads(int num_threads) { + if (wrapper == null) { + throw new IllegalStateException("The interpreter has already been closed."); + } + wrapper.setNumThreads(num_threads); + } + /** Release resources associated with the {@code Interpreter}. */ @Override public void close() { diff --git a/tensorflow/contrib/lite/java/src/main/java/org/tensorflow/lite/NativeInterpreterWrapper.java b/tensorflow/contrib/lite/java/src/main/java/org/tensorflow/lite/NativeInterpreterWrapper.java index dfc8ac111a..2fc803715b 100644 --- a/tensorflow/contrib/lite/java/src/main/java/org/tensorflow/lite/NativeInterpreterWrapper.java +++ b/tensorflow/contrib/lite/java/src/main/java/org/tensorflow/lite/NativeInterpreterWrapper.java @@ -153,6 +153,10 @@ final class NativeInterpreterWrapper implements AutoCloseable { useNNAPI(interpreterHandle, useNNAPI); } + void setNumThreads(int num_threads) { + numThreads(interpreterHandle, num_threads); + } + /** Gets index of an input given its name. */ int getInputIndex(String name) { if (inputsIndexes == null) { @@ -324,6 +328,8 @@ final class NativeInterpreterWrapper implements AutoCloseable { private static native void useNNAPI(long interpreterHandle, boolean state); + private static native void numThreads(long interpreterHandle, int num_threads); + private static native long createErrorReporter(int size); private static native long createModel(String modelPathOrBuffer, long errorHandle); diff --git a/tensorflow/contrib/lite/java/src/main/native/nativeinterpreterwrapper_jni.cc b/tensorflow/contrib/lite/java/src/main/native/nativeinterpreterwrapper_jni.cc index ccfdfd829b..45f510da1d 100644 --- a/tensorflow/contrib/lite/java/src/main/native/nativeinterpreterwrapper_jni.cc +++ b/tensorflow/contrib/lite/java/src/main/native/nativeinterpreterwrapper_jni.cc @@ -320,6 +320,16 @@ Java_org_tensorflow_lite_NativeInterpreterWrapper_useNNAPI(JNIEnv* env, interpreter->UseNNAPI(static_cast<bool>(state)); } +JNIEXPORT void JNICALL +Java_org_tensorflow_lite_NativeInterpreterWrapper_numThreads(JNIEnv* env, + jclass clazz, + jlong handle, + jint num_threads) { + tflite::Interpreter* interpreter = convertLongToInterpreter(env, handle); + if (interpreter == nullptr) return; + interpreter->SetNumThreads(static_cast<int>(num_threads)); +} + JNIEXPORT jlong JNICALL Java_org_tensorflow_lite_NativeInterpreterWrapper_createErrorReporter( JNIEnv* env, jclass clazz, jint size) { diff --git a/tensorflow/contrib/lite/java/src/main/native/nativeinterpreterwrapper_jni.h b/tensorflow/contrib/lite/java/src/main/native/nativeinterpreterwrapper_jni.h index 0e28a77fee..eaa765cb34 100644 --- a/tensorflow/contrib/lite/java/src/main/native/nativeinterpreterwrapper_jni.h +++ b/tensorflow/contrib/lite/java/src/main/native/nativeinterpreterwrapper_jni.h @@ -61,7 +61,7 @@ Java_org_tensorflow_lite_NativeInterpreterWrapper_getOutputNames(JNIEnv* env, /* * Class: org_tensorflow_lite_NativeInterpreterWrapper * Method: - * Signature: (JZ) + * Signature: (JZ)V */ JNIEXPORT void JNICALL Java_org_tensorflow_lite_NativeInterpreterWrapper_useNNAPI(JNIEnv* env, @@ -72,6 +72,16 @@ Java_org_tensorflow_lite_NativeInterpreterWrapper_useNNAPI(JNIEnv* env, /* * Class: org_tensorflow_lite_NativeInterpreterWrapper * Method: + * Signature: (JI)V + */ +JNIEXPORT void JNICALL +Java_org_tensorflow_lite_NativeInterpreterWrapper_numThreads(JNIEnv* env, + jclass clazz, + jlong handle, + jint num_threads); +/* + * Class: org_tensorflow_lite_NativeInterpreterWrapper + * Method: * Signature: (I)J */ JNIEXPORT jlong JNICALL diff --git a/tensorflow/contrib/lite/kernels/add.cc b/tensorflow/contrib/lite/kernels/add.cc index 63ea89df56..e0aa070e2d 100644 --- a/tensorflow/contrib/lite/kernels/add.cc +++ b/tensorflow/contrib/lite/kernels/add.cc @@ -176,7 +176,7 @@ TfLiteStatus Eval(TfLiteContext* context, TfLiteNode* node) { output); } else { context->ReportError(context, - "Inputs and outputs not all float|unit8 types."); + "Inputs and outputs not all float|uint8 types."); return kTfLiteError; } diff --git a/tensorflow/contrib/lite/kernels/div.cc b/tensorflow/contrib/lite/kernels/div.cc index 6dd243ad62..ec380c8e49 100644 --- a/tensorflow/contrib/lite/kernels/div.cc +++ b/tensorflow/contrib/lite/kernels/div.cc @@ -106,6 +106,8 @@ void EvalFloat(TfLiteContext* context, TfLiteNode* node, #undef TF_LITE_DIV } + + template <KernelType kernel_type> TfLiteStatus Eval(TfLiteContext* context, TfLiteNode* node) { auto* params = reinterpret_cast<TfLiteDivParams*>(node->builtin_data); @@ -118,7 +120,8 @@ TfLiteStatus Eval(TfLiteContext* context, TfLiteNode* node) { if (output->type == kTfLiteFloat32) { EvalFloat<kernel_type>(context, node, params, data, input1, input2, output); } else { - context->ReportError(context, "Inputs and outputs not all float types."); + context->ReportError(context, + "Div only supports FLOAT32 and quantized UINT8 now."); return kTfLiteError; } diff --git a/tensorflow/contrib/lite/kernels/internal/optimized/optimized_ops.h b/tensorflow/contrib/lite/kernels/internal/optimized/optimized_ops.h index d585bcca0e..9e9aba0169 100644 --- a/tensorflow/contrib/lite/kernels/internal/optimized/optimized_ops.h +++ b/tensorflow/contrib/lite/kernels/internal/optimized/optimized_ops.h @@ -4374,7 +4374,7 @@ inline void Softmax(const uint8* input_data, const Dims<4>& input_dims, using FixedPointAccum = gemmlowp::FixedPoint<int32, kAccumulationIntegerBits>; using FixedPoint0 = gemmlowp::FixedPoint<int32, 0>; - gemmlowp::ScopedProfilingLabel label("Softmax/8bit"); +gemmlowp::ScopedProfilingLabel label("Softmax/8bit"); const int batches = MatchingArraySize(input_dims, 3, output_dims, 3); const int height = MatchingArraySize(input_dims, 2, output_dims, 2); const int width = MatchingArraySize(input_dims, 1, output_dims, 1); diff --git a/tensorflow/contrib/lite/kernels/internal/reference/reference_ops.h b/tensorflow/contrib/lite/kernels/internal/reference/reference_ops.h index ae295cc8b5..4c8cbe4275 100644 --- a/tensorflow/contrib/lite/kernels/internal/reference/reference_ops.h +++ b/tensorflow/contrib/lite/kernels/internal/reference/reference_ops.h @@ -1403,6 +1403,33 @@ inline void BroadcastMul(const uint8* input1_data, const Dims<4>& input1_dims, output_data, output_dims); } +inline void Div(const float* input1_data, const Dims<4>& input1_dims, + const float* input2_data, const Dims<4>& input2_dims, + float output_activation_min, float output_activation_max, + float* output_data, const Dims<4>& output_dims) { + const int batches = + MatchingArraySize(input1_dims, 3, input2_dims, 3, output_dims, 3); + const int height = + MatchingArraySize(input1_dims, 2, input2_dims, 2, output_dims, 2); + const int width = + MatchingArraySize(input1_dims, 1, input2_dims, 1, output_dims, 1); + const int depth = + MatchingArraySize(input1_dims, 0, input2_dims, 0, output_dims, 0); + for (int b = 0; b < batches; ++b) { + for (int y = 0; y < height; ++y) { + for (int x = 0; x < width; ++x) { + for (int c = 0; c < depth; ++c) { + output_data[Offset(output_dims, c, x, y, b)] = + ActivationFunctionWithMinMax( + input1_data[Offset(input1_dims, c, x, y, b)] / + input2_data[Offset(input2_dims, c, x, y, b)], + output_activation_min, output_activation_max); + } + } + } + } +} + // TODO(jiawen): We can implement BroadcastDiv on buffers of arbitrary // dimensionality if the runtime code does a single loop over one dimension // that handles broadcasting as the base case. The code generator would then @@ -1444,18 +1471,6 @@ void BroadcastDiv(const T* input1_data, const Dims<4>& input1_dims, } } -inline void Div(const float* input1_data, const Dims<4>& input1_dims, - const float* input2_data, const Dims<4>& input2_dims, - float output_activation_min, float output_activation_max, - float* output_data, const Dims<4>& output_dims) { - const int flat_size = MatchingFlatSize(input1_dims, input2_dims, output_dims); - for (int i = 0; i < flat_size; ++i) { - output_data[i] = ActivationFunctionWithMinMax( - input1_data[i] / input2_data[i], output_activation_min, - output_activation_max); - } -} - inline void Sub(const float* input1_data, const Dims<4>& input1_dims, const float* input2_data, const Dims<4>& input2_dims, float output_activation_min, float output_activation_max, diff --git a/tensorflow/contrib/lite/kernels/sub.cc b/tensorflow/contrib/lite/kernels/sub.cc index 66b06aeaec..7c60a4fdbf 100644 --- a/tensorflow/contrib/lite/kernels/sub.cc +++ b/tensorflow/contrib/lite/kernels/sub.cc @@ -174,7 +174,8 @@ TfLiteStatus Eval(TfLiteContext* context, TfLiteNode* node) { EvalQuantized<kernel_type>(context, node, params, data, input1, input2, output); } else { - context->ReportError(context, "Inputs and outputs not all float types."); + context->ReportError(context, + "Inputs and outputs not all float|uint8 types."); return kTfLiteError; } diff --git a/tensorflow/contrib/lite/toco/graph_transformations/resolve_tensorflow_merge.cc b/tensorflow/contrib/lite/toco/graph_transformations/resolve_tensorflow_merge.cc index 477e7f13da..38e0005890 100644 --- a/tensorflow/contrib/lite/toco/graph_transformations/resolve_tensorflow_merge.cc +++ b/tensorflow/contrib/lite/toco/graph_transformations/resolve_tensorflow_merge.cc @@ -32,7 +32,7 @@ bool ResolveTensorFlowMerge::Run(Model* model, std::size_t op_index) { } // We need to yield until this Merge node has only 1 input, which will mean - // that that is the selected input. Other graph transformations on other nodes + // that is the selected input. Other graph transformations on other nodes // such as ResolveTensorFlowSwitch, will take care of trimming the // non-selected inputs, so that at some point there will be only 1 input left. if (merge_op->inputs.size() > 1) { diff --git a/tensorflow/contrib/lite/toco/model.h b/tensorflow/contrib/lite/toco/model.h index 705a9d69a6..482cc71d8b 100644 --- a/tensorflow/contrib/lite/toco/model.h +++ b/tensorflow/contrib/lite/toco/model.h @@ -152,9 +152,9 @@ enum class AxesOrder { }; // The type of the scalars in an array. -// Note that that does not by itself tell whether the values in the array are -// real (are literally interpreted as real numbers) or quantized (only acquire -// a meaning as real numbers in conjunction with QuantizationParams). +// Note that the type does not by itself tell whether the values in the array +// are real (are literally interpreted as real numbers) or quantized (only +// acquire a meaning as real numbers in conjunction with QuantizationParams). // // In practice though: // float values are always real |