Reduce memory usage and increase performance for convolution on iOS (#…

…3778) * Reduce memory usage and increase performance for convolution on iOS * Switched to persistent memory allocation, added documentation, and make all the implementations functors * Updated conv kernel based on review comments
tensorflow · Aug 23, 2016 · 459c2fe · 459c2fe
1 parent 6d04d60
commit 459c2fe
Show file tree

Hide file tree

Showing 8 changed files with 653 additions and 0 deletions.
diff --git a/tensorflow/contrib/ios_examples/README.md b/tensorflow/contrib/ios_examples/README.md
@@ -71,6 +71,9 @@ rundown:
    inside the library are not stripped out. To the linker, they can appear
    unused because no other code references the variables, but in fact their
    constructors have the important side effect of registering the class.
+
+ - You'll need to include the Accelerate framework in the "Link Binary with
+   Libraries" build phase of your project.
 
  - C++11 support (or later) should be enabled by setting `C++ Language Dialect` to
    `GNU++11` (or `GNU++14`), and `C++ Standard Library` to `libc++`.

diff --git a/tensorflow/contrib/ios_examples/benchmark/benchmark.xcodeproj/project.pbxproj b/tensorflow/contrib/ios_examples/benchmark/benchmark.xcodeproj/project.pbxproj
@@ -9,6 +9,7 @@
 /* Begin PBXBuildFile section */
 		590E7D881D02091F00DF5523 /* libprotobuf-lite.a in Frameworks */ = {isa = PBXBuildFile; fileRef = 590E7D861D02091F00DF5523 /* libprotobuf-lite.a */; };
 		590E7D8A1D0209DD00DF5523 /* libprotobuf.a in Frameworks */ = {isa = PBXBuildFile; fileRef = 590E7D871D02091F00DF5523 /* libprotobuf.a */; };
+		5993C7701D5D4E7F0048CE6A /* Accelerate.framework in Frameworks */ = {isa = PBXBuildFile; fileRef = 5993C76F1D5D4E7F0048CE6A /* Accelerate.framework */; };
 		59A3D0011CF4E68100C4259F /* AppDelegate.mm in Sources */ = {isa = PBXBuildFile; fileRef = 59A3CFF21CF4E68100C4259F /* AppDelegate.mm */; };
 		59A3D0031CF4E68100C4259F /* grace_hopper.jpg in Resources */ = {isa = PBXBuildFile; fileRef = 59A3CFF51CF4E68100C4259F /* grace_hopper.jpg */; };
 		59A3D0051CF4E68100C4259F /* imagenet_comp_graph_label_strings.txt in Resources */ = {isa = PBXBuildFile; fileRef = 59A3CFF71CF4E68100C4259F /* imagenet_comp_graph_label_strings.txt */; };
@@ -25,6 +26,7 @@
 		590E7D861D02091F00DF5523 /* libprotobuf-lite.a */ = {isa = PBXFileReference; lastKnownFileType = archive.ar; name = "libprotobuf-lite.a"; path = "../../makefile/gen/protobuf_ios/lib/libprotobuf-lite.a"; sourceTree = "<group>"; };
 		590E7D871D02091F00DF5523 /* libprotobuf.a */ = {isa = PBXFileReference; lastKnownFileType = archive.ar; name = libprotobuf.a; path = ../../makefile/gen/protobuf_ios/lib/libprotobuf.a; sourceTree = "<group>"; };
 		5911579B1CF4011C00C31E3A /* benchmark.app */ = {isa = PBXFileReference; explicitFileType = wrapper.application; includeInIndex = 0; path = benchmark.app; sourceTree = BUILT_PRODUCTS_DIR; };
+		5993C76F1D5D4E7F0048CE6A /* Accelerate.framework */ = {isa = PBXFileReference; lastKnownFileType = wrapper.framework; name = Accelerate.framework; path = System/Library/Frameworks/Accelerate.framework; sourceTree = SDKROOT; };
 		59A3CFF11CF4E68100C4259F /* AppDelegate.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = AppDelegate.h; sourceTree = "<group>"; };
 		59A3CFF21CF4E68100C4259F /* AppDelegate.mm */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.objcpp; path = AppDelegate.mm; sourceTree = "<group>"; };
 		59A3CFF41CF4E68100C4259F /* cropped_panda.jpg */ = {isa = PBXFileReference; lastKnownFileType = image.jpeg; path = cropped_panda.jpg; sourceTree = "<group>"; };
@@ -50,6 +52,7 @@
 			isa = PBXFrameworksBuildPhase;
 			buildActionMask = 2147483647;
 			files = (
+				5993C7701D5D4E7F0048CE6A /* Accelerate.framework in Frameworks */,
 				590E7D8A1D0209DD00DF5523 /* libprotobuf.a in Frameworks */,
 				590E7D881D02091F00DF5523 /* libprotobuf-lite.a in Frameworks */,
 				59A3D0181CF4E86100C4259F /* UIKit.framework in Frameworks */,
@@ -63,6 +66,7 @@
 		591157921CF4011C00C31E3A = {
 			isa = PBXGroup;
 			children = (
+				5993C76F1D5D4E7F0048CE6A /* Accelerate.framework */,
 				590E7D861D02091F00DF5523 /* libprotobuf-lite.a */,
 				590E7D871D02091F00DF5523 /* libprotobuf.a */,
 				59A3D0171CF4E86100C4259F /* UIKit.framework */,

diff --git a/tensorflow/contrib/ios_examples/camera/camera_example.xcodeproj/project.pbxproj b/tensorflow/contrib/ios_examples/camera/camera_example.xcodeproj/project.pbxproj
@@ -24,6 +24,7 @@
 		592FF90D18EDD0DA00C164F8 /* MainStoryboard_iPhone.storyboard in Resources */ = {isa = PBXBuildFile; fileRef = 592FF90A18EDD0DA00C164F8 /* MainStoryboard_iPhone.storyboard */; };
 		592FF92518EE240200C164F8 /* CameraExampleAppDelegate.m in Sources */ = {isa = PBXBuildFile; fileRef = 592FF92218EE240200C164F8 /* CameraExampleAppDelegate.m */; };
 		592FF92618EE240200C164F8 /* CameraExampleViewController.mm in Sources */ = {isa = PBXBuildFile; fileRef = 592FF92418EE240200C164F8 /* CameraExampleViewController.mm */; };
+		5993C7721D5D4E980048CE6A /* Accelerate.framework in Frameworks */ = {isa = PBXBuildFile; fileRef = 5993C7711D5D4E980048CE6A /* Accelerate.framework */; };
 /* End PBXBuildFile section */
 
 /* Begin PBXFileReference section */
@@ -52,13 +53,15 @@
 		592FF92218EE240200C164F8 /* CameraExampleAppDelegate.m */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.objc; path = CameraExampleAppDelegate.m; sourceTree = SOURCE_ROOT; };
 		592FF92318EE240200C164F8 /* CameraExampleViewController.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = CameraExampleViewController.h; sourceTree = SOURCE_ROOT; };
 		592FF92418EE240200C164F8 /* CameraExampleViewController.mm */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.objcpp; path = CameraExampleViewController.mm; sourceTree = SOURCE_ROOT; };
+		5993C7711D5D4E980048CE6A /* Accelerate.framework */ = {isa = PBXFileReference; lastKnownFileType = wrapper.framework; name = Accelerate.framework; path = Platforms/iPhoneOS.platform/Developer/SDKs/iPhoneOS9.3.sdk/System/Library/Frameworks/Accelerate.framework; sourceTree = DEVELOPER_DIR; };
 /* End PBXFileReference section */
 
 /* Begin PBXFrameworksBuildPhase section */
 		592FF8B218ECBD7600C164F8 /* Frameworks */ = {
 			isa = PBXFrameworksBuildPhase;
 			buildActionMask = 2147483647;
 			files = (
+				5993C7721D5D4E980048CE6A /* Accelerate.framework in Frameworks */,
 				591D3EDF1CFFAD230059011C /* libprotobuf-lite.a in Frameworks */,
 				591D3EE01CFFAD230059011C /* libprotobuf.a in Frameworks */,
 				591D3ECF1CFF7FCE0059011C /* ImageIO.framework in Frameworks */,
@@ -103,6 +106,7 @@
 		592FF8B718ECBD7600C164F8 /* Frameworks */ = {
 			isa = PBXGroup;
 			children = (
+				5993C7711D5D4E980048CE6A /* Accelerate.framework */,
 				591D3EDD1CFFAD230059011C /* libprotobuf-lite.a */,
 				591D3EDE1CFFAD230059011C /* libprotobuf.a */,
 				591D3ECE1CFF7FCE0059011C /* ImageIO.framework */,

diff --git a/tensorflow/contrib/ios_examples/simple/tf_ios_makefile_example.xcodeproj/project.pbxproj b/tensorflow/contrib/ios_examples/simple/tf_ios_makefile_example.xcodeproj/project.pbxproj
@@ -9,6 +9,7 @@
 /* Begin PBXBuildFile section */
 		590E7D881D02091F00DF5523 /* libprotobuf-lite.a in Frameworks */ = {isa = PBXBuildFile; fileRef = 590E7D861D02091F00DF5523 /* libprotobuf-lite.a */; };
 		590E7D8A1D0209DD00DF5523 /* libprotobuf.a in Frameworks */ = {isa = PBXBuildFile; fileRef = 590E7D871D02091F00DF5523 /* libprotobuf.a */; };
+		5993C7741D5D4EAF0048CE6A /* Accelerate.framework in Frameworks */ = {isa = PBXBuildFile; fileRef = 5993C7731D5D4EAF0048CE6A /* Accelerate.framework */; };
 		59A3D0011CF4E68100C4259F /* AppDelegate.mm in Sources */ = {isa = PBXBuildFile; fileRef = 59A3CFF21CF4E68100C4259F /* AppDelegate.mm */; };
 		59A3D0031CF4E68100C4259F /* grace_hopper.jpg in Resources */ = {isa = PBXBuildFile; fileRef = 59A3CFF51CF4E68100C4259F /* grace_hopper.jpg */; };
 		59A3D0051CF4E68100C4259F /* imagenet_comp_graph_label_strings.txt in Resources */ = {isa = PBXBuildFile; fileRef = 59A3CFF71CF4E68100C4259F /* imagenet_comp_graph_label_strings.txt */; };
@@ -25,6 +26,7 @@
 		590E7D861D02091F00DF5523 /* libprotobuf-lite.a */ = {isa = PBXFileReference; lastKnownFileType = archive.ar; name = "libprotobuf-lite.a"; path = "../../makefile/gen/protobuf_ios/lib/libprotobuf-lite.a"; sourceTree = "<group>"; };
 		590E7D871D02091F00DF5523 /* libprotobuf.a */ = {isa = PBXFileReference; lastKnownFileType = archive.ar; name = libprotobuf.a; path = ../../makefile/gen/protobuf_ios/lib/libprotobuf.a; sourceTree = "<group>"; };
 		5911579B1CF4011C00C31E3A /* tf_ios_makefile_example.app */ = {isa = PBXFileReference; explicitFileType = wrapper.application; includeInIndex = 0; path = tf_ios_makefile_example.app; sourceTree = BUILT_PRODUCTS_DIR; };
+		5993C7731D5D4EAF0048CE6A /* Accelerate.framework */ = {isa = PBXFileReference; lastKnownFileType = wrapper.framework; name = Accelerate.framework; path = System/Library/Frameworks/Accelerate.framework; sourceTree = SDKROOT; };
 		59A3CFF11CF4E68100C4259F /* AppDelegate.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = AppDelegate.h; sourceTree = "<group>"; };
 		59A3CFF21CF4E68100C4259F /* AppDelegate.mm */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.objcpp; path = AppDelegate.mm; sourceTree = "<group>"; };
 		59A3CFF41CF4E68100C4259F /* cropped_panda.jpg */ = {isa = PBXFileReference; lastKnownFileType = image.jpeg; path = cropped_panda.jpg; sourceTree = "<group>"; };
@@ -50,6 +52,7 @@
 			isa = PBXFrameworksBuildPhase;
 			buildActionMask = 2147483647;
 			files = (
+				5993C7741D5D4EAF0048CE6A /* Accelerate.framework in Frameworks */,
 				590E7D8A1D0209DD00DF5523 /* libprotobuf.a in Frameworks */,
 				590E7D881D02091F00DF5523 /* libprotobuf-lite.a in Frameworks */,
 				59A3D0181CF4E86100C4259F /* UIKit.framework in Frameworks */,
@@ -63,6 +66,7 @@
 		591157921CF4011C00C31E3A = {
 			isa = PBXGroup;
 			children = (
+				5993C7731D5D4EAF0048CE6A /* Accelerate.framework */,
 				590E7D861D02091F00DF5523 /* libprotobuf-lite.a */,
 				590E7D871D02091F00DF5523 /* libprotobuf.a */,
 				59A3D0171CF4E86100C4259F /* UIKit.framework */,

diff --git a/tensorflow/contrib/makefile/Makefile b/tensorflow/contrib/makefile/Makefile
@@ -284,6 +284,7 @@ ifeq ($(TARGET),IOS)
 		CXXFLAGS += -miphoneos-version-min=$(MIN_SDK_VERSION) \
 		-arch armv7 \
 		-D__thread= \
+		-DUSE_GEMM_FOR_CONV \
 		-Wno-c++11-narrowing \
 		-mno-thumb \
 		-DTF_LEAN_BINARY \
@@ -294,6 +295,7 @@ ifeq ($(TARGET),IOS)
 		${IPHONEOS_SYSROOT}
 		LDFLAGS := -arch armv7 \
 		-miphoneos-version-min=${MIN_SDK_VERSION} \
+		-framework Accelerate \
 		-Xlinker -S \
 		-Xlinker -x \
 		-Xlinker -dead_strip \
@@ -305,6 +307,7 @@ ifeq ($(TARGET),IOS)
 		CXXFLAGS += -miphoneos-version-min=$(MIN_SDK_VERSION) \
 		-arch armv7s \
 		-D__thread= \
+		-DUSE_GEMM_FOR_CONV \
 		-Wno-c++11-narrowing \
 		-mno-thumb \
 		-DTF_LEAN_BINARY \
@@ -315,6 +318,7 @@ ifeq ($(TARGET),IOS)
 		${IPHONEOS_SYSROOT}
 		LDFLAGS := -arch armv7s \
 		-miphoneos-version-min=${MIN_SDK_VERSION} \
+		-framework Accelerate \
 		-Xlinker -S \
 		-Xlinker -x \
 		-Xlinker -dead_strip \
@@ -326,6 +330,7 @@ ifeq ($(TARGET),IOS)
 		CXXFLAGS += -miphoneos-version-min=$(MIN_SDK_VERSION) \
 		-arch arm64 \
 		-D__thread= \
+		-DUSE_GEMM_FOR_CONV \
 		-Wno-c++11-narrowing \
 		-DTF_LEAN_BINARY \
 		-D__ANDROID_TYPES_SLIM__ \
@@ -335,6 +340,7 @@ ifeq ($(TARGET),IOS)
 		${IPHONEOS_SYSROOT}
 		LDFLAGS := -arch arm64 \
 		-miphoneos-version-min=${MIN_SDK_VERSION} \
+		-framework Accelerate \
 		-Xlinker -S \
 		-Xlinker -x \
 		-Xlinker -dead_strip \
@@ -346,6 +352,7 @@ ifeq ($(TARGET),IOS)
 		CXXFLAGS += -mios-simulator-version-min=$(MIN_SDK_VERSION) \
 		-arch i386 \
 		-D__thread= \
+		-DUSE_GEMM_FOR_CONV \
 		-Wno-c++11-narrowing \
 		-DTF_LEAN_BINARY \
 		-D__ANDROID_TYPES_SLIM__ \
@@ -355,6 +362,7 @@ ifeq ($(TARGET),IOS)
 		${IPHONESIMULATOR_SYSROOT}
 		LDFLAGS := -arch i386 \
 		-mios-simulator-version-min=${MIN_SDK_VERSION} \
+		-framework Accelerate \
 		-Xlinker -S \
 		-Xlinker -x \
 		-Xlinker -dead_strip \
@@ -366,6 +374,7 @@ ifeq ($(TARGET),IOS)
 		CXXFLAGS += -mios-simulator-version-min=$(MIN_SDK_VERSION) \
 		-arch x86_64 \
 		-D__thread= \
+		-DUSE_GEMM_FOR_CONV \
 		-Wno-c++11-narrowing \
 		-DTF_LEAN_BINARY \
 		-D__ANDROID_TYPES_SLIM__ \
@@ -375,6 +384,7 @@ ifeq ($(TARGET),IOS)
 		${IPHONESIMULATOR_SYSROOT}
 		LDFLAGS := -arch x86_64 \
 		-mios-simulator-version-min=${MIN_SDK_VERSION} \
+		-framework Accelerate \
 		-Xlinker -S \
 		-Xlinker -x \
 		-Xlinker -dead_strip \

diff --git a/tensorflow/contrib/makefile/tf_op_files.txt b/tensorflow/contrib/makefile/tf_op_files.txt
@@ -98,6 +98,7 @@ tensorflow/core/kernels/cwise_op_equal_to.cc
 tensorflow/core/kernels/cwise_op_div.cc
 tensorflow/core/kernels/cwise_op_add.cc
 tensorflow/core/kernels/ctc_decoder_ops.cc
+tensorflow/core/kernels/conv_ops_using_gemm.cc
 tensorflow/core/kernels/conv_ops.cc
 tensorflow/core/kernels/conv_grad_ops.cc
 tensorflow/core/kernels/control_flow_ops.cc

diff --git a/tensorflow/core/kernels/conv_ops.cc b/tensorflow/core/kernels/conv_ops.cc
@@ -242,8 +242,12 @@ class Conv2DOp : public BinaryOp<T> {
       Name("Conv2D").Device(DEVICE_CPU).TypeConstraint<T>("T"), \
       Conv2DOp<CPUDevice, T>);
 
+// If we're using the alternative GEMM-based implementation of Conv2D for the
+// CPU implementation, don't register this EigenTensor-based version.
+#if !defined(USE_GEMM_FOR_CONV)
 TF_CALL_half(REGISTER_CPU);
 TF_CALL_float(REGISTER_CPU);
+#endif  // USE_GEMM_FOR_CONV
 
 // To be used inside depthwise_conv_op.cc.
 template class LaunchConv2DOp<CPUDevice, float>;