Add SGD Kernel Go Package (sql-machine-learning#1637)

* add sgd kernel using eigen

* test style

* move kernel to go dir

* add kernel test

* refine test

* fix eigen3 header file

* trigger go unittest

* fix ci

* fix ci

* fix ci

* follow comments

* trigger go test

* update

* follow comments

* update

* update

* fix code style

* fix dockerfile

.clang-format

@@ -0,0 +1,18 @@
+---
+Language:        Cpp
+BasedOnStyle:  Google
+IndentWidth:     2
+TabWidth:        2
+ContinuationIndentWidth: 4
+AccessModifierOffset: -1  # The private/protected/public has no indent in class
+Standard:  Cpp11
+AllowAllParametersOfDeclarationOnNextLine: true
+BinPackParameters: false
+BinPackArguments: false
+---
+Language: Proto
+BasedOnStyle:  Google
+IndentWidth:     2
+TabWidth:        2
+ContinuationIndentWidth: 4
+---

.gitignore

@@ -14,6 +14,8 @@ __pycache__/
 
 # C extensions
 *.so
+*.o
+*.a
 
 # Distribution / packaging
 .Python

.pre-commit-config.yaml

@@ -21,11 +21,26 @@ repos:
     hooks:
     - id: mypy
       args: [--ignore-missing-imports, --follow-imports=skip]
+-   repo: local
+    hooks:
+    - id: clang-format
+      name: clang-format
+      description: Format files with ClangFormat.
+      entry: bash ./scripts/codestyle/clang_format.hook -i
+      language: system
+      files: \.(c|cc|cxx|cpp|cu|h|hpp|hxx|proto)$
+-   repo: local
+    hooks:
+    - id: cpplint-cpp-source
+      name: cpplint
+      description: Check C++ code style using cpplint.py.
+      entry: bash ./scripts/codestyle/cpplint_precommit.hook
+      language: system
+      files: \.(c|cc|cxx|cpp|cu|h|hpp|hxx)$
 - repo: git://github.com/dnephin/pre-commit-golang
   rev: v0.3.3
   hooks:
     - id: go-fmt
     - id: go-lint
     - id: validate-toml
-    - id: no-go-testing
-
+    - id: no-go-testing

.travis.yml

@@ -8,6 +8,7 @@ addons:
       - docker-ce
       - python3-pip
       - python3-setuptools
+      - clang-format
 
 install:
   - docker version

elasticdl/Makefile

@@ -5,11 +5,20 @@ EDL_PROTO_FILE = elasticdl/proto/elasticdl.proto
 GO_FILE_DIR = elasticdl/pkg/proto
 GO_PB_FILE = $(GO_FILE_DIR)/elasticdl.pb.go
 
-all: python_pb $(GO_PB_FILE)
+CXX = g++
+CAPI_DIR = elasticdl/pkg/kernel/capi
+
+all: python_pb $(GO_PB_FILE) kernel
 
 python_pb:
 	$(PYTHON) -m grpc_tools.protoc -I . $(EDL_PROTO_FILE) --python_out=. --grpc_python_out=.
 
 $(GO_PB_FILE): $(EDL_PROTO_FILE)
 	mkdir -p $(GO_FILE_DIR)
 	protoc -I . $< --go_out=plugins=grpc:.
+
+kernel:
+	${CXX} -O3 -std=c++11 ${CAPI_DIR}/kernel_api.cc -c -o ${CAPI_DIR}/libkernel_api.o
+	ar crv ${CAPI_DIR}/libkernel_api.a ${CAPI_DIR}/libkernel_api.o
+clean:
+	rm -rf ${CAPI_DIR}/*.o ${CAPI_DIR}/*.a

elasticdl/docker/Dockerfile

@@ -6,7 +6,7 @@ FROM ${BASE_IMAGE}
 COPY elasticdl/docker/bashrc /etc/bash.bashrc
 RUN chmod a+rwx /etc/bash.bashrc
 
-RUN apt-get update && apt-get install -y unzip curl git software-properties-common
+RUN apt-get update && apt-get install -y unzip curl git software-properties-common libeigen3-dev
 
 COPY elasticdl/requirements.txt /requirements.txt
 ARG EXTRA_PYPI_INDEX=https://pypi.org/simple
@@ -15,4 +15,4 @@ RUN pip install -r /requirements.txt --extra-index-url=${EXTRA_PYPI_INDEX}
 WORKDIR /
 ENV PYTHONPATH=/
 COPY elasticdl /elasticdl
-RUN make -f elasticdl/Makefile
+RUN make -f elasticdl/Makefile python_pb

elasticdl/docker/Dockerfile.dev

@@ -6,7 +6,9 @@ FROM ${BASE_IMAGE}
 COPY elasticdl/docker/bashrc /etc/bash.bashrc
 RUN chmod a+rwx /etc/bash.bashrc
 
-RUN apt-get update && apt-get install -y unzip curl git software-properties-common wget
+RUN apt update && \
+    apt install -y unzip curl git software-properties-common g++ wget \
+    libeigen3-dev
 
 COPY elasticdl/requirements.txt /requirements.txt
 COPY elasticdl/requirements-dev.txt /requirements-dev.txt

elasticdl/pkg/kernel/capi/kernel_api.cc

@@ -0,0 +1,13 @@
+#include "kernel_api.h"
+
+#include <eigen3/Eigen/Dense>
+
+void SGD(float* grad, float* param, double lr, long long size) {
+  Eigen::Map<Eigen::Array<float, 1, Eigen::Dynamic>> eg{
+      grad, static_cast<Eigen::Index>(size)};
+
+  Eigen::Map<Eigen::Array<float, 1, Eigen::Dynamic>> ep{
+      param, static_cast<Eigen::Index>(size)};
+
+  ep -= lr * eg;
+}

elasticdl/pkg/kernel/capi/kernel_api.h

@@ -0,0 +1,13 @@
+#ifndef ELASTICDL_PKG_KERNEL_CAPI_KERNEL_API_H_
+#define ELASTICDL_PKG_KERNEL_CAPI_KERNEL_API_H_
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+void SGD(float* grad, float* param, double lr, long long size);
+
+#ifdef __cplusplus
+}
+#endif
+#endif

elasticdl/pkg/kernel/kernel.go

@@ -0,0 +1,13 @@
+package kernel
+
+// #cgo LDFLAGS: -L./capi -lkernel_api
+// #include "capi/kernel_api.h"
+import "C"
+import "unsafe"
+
+// SGD kernel
+func SGD(grad []float32, param []float32, lr float64, size int64) {
+	gradPtr := (*C.float)(unsafe.Pointer(&grad[0]))
+	paramPtr := (*C.float)(unsafe.Pointer(&param[0]))
+	C.SGD(gradPtr, paramPtr, C.double(lr), C.longlong(size))
+}

elasticdl/pkg/kernel/kernel_test.go

@@ -0,0 +1,27 @@
+package kernel
+
+import "testing"
+import "math/rand"
+import "github.com/stretchr/testify/assert"
+
+func TestSGD(t *testing.T) {
+	const size int = 10
+	a := make([]float32, size)
+	b := make([]float32, size)
+	var lr float32 = 0.1
+
+	for i := 0; i < size; i++ {
+		a[i] = rand.Float32()
+		b[i] = rand.Float32()
+	}
+
+	expected := make([]float32, size)
+
+	for i := 0; i < size; i++ {
+		expected[i] = b[i] - lr*a[i]
+	}
+
+	SGD(a, b, float64(lr), int64(size))
+
+	assert.Equal(t, b, expected)
+}

scripts/codestyle/clang_format.hook

@@ -0,0 +1 @@
+clang-format $@

scripts/codestyle/cpplint_precommit.hook

@@ -0,0 +1,27 @@
+#!/bin/bash
+
+TOTAL_ERRORS=0
+if [[ ! $TRAVIS_BRANCH ]]; then
+  # install cpplint on local machine.
+  if [[ ! $(which cpplint) ]]; then
+    pip install cpplint
+  fi
+  # diff files on local machine.
+  files=$(git diff --cached --name-status | awk 'Extra open brace or missing close brace2}')
+else
+  # diff files between PR and latest commit on Travis CI.
+  branch_ref=$(git rev-parse "$TRAVIS_BRANCH")
+  head_ref=$(git rev-parse HEAD)
+  files=$(git diff --name-status $branch_ref $head_ref | awk 'Extra open brace or missing close brace2}')
+fi
+# The trick to remove deleted files: https://stackoverflow.com/a/2413151
+for file in $files; do
+    if [[ $file =~ ^(patches/.*) ]]; then
+        continue;
+    else
+        cpplint --filter=-readability/fn_size $file;
+        TOTAL_ERRORS=$(expr $TOTAL_ERRORS + $?);
+    fi
+done
+
+exit $TOTAL_ERRORS