From 14bf7d003c8541debcbe090bcfd8d6bf11cc49aa Mon Sep 17 00:00:00 2001
From: Christian Zimmermann <christian.zimmermann@ur.de>
Date: Wed, 1 Nov 2017 21:26:45 +0100
Subject: [PATCH] start re-implementing Contraction + performance check
 (AWESOME) + optimization level 3

---
 CMakeLists.txt               |   2 +-
 src/multi_array_operation.cc |   8 ---
 src/multi_array_operation.h  |  22 ++++++++
 src/op_unit_test.cc          | 100 +++++++++++++++++++++++++++++++++++
 4 files changed, 123 insertions(+), 9 deletions(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index eb355d8..6106bca 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -2,7 +2,7 @@ cmake_minimum_required(VERSION 2.8)
 
 project(multi_array)
 
-set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wall -Werror -std=c++11 --pedantic")
+set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wall -Werror -std=c++11 --pedantic -O3")
 
 enable_testing()
 
diff --git a/src/multi_array_operation.cc b/src/multi_array_operation.cc
index dd8ac64..84ccfa5 100644
--- a/src/multi_array_operation.cc
+++ b/src/multi_array_operation.cc
@@ -122,14 +122,6 @@ namespace MultiArrayTools
 	return mp.begin()->first;
     }
 
-    
-    /*********************************
-     *   MultiArrayOperationBase     *
-     *********************************/
-    
-    // purely virtual at the moment
-
-
     /***************************
      *   OperationTemplate     *
      ***************************/
diff --git a/src/multi_array_operation.h b/src/multi_array_operation.h
index b5b883d..7f3db9c 100644
--- a/src/multi_array_operation.h
+++ b/src/multi_array_operation.h
@@ -223,6 +223,28 @@ namespace MultiArrayTools
 	mutable BlockResult<T> mRes;
     };
 
+    template <typename T, class Op>
+    class Contraction : public OperationTemplate<T,Contraction<T,Op> >
+    {
+    public:
+
+	typedef T value_type;
+	typedef OperationTemplate<T,Contraction<T,Op> > OT;
+
+	Contraction(const Op& op);
+	
+	const BlockResult<T>& get() const;
+
+	std::vector<BTSS> block(const std::shared_ptr<IndexBase> blockIndex) const;
+	const Contraction& block() const;
+	
+    protected:
+
+	const Op& mOp;
+	mutable BlockResult<T> mRes;
+	
+    };
+    
 }
 
 #include "multi_array_operation.cc"
diff --git a/src/op_unit_test.cc b/src/op_unit_test.cc
index 644f276..3085da7 100644
--- a/src/op_unit_test.cc
+++ b/src/op_unit_test.cc
@@ -6,6 +6,8 @@
 
 #include "multi_array_header.h"
 
+#include <ctime>
+
 namespace MAT = MultiArrayTools;
 
 namespace {
@@ -25,6 +27,15 @@ namespace {
 	fptr = nptr;
     }
 
+    template <class Factory, typename T>
+    void swapFactory(std::shared_ptr<RangeFactoryBase>& fptr, std::vector<T>& ilist)
+    {
+	std::vector<T> tmp = ilist;
+	auto nptr = std::make_shared<Factory>( tmp );
+	fptr = nptr;
+    }
+
+    
     template <class Factory, class... Rs>
     void swapMFactory(std::shared_ptr<RangeFactoryBase>& fptr, const Rs&... rs)
     {
@@ -101,6 +112,95 @@ namespace {
 	std::vector<double> v4 = { 1.470, 2.210 };
     };
 
+    class OpTest_Performance : public ::testing::Test
+    {
+    protected:
+
+	typedef SingleRangeFactory<size_t,RangeType::ANY> SRF;
+	typedef SRF::oType SRange;
+
+	typedef MultiRangeFactory<SRange,SRange> MRF;
+	typedef MRF::oType MRange;
+
+	OpTest_Performance()
+	{
+	
+	    std::vector<size_t> initvec1(vs1);
+	    cv1.resize(vs1);
+	    for(size_t i = 0; i != vs1; ++i){
+		initvec1[i] = i;
+		cv1[i] = sqrt( static_cast<double>(i)*0.53 );
+	    }
+
+	    std::vector<size_t> initvec2(vs2);
+	    cv2.resize(vs2*vs1);
+	    for(size_t i = 0; i != vs2; ++i){
+		initvec2[i] = i;
+		for(size_t j = 0; j != vs1; ++j){
+		    cv2[i*vs1 + j] = static_cast<double>(i) * sin(static_cast<double>(j)*0.4);
+		}
+	    }
+
+	    swapFactory<SRF>(rfbptr, initvec1);
+	    sr1ptr = std::dynamic_pointer_cast<SRange>(rfbptr->create());
+
+	    swapFactory<SRF>(rfbptr, initvec2);
+	    sr2ptr = std::dynamic_pointer_cast<SRange>(rfbptr->create());
+
+	    swapMFactory<MRF>(rfbptr, sr2ptr, sr1ptr);
+	    mrptr = std::dynamic_pointer_cast<MRange>(rfbptr->create());
+	}
+
+	const size_t vs1 = 10000;
+	const size_t vs2 = 1000;	
+
+	std::shared_ptr<RangeFactoryBase> rfbptr;
+	std::shared_ptr<SRange> sr1ptr;
+	std::shared_ptr<SRange> sr2ptr;
+	std::shared_ptr<MRange> mrptr;
+
+	std::vector<double> cv1;
+	std::vector<double> cv2;
+    };
+
+    TEST_F(OpTest_Performance, PCheck)
+    {
+	MultiArray<double,MRange> ma2(mrptr, cv2);
+	MultiArray<double,SRange> ma1(sr1ptr, cv1);
+	MultiArray<double,MRange> res(mrptr);
+
+	auto si1 = std::dynamic_pointer_cast<SRange::IndexType>( sr1ptr->index() );
+	auto si2 = std::dynamic_pointer_cast<SRange::IndexType>( sr2ptr->index() );
+	auto mi = std::dynamic_pointer_cast<MRange::IndexType>( mrptr->index() );
+	(*mi)(si2, si1);
+
+	std::clock_t begin = std::clock();
+	
+	res(mi) = ma2(mi) * ma1(si1);
+
+	std::clock_t end = std::clock();
+	std::cout << "MultiArray time: " << static_cast<double>( end - begin ) / CLOCKS_PER_SEC
+		  << std::endl;
+
+	std::clock_t begin2 = std::clock();
+	
+	std::vector<double> res2(vs1*vs2);
+	for(size_t i = 0; i != vs2; ++i){
+	    for(size_t j = 0; j != vs1; ++j){
+		res2[i*vs1 + j] = cv1[j] * cv2[i*vs1 + j];
+	    }
+	}
+
+	std::clock_t end2 = std::clock();
+	std::cout << "std::vector - for loop time: " << static_cast<double>( end2 - begin2 ) / CLOCKS_PER_SEC
+		  << std::endl;
+
+	std::cout << "ratio: " << static_cast<double>( end - begin ) / static_cast<double>( end2 - begin2 ) << std::endl;
+
+	EXPECT_EQ( xround( res.at(mkt(7,9)) ), xround(res2[7*vs1 + 9]) );
+	
+    }
+    
     TEST_F(OpTest_1Dim, ExecOp)
     {
 	MultiArray<double,SRange> ma1(srptr, v1);