Logo ROOT   6.16/01
Reference Guide
TTreeCacheUnzip.h
Go to the documentation of this file.
1// @(#)root/tree:$Id$
2// Author: Rene Brun 04/06/2006
3
4/*************************************************************************
5 * Copyright (C) 1995-2000, Rene Brun and Fons Rademakers. *
6 * All rights reserved. *
7 * *
8 * For the licensing terms see $ROOTSYS/LICENSE. *
9 * For the list of contributors see $ROOTSYS/README/CREDITS. *
10 *************************************************************************/
11
12#ifndef ROOT_TTreeCacheUnzip
13#define ROOT_TTreeCacheUnzip
14
15
16//////////////////////////////////////////////////////////////////////////
17// //
18// TTreeCacheUnzip //
19// //
20// Specialization of TTreeCache for parallel Unzipping //
21// //
22// Fabrizio Furano (CERN) Aug 2009 //
23// Core TTree-related code borrowed from the previous version //
24// by Leandro Franco and Rene Brun //
25// //
26//////////////////////////////////////////////////////////////////////////
27
28#include "Bytes.h"
29#include "TTreeCache.h"
30#include "ROOT/TTaskGroup.hxx"
31#include <atomic>
32#include <queue>
33#include <memory>
34#include <vector>
35
36class TBasket;
37class TBranch;
38class TMutex;
39class TTree;
40
42
43public:
44 // We have three possibilities for the unzipping mode:
45 // enable, disable and force
47
48 // Unzipping states for a basket:
50
51protected:
52 // Unzipping state for baskets
53 struct UnzipState {
54 // Note: we cannot use std::unique_ptr<std::unique_ptr<char[]>[]> or vector of unique_ptr
55 // for fUnzipChunks since std::unique_ptr is not copy constructable.
56 // However, in future upgrade we cannot use make_vector in C++14.
57 std::unique_ptr<char[]> *fUnzipChunks; ///<! [fNseek] Individual unzipped chunks. Their summed size is kept under control.
58 std::vector<Int_t> fUnzipLen; ///<! [fNseek] Length of the unzipped buffers
59 std::atomic<Byte_t> *fUnzipStatus; ///<! [fNSeek]
60
62 fUnzipChunks = nullptr;
63 fUnzipStatus = nullptr;
64 }
66 if (fUnzipChunks) delete [] fUnzipChunks;
67 if (fUnzipStatus) delete [] fUnzipStatus;
68 }
69 void Clear(Int_t size);
70 Bool_t IsUntouched(Int_t index) const;
71 Bool_t IsProgress(Int_t index) const;
72 Bool_t IsFinished(Int_t index) const;
73 Bool_t IsUnzipped(Int_t index) const;
74 void Reset(Int_t oldSize, Int_t newSize);
75 void SetUntouched(Int_t index);
76 void SetProgress(Int_t index);
77 void SetFinished(Int_t index);
78 void SetMissed(Int_t index);
79 void SetUnzipped(Int_t index, char* buf, Int_t len);
81 };
82
83 typedef struct UnzipState UnzipState_t;
85
86 // Members for paral. managing
90 Bool_t fParallel; ///< Indicate if we want to activate the parallelism (for this instance)
91
93
94 static TTreeCacheUnzip::EParUnzipMode fgParallel; ///< Indicate if we want to activate the parallelism
95
96 // IMT TTaskGroup Manager
97#ifdef R__USE_IMT
98 std::unique_ptr<ROOT::Experimental::TTaskGroup> fUnzipTaskGroup;
99#endif
100
101 // Unzipping related members
102 Int_t fNseekMax; ///<! fNseek can change so we need to know its max size
103 Int_t fUnzipGroupSize; ///<! Min accumulated size of a group of baskets ready to be unzipped by a IMT task
104 Long64_t fUnzipBufferSize; ///<! Max Size for the ready unzipped blocks (default is 2*fBufferSize)
105
106 static Double_t fgRelBuffSize; ///< This is the percentage of the TTreeCacheUnzip that will be used
107
108 // Members use to keep statistics
109 Int_t fNFound; ///<! number of blocks that were found in the cache
110 Int_t fNMissed; ///<! number of blocks that were not found in the cache and were unzipped
111 Int_t fNStalls; ///<! number of hits which caused a stall
112 Int_t fNUnzip; ///<! number of blocks that were unzipped
113
114private:
115 TTreeCacheUnzip(const TTreeCacheUnzip &); //this class cannot be copied
117
120
121 // Private methods
122 void Init();
123
124public:
126 TTreeCacheUnzip(TTree *tree, Int_t buffersize=0);
127 virtual ~TTreeCacheUnzip();
128
129 virtual Int_t AddBranch(TBranch *b, Bool_t subbranches = kFALSE);
130 virtual Int_t AddBranch(const char *branch, Bool_t subbranches = kFALSE);
132 virtual Int_t ReadBufferExt(char *buf, Long64_t pos, Int_t len, Int_t &loc);
133 void SetEntryRange(Long64_t emin, Long64_t emax);
134 virtual void StopLearningPhase();
136
137 // Methods related to the thread
139 static Bool_t IsParallelUnzip();
141
142 // Unzipping related methods
143#ifdef R__USE_IMT
145#endif
146 Int_t GetRecordHeader(char *buf, Int_t maxbytes, Int_t &nbytes, Int_t &objlen, Int_t &keylen);
147 virtual Int_t GetUnzipBuffer(char **buf, Long64_t pos, Int_t len, Bool_t *free);
149 virtual void ResetCache();
150 virtual Int_t SetBufferSize(Int_t buffersize);
151 void SetUnzipBufferSize(Long64_t bufferSize);
152 void SetUnzipGroupSize(Int_t groupSize) { fUnzipGroupSize = groupSize; }
153 static void SetUnzipRelBufferSize(Float_t relbufferSize);
154 Int_t UnzipBuffer(char **dest, char *src);
155 Int_t UnzipCache(Int_t index);
156
157 // Methods to get stats
158 Int_t GetNUnzip() { return fNUnzip; }
160 Int_t GetNFound() { return fNFound; }
161
162 void Print(Option_t* option = "") const;
163
164 // static members
165 ClassDef(TTreeCacheUnzip,0) //Specialization of TTreeCache for parallel unzipping
166};
167
168#endif
#define b(i)
Definition: RSha256.hxx:100
int Int_t
Definition: RtypesCore.h:41
const Bool_t kFALSE
Definition: RtypesCore.h:88
bool Bool_t
Definition: RtypesCore.h:59
double Double_t
Definition: RtypesCore.h:55
long long Long64_t
Definition: RtypesCore.h:69
float Float_t
Definition: RtypesCore.h:53
const char Option_t
Definition: RtypesCore.h:62
#define ClassDef(name, id)
Definition: Rtypes.h:324
#define free
Definition: civetweb.c:1539
Manages buffers for branches of a Tree.
Definition: TBasket.h:34
A TTree is a list of TBranches.
Definition: TBranch.h:64
Definition: TMutex.h:30
Specialization of TTreeCache for parallel Unzipping.
virtual void StopLearningPhase()
It's the same as TTreeCache::StopLearningPhase but we guarantee that we start the unzipping just afte...
UnzipState_t fUnzipState
Bool_t FillBuffer()
Fill the cache buffer with the branches in the cache.
void SetUnzipGroupSize(Int_t groupSize)
void Init()
Initialization procedure common to all the constructors.
Int_t UnzipCache(Int_t index)
This inflates a basket in the cache.
Int_t fNMissed
! number of blocks that were not found in the cache and were unzipped
Int_t GetUnzipGroupSize()
void SetEntryRange(Long64_t emin, Long64_t emax)
Set the minimum and maximum entry number to be processed this information helps to optimize the numbe...
TTreeCacheUnzip(const TTreeCacheUnzip &)
virtual Int_t ReadBufferExt(char *buf, Long64_t pos, Int_t len, Int_t &loc)
Int_t fNseekMax
! fNseek can change so we need to know its max size
virtual ~TTreeCacheUnzip()
Destructor. (in general called by the TFile destructor)
void UpdateBranches(TTree *tree)
update pointer to current Tree and recompute pointers to the branches in the cache
Int_t fNStalls
! number of hits which caused a stall
static TTreeCacheUnzip::EParUnzipMode fgParallel
Indicate if we want to activate the parallelism.
std::unique_ptr< ROOT::Experimental::TTaskGroup > fUnzipTaskGroup
static Int_t SetParallelUnzip(TTreeCacheUnzip::EParUnzipMode option=TTreeCacheUnzip::kEnable)
Static function that (de)activates multithreading unzipping.
Int_t UnzipBuffer(char **dest, char *src)
Unzips a ROOT specific buffer... by reading the header at the beginning.
Int_t GetRecordHeader(char *buf, Int_t maxbytes, Int_t &nbytes, Int_t &objlen, Int_t &keylen)
Read the logical record header from the buffer buf.
static void SetUnzipRelBufferSize(Float_t relbufferSize)
static function: Sets the unzip relatibe buffer size
void Print(Option_t *option="") const
Print cache statistics.
Int_t fNUnzip
! number of blocks that were unzipped
Int_t CreateTasks()
We create a TTaskGroup and asynchronously maps each group of baskets(> 100 kB in total) to a task.
Bool_t fParallel
Indicate if we want to activate the parallelism (for this instance)
Long64_t fUnzipBufferSize
! Max Size for the ready unzipped blocks (default is 2*fBufferSize)
static Bool_t IsParallelUnzip()
Static function that tells wether the multithreading unzipping is activated.
void SetUnzipBufferSize(Long64_t bufferSize)
Sets the size for the unzipping cache... by default it should be two times the size of the prefetchin...
Int_t fUnzipGroupSize
! Min accumulated size of a group of baskets ready to be unzipped by a IMT task
virtual Int_t SetBufferSize(Int_t buffersize)
Change the underlying buffer size of the cache.
Int_t fNFound
! number of blocks that were found in the cache
virtual Int_t GetUnzipBuffer(char **buf, Long64_t pos, Int_t len, Bool_t *free)
We try to read a buffer that has already been unzipped Returns -1 in case of read failure,...
virtual Int_t AddBranch(TBranch *b, Bool_t subbranches=kFALSE)
Add a branch to the list of branches to be stored in the cache this function is called by TBranch::Ge...
static Double_t fgRelBuffSize
This is the percentage of the TTreeCacheUnzip that will be used.
virtual void ResetCache()
This will delete the list of buffers that are in the unzipping cache and will reset certain values in...
static EParUnzipMode GetParallelUnzip()
Static function that returns the parallel option (to indicate an additional thread)
TTreeCacheUnzip & operator=(const TTreeCacheUnzip &)
A TTree object has a header with a name and a title.
Definition: TTree.h:71
Definition: tree.py:1
Bool_t IsFinished(Int_t index) const
void Reset(Int_t oldSize, Int_t newSize)
Reset all baskets' state arrays.
void Clear(Int_t size)
Clear all baskets' state arrays.
Bool_t IsProgress(Int_t index) const
std::atomic< Byte_t > * fUnzipStatus
! [fNSeek]
void SetUnzipped(Int_t index, char *buf, Int_t len)
std::vector< Int_t > fUnzipLen
! [fNseek] Length of the unzipped buffers
Bool_t TryUnzipping(Int_t index)
Start unzipping the basket if it is untouched yet.
void SetProgress(Int_t index)
void SetUntouched(Int_t index)
Bool_t IsUnzipped(Int_t index) const
Check if the basket is unzipped already.
Bool_t IsUntouched(Int_t index) const
void SetFinished(Int_t index)
Set cache as finished.
std::unique_ptr< char[]> * fUnzipChunks
! [fNseek] Individual unzipped chunks. Their summed size is kept under control.
#define dest(otri, vertexptr)
Definition: triangle.c:1040