[Genabel-commits] r1469 - in pkg: . OmicABELnoMM

noreply at r-forge.r-project.org noreply at r-forge.r-project.org
Mon Dec 16 09:37:16 CET 2013


Author: lckarssen
Date: 2013-12-16 09:37:15 +0100 (Mon, 16 Dec 2013)
New Revision: 1469

Added:
   pkg/OmicABELnoMM/
   pkg/OmicABELnoMM/AIOwrapper.cpp
   pkg/OmicABELnoMM/AIOwrapper.h
   pkg/OmicABELnoMM/Algorithm.cpp
   pkg/OmicABELnoMM/Algorithm.h
   pkg/OmicABELnoMM/CreateData.R
   pkg/OmicABELnoMM/Definitions.h
   pkg/OmicABELnoMM/Utility.cpp
   pkg/OmicABELnoMM/Utility.h
   pkg/OmicABELnoMM/XL.fvd
   pkg/OmicABELnoMM/XL.fvi
   pkg/OmicABELnoMM/XR.fvd
   pkg/OmicABELnoMM/XR.fvi
   pkg/OmicABELnoMM/Y.fvd
   pkg/OmicABELnoMM/Y.fvi
   pkg/OmicABELnoMM/howtocompile.txt
   pkg/OmicABELnoMM/main.cpp
Log:
First commit of OmicABELnoMM, written by Alvaro Jesus Frank. The name refers to OmicABEL, but without support for mixed models. 

Note that OmicABELnoMM is under full development.


Added: pkg/OmicABELnoMM/AIOwrapper.cpp
===================================================================
--- pkg/OmicABELnoMM/AIOwrapper.cpp	                        (rev 0)
+++ pkg/OmicABELnoMM/AIOwrapper.cpp	2013-12-16 08:37:15 UTC (rev 1469)
@@ -0,0 +1,778 @@
+#include "AIOwrapper.h"
+
+
+
+AIOwrapper::AIOwrapper()
+{
+    Fhandler = &FHandler;
+    io_overhead = "*";
+}
+
+AIOwrapper::~AIOwrapper()
+{
+
+}
+
+
+void AIOwrapper::initialize(struct Settings &params)
+{
+
+
+    pthread_mutex_init( &(FHandler.m_more), NULL);
+    pthread_mutex_init( &(FHandler.m_read), NULL);
+    pthread_mutex_init( &(FHandler.m_buff_upd), NULL);
+
+    pthread_cond_init(&(FHandler.condition_more), NULL);
+    pthread_cond_init(&(FHandler.condition_read), NULL);
+
+    pthread_barrier_init(&(FHandler.finalize_barrier),NULL,2);
+
+
+    Fhandler->fakefiles = params.use_fake_files;
+
+    databel_fvi* Yfvi;
+    databel_fvi* ALfvi;
+    databel_fvi* ARfvi;
+
+    if(!Fhandler->fakefiles)
+    {
+        Fhandler->fnameAL = params.fnameAL;
+        Fhandler->fnameAR = params.fnameAR;
+        Fhandler->fnameY = params.fnameY;
+        Fhandler->fnameOutB = params.fnameOutB;
+
+        databel_fvi* Yfvi  = load_databel_fvi( (Fhandler->fnameY+".fvi").c_str() );
+        databel_fvi* ALfvi = load_databel_fvi( (Fhandler->fnameAL+".fvi").c_str() );
+        databel_fvi* ARfvi = load_databel_fvi( (Fhandler->fnameAR+".fvi").c_str() );
+        params.n = ALfvi->fvi_header.numObservations;
+        params.m = ARfvi->fvi_header.numVariables/params.r;
+        params.t = Yfvi->fvi_header.numVariables;
+        params.l = ALfvi->fvi_header.numVariables;
+
+        //block size to keep mem under 1 gigabyte
+        int opt_block = params.n/(4*1000^3)*(1/(2*params.r));
+        int opt_tb = 1000;
+        int opt_mb = 1000;
+
+        params.mb = min(params.m,opt_tb);
+        params.tb = min(params.t,opt_mb);
+        }
+    else
+    {
+
+    }
+
+    params.p = params.l + params.r;
+
+
+    //block size to keep mem under 1 gigabyte
+//    int opt_block = params.n/(4*1000^3)*(1/(2*params.r));
+//    int opt_tb = max(4*2000,opt_block);
+//    int opt_mb = max(2000,opt_block);
+//
+//    params.mb = min(params.m,opt_tb);
+//    params.tb = min(params.t,opt_mb);
+
+    prepare_B();
+    prepare_AL(params.l,params.n);
+    prepare_AR(  params.mb,  params.n,  params.m,  params.r);
+    prepare_Y(params.tb, params.n, params.t);
+
+
+
+
+
+}
+
+void AIOwrapper::finalize()
+{
+    //cout << "f";
+    void *status;
+
+    Fhandler->not_done = false;
+    pthread_mutex_lock(&(Fhandler->m_more));
+    pthread_cond_signal( &(Fhandler->condition_more ));
+    pthread_mutex_unlock(&(Fhandler->m_more));
+
+    pthread_barrier_wait(&(Fhandler->finalize_barrier));
+    finalize_Y();
+    finalize_AR();
+    finalize_AL();
+    finalize_B();
+
+    pthread_attr_destroy(&(Fhandler->attr));
+
+    pthread_mutex_destroy(&(Fhandler->m_more));
+    pthread_cond_destroy(&(Fhandler->condition_more));
+
+    pthread_mutex_destroy(&(Fhandler->m_read));
+    pthread_cond_destroy(&(Fhandler->condition_read));
+
+
+}
+
+
+void AIOwrapper::prepare_B()
+{
+    Fhandler->fp_B = fopen((Fhandler->fnameOutB+".fvd").c_str(), "w+b");
+    if(Fhandler->fp_B == 0)
+    {
+        cout << "Error Opening File B " << Fhandler->fnameOutB << endl;
+        exit(1);
+    }
+}
+
+
+void AIOwrapper::finalize_B()
+{
+    fclose(Fhandler->fp_B);
+}
+
+
+void* AIOwrapper::async_io( void *ptr )
+{
+    //cout << "async_io\n" << flush;
+    type_fileh* Fhandler = (type_fileh *)ptr;
+    int size_buff,tmp_y_blockSize,tmp_ar_blockSize;
+
+    struct timespec timeToWait;
+    FILE*  fp_Y;
+
+    FILE*  fp_Ar;
+    if(!Fhandler->fakefiles)
+    {
+        fp_Y = fopen((Fhandler->fnameY+".fvd").c_str(), "rb");
+        if(fp_Y == 0)
+        {
+            cout << "Error Reading File Y " << Fhandler->fnameY << endl;
+            exit(1);
+        }
+
+        fp_Ar = fopen((Fhandler->fnameAR+".fvd").c_str(), "rb");
+        if(fp_Ar == 0)
+        {
+            cout << "Error Reading File Xr " << Fhandler->fnameAR << endl;
+            exit(1);
+        }
+
+
+    }
+    else
+    {
+//        fp_Y = fopen("tempY.bin", "w+b");
+//        if(fp_Y == 0)
+//        {
+//            cout << "Error creating temp File Y " << Fhandler->fnameY << endl;
+//            exit(1);
+//        }
+//        fwrite(Fhandler->Yb, sizeof(type_precision), Fhandler->n*Fhandler->y_blockSize, fp_Y);
+//        fclose(fp_Y);
+//
+//
+//        fp_Ar = fopen("tempAR.bin", "w+b");
+//        if(fp_Ar == 0)
+//        {
+//            cout << "Error creating temp File Y " << Fhandler->fnameY << endl;
+//            exit(1);
+//        }
+//        fwrite(Fhandler->Yb, sizeof(type_precision), Fhandler->n*Fhandler->y_blockSize, fp_Ar);
+//        fclose(fp_Ar);
+//
+//        fp_Y = fopen("tempY.bin", "rb");
+//        if(fp_Y == 0)
+//        {
+//            cout << "Error Reading File Y " << Fhandler->fnameY << endl;
+//            exit(1);
+//        }
+//
+//        fp_Ar = fopen("tempAR.bin", "rb");
+//        if(fp_Ar == 0)
+//        {
+//            cout << "Error Reading File Xr " << Fhandler->fnameAR << endl;
+//            exit(1);
+//        }
+
+    }
+
+
+    Fhandler->not_done = true;
+    Fhandler->reset_wait = false;
+
+    //cout << "c" << flush;
+
+    while(Fhandler->not_done)
+    {
+
+        while(!Fhandler->empty_buffers.empty() && Fhandler->y_to_readSize)
+        {
+
+            tmp_y_blockSize = Fhandler->y_blockSize;
+            if(Fhandler->y_to_readSize < Fhandler->y_blockSize)
+                tmp_y_blockSize = Fhandler->y_to_readSize;
+
+            Fhandler->y_to_readSize -= tmp_y_blockSize;
+            size_buff = Fhandler->n * tmp_y_blockSize;
+            //cout << Fhandler->y_to_readSize << endl;
+
+            pthread_mutex_lock(&(Fhandler->m_buff_upd));
+            //cout << " pre;" << Fhandler->full_buffers.size() << ";" << Fhandler->empty_buffers.size() << endl;
+            type_buffElement* tobeFilled = Fhandler->empty_buffers.front();
+            Fhandler->empty_buffers.pop();
+            //pthread_mutex_unlock(&(Fhandler->m_buff_upd));
+
+            tobeFilled->size = tmp_y_blockSize;
+
+            if(Fhandler->fakefiles)
+            {
+                //fread (tobeFilled->buff,sizeof(type_precision),size,fp_Y);
+                //fseek ( fp_Y , 0 , SEEK_SET );
+                int old_seed = Fhandler->seed;
+                srand (old_seed);
+                re_random_vec(tobeFilled->buff, size_buff );
+                re_random_vec_nan(tobeFilled->buff, size_buff );
+                Fhandler->seed += 75;
+            }
+            else
+            {
+                fread (tobeFilled->buff,sizeof(type_precision),size_buff,fp_Y);
+                if(Fhandler->y_to_readSize <= 0)
+                {
+                    fseek ( fp_Y , 0 , SEEK_SET );
+                }
+            }
+
+
+            //pthread_mutex_lock(&(Fhandler->m_buff_upd));
+            Fhandler->full_buffers.push(tobeFilled);
+            //  cout << "\nStoring " << tobeFilled << endl;
+            //cout << " post;" << Fhandler->full_buffers.size() << ";" << Fhandler->empty_buffers.size() << endl;
+            pthread_mutex_unlock(&(Fhandler->m_buff_upd));
+
+            pthread_mutex_lock(&(Fhandler->m_read));
+            pthread_cond_signal( &(Fhandler->condition_read ));
+            pthread_mutex_unlock(&(Fhandler->m_read));
+
+        }
+
+        while(!Fhandler->ar_empty_buffers.empty() && Fhandler->Ar_to_readSize)
+        {
+            tmp_ar_blockSize = Fhandler->Ar_blockSize;
+            if(Fhandler->Ar_to_readSize < Fhandler->Ar_blockSize)
+                tmp_ar_blockSize = Fhandler->Ar_to_readSize;
+
+            Fhandler->Ar_to_readSize -= tmp_ar_blockSize;
+            size_buff = Fhandler->n * tmp_ar_blockSize*Fhandler->r;
+
+            pthread_mutex_lock(&(Fhandler->m_buff_upd));
+            type_buffElement* tobeFilled = Fhandler->ar_empty_buffers.front();
+            Fhandler->ar_empty_buffers.pop();
+
+
+            tobeFilled->size = tmp_ar_blockSize;
+
+            if(Fhandler->fakefiles)
+            {
+                //fread (tobeFilled->buff,sizeof(type_precision),size,fp_Ar);
+                //fseek ( fp_Ar , 0 , SEEK_SET );
+                re_random_vec(tobeFilled->buff , Fhandler->n * tmp_ar_blockSize*Fhandler->r );
+                re_random_vec_nan(tobeFilled->buff , Fhandler->n * tmp_ar_blockSize*Fhandler->r );
+
+            }
+            else
+            {
+                fread (tobeFilled->buff,sizeof(type_precision),size_buff,fp_Ar);
+                if(Fhandler->Ar_to_readSize <= 0)
+                {
+                    fseek ( fp_Ar , 0 , SEEK_SET );
+                }
+            }
+
+            Fhandler->ar_full_buffers.push(tobeFilled);
+            //  cout << "\nStoring " << tobeFilled << endl;
+            pthread_mutex_unlock(&(Fhandler->m_buff_upd));
+
+            pthread_mutex_lock(&(Fhandler->m_read));
+            pthread_cond_signal( &(Fhandler->condition_read ));
+            pthread_mutex_unlock(&(Fhandler->m_read));
+
+        }
+
+
+
+
+#ifdef WINDOWS
+        SYSTEMTIME time;
+        GetSystemTime(&time);
+
+        timeToWait.tv_sec = time.wSecond + 500/1000;
+        long int morenanos = (500%1000)*1000000;
+        timeToWait.tv_nsec = time.wMilliseconds*1000 + morenanos ;
+#else
+        clock_gettime(CLOCK_REALTIME, &timeToWait);
+        timeToWait.tv_nsec += 150;
+#endif
+
+        pthread_mutex_lock(&(Fhandler->m_more));
+        pthread_cond_timedwait( &(Fhandler->condition_more), &(Fhandler->m_more), &timeToWait );
+        pthread_mutex_unlock( &(Fhandler->m_more ));
+
+        pthread_mutex_lock(&(Fhandler->m_read));
+        pthread_cond_signal( &(Fhandler->condition_read ));
+        pthread_mutex_unlock(&(Fhandler->m_read));
+
+        if(Fhandler->reset_wait)
+        {
+            pthread_barrier_wait(&(Fhandler->finalize_barrier));
+            //wait for main thread to reset everything
+            pthread_barrier_wait(&(Fhandler->finalize_barrier));
+        }
+
+
+    }
+    //cout << "k" << flush;
+    //barrier
+    pthread_barrier_wait(&(Fhandler->finalize_barrier));
+
+    type_buffElement* tmp;
+    while(!Fhandler->full_buffers.empty())
+    {
+       tmp= Fhandler->full_buffers.front();
+       Fhandler->full_buffers.pop();
+       delete []tmp->buff;
+       delete tmp;
+    }
+
+    while(!Fhandler->empty_buffers.empty())
+    {
+       tmp= Fhandler->empty_buffers.front();
+       Fhandler->empty_buffers.pop();
+       delete []tmp->buff;
+       delete tmp;
+    }
+
+    while(!Fhandler->ar_full_buffers.empty())
+    {
+       tmp= Fhandler->ar_full_buffers.front();
+       Fhandler->ar_full_buffers.pop();
+       delete tmp->buff;
+       delete tmp;
+    }
+
+    while(!Fhandler->ar_empty_buffers.empty())
+    {
+       tmp= Fhandler->ar_empty_buffers.front();
+       Fhandler->ar_empty_buffers.pop();
+       delete []tmp->buff;
+       delete tmp;
+    }
+
+    //
+
+    //pthread_exit(NULL);
+
+    if(!Fhandler->fakefiles)
+    {
+        fclose(fp_Y);
+        fclose(fp_Ar);
+
+    }
+    else
+    {
+        fclose(fp_Y);
+        fclose(fp_Ar);
+    }
+
+//
+//            //!induce realistic fileread delay
+
+}
+
+void AIOwrapper::load_ARblock(type_precision** Ar, int &Ar_blockSize)
+{
+
+    int status;
+    int createstatus = 0;
+    //cout<<"^";
+
+    while(Fhandler->ar_full_buffers.empty())
+    {
+        pthread_mutex_lock(&(Fhandler->m_more));
+        pthread_cond_signal( &(Fhandler->condition_more ));
+        pthread_mutex_unlock(&(Fhandler->m_more));
+
+        io_overhead = "#";
+
+        pthread_mutex_lock(&(Fhandler->m_read));
+        pthread_cond_wait( &(Fhandler->condition_read), &(Fhandler->m_read ));
+        pthread_mutex_unlock(&(Fhandler->m_read));
+    }
+
+
+    //!read new rdy buffer
+    pthread_mutex_lock(&(Fhandler->m_buff_upd));
+        if(Fhandler->Ar_currentReadBuff)
+        {
+            Fhandler->ar_empty_buffers.push(Fhandler->Ar_currentReadBuff);
+        }
+
+        Fhandler->Ar_currentReadBuff = Fhandler->ar_full_buffers.front();
+        Fhandler->ar_full_buffers.pop();
+
+    //cout << "\nReading " << Fhandler->Ar_currentReadBuff << endl;
+    Fhandler->Ar = Fhandler->Ar_currentReadBuff->buff;
+    Ar_blockSize = Fhandler->Ar_currentReadBuff->size;
+
+    pthread_mutex_unlock(&(Fhandler->m_buff_upd));
+
+
+
+    pthread_mutex_lock(&(Fhandler->m_more));
+    pthread_cond_signal( &(Fhandler->condition_more ));
+    pthread_mutex_unlock(&(Fhandler->m_more));
+
+
+
+//secueantial
+//    int tmp_ar_blockSize = Fhandler->Ar_blockSize;
+//    if(Fhandler->Ar_to_readSize < Fhandler->Ar_blockSize)
+//        tmp_ar_blockSize = Fhandler->Ar_to_readSize;
+//
+//    Fhandler->Ar_to_readSize -= tmp_ar_blockSize;
+//    int size_buff = Fhandler->n * tmp_ar_blockSize * Fhandler->r;
+//    re_random_vec(Fhandler->Ar,size_buff);
+//Ar_blockSize=tmp_ar_blockSize;
+
+
+    (*Ar) = Fhandler->Ar;
+
+
+
+
+
+
+}
+
+void AIOwrapper::load_Yblock(type_precision** Y, int &y_blockSize)
+{
+
+    int status;
+    int createstatus = 0;
+
+    while(Fhandler->full_buffers.empty())
+    {
+
+        pthread_mutex_lock(&(Fhandler->m_more));
+        pthread_cond_signal( &(Fhandler->condition_more ));
+        pthread_mutex_unlock(&(Fhandler->m_more));
+
+        io_overhead = "!";
+
+        pthread_mutex_lock(&(Fhandler->m_read));
+        pthread_cond_wait( &(Fhandler->condition_read), &(Fhandler->m_read ));
+        pthread_mutex_unlock(&(Fhandler->m_read));
+
+    }
+
+    //!read new rdy buffer
+    pthread_mutex_lock(&(Fhandler->m_buff_upd));
+    //cout << " pre," << Fhandler->full_buffers.size() << ";" << Fhandler->empty_buffers.size() << endl;
+
+        if(Fhandler->currentReadBuff)
+        {
+            //memset(Fhandler->currentReadBuff->buff,0,y_blockSize);
+            Fhandler->empty_buffers.push(Fhandler->currentReadBuff);
+        }
+        Fhandler->currentReadBuff = Fhandler->full_buffers.front();
+        Fhandler->full_buffers.pop();
+
+    //cout << "\nReading " << Fhandler->currentReadBuff << endl;
+    Fhandler->Yb = Fhandler->currentReadBuff->buff;
+    y_blockSize = Fhandler->currentReadBuff->size;
+
+    (*Y) = Fhandler->Yb;
+
+     //cout << " post," << Fhandler->full_buffers.size() << ";" << Fhandler->empty_buffers.size() << endl;
+
+    pthread_mutex_unlock(&(Fhandler->m_buff_upd));
+
+
+
+    pthread_mutex_lock(&(Fhandler->m_more));
+    pthread_cond_signal( &(Fhandler->condition_more ));
+    pthread_mutex_unlock(&(Fhandler->m_more));
+
+    //matlab_print_matrix("Y",Fhandler->n,y_blockSize,*Y);
+
+
+}
+
+void AIOwrapper::write_B(type_precision* B, int p, int blockSize)
+{
+    fwrite (B , sizeof(type_precision), p*blockSize, Fhandler->fp_B);
+}
+
+void AIOwrapper::prepare_Y(int y_blockSize, int n, int totalY)
+{
+    //for fake files
+
+    Fhandler->seed = 1337;
+    srand (Fhandler->seed);
+
+    Fhandler->y_blockSize = y_blockSize;
+
+    Fhandler->n= n;
+    Fhandler->Y_Amount=totalY;
+    Fhandler->y_to_readSize = Fhandler->Y_Amount;
+    Fhandler->buff_count = min(2,(totalY+ y_blockSize - 1)/y_blockSize) ;
+    //cout << "buffcount " << Fhandler->buff_count;
+
+
+    Fhandler->currentReadBuff = 0;
+    type_buffElement* tmp;
+
+    for(int i = 0; i< Fhandler->buff_count  ; i++)
+    {
+        tmp = new type_buffElement();
+        tmp->buff = new type_precision[Fhandler->n*Fhandler->y_blockSize];
+        tmp->size = y_blockSize;
+//        for( int i = 0; i < Fhandler->n*Fhandler->y_blockSize; i++)
+//        {
+//            (tmp->buff)[i] = 0;
+//        }
+        Fhandler->empty_buffers.push(tmp);
+    }
+    Fhandler->Yb = tmp->buff;
+
+
+
+    pthread_mutex_init(&(Fhandler->m_buff_upd), NULL);
+    pthread_mutex_init(&(Fhandler->m_more), NULL);
+    pthread_mutex_init(&(Fhandler->m_read), NULL);
+
+    pthread_attr_init(&(Fhandler->attr));
+    pthread_attr_setdetachstate(&(Fhandler->attr), PTHREAD_CREATE_JOINABLE);
+
+    pthread_create( &(Fhandler->iothread),&(Fhandler->attr), AIOwrapper::async_io, (void*)Fhandler);
+
+
+
+}
+
+void AIOwrapper::reset_Y()
+{
+    void *status;
+
+    Fhandler->seed = 1337;
+
+    //cout << "ry" << flush;
+
+    Fhandler->reset_wait = true;
+    pthread_barrier_wait(&(Fhandler->finalize_barrier));
+
+    pthread_mutex_lock(&(Fhandler->m_buff_upd));
+    Fhandler->y_to_readSize = Fhandler->Y_Amount;
+
+    if(Fhandler->currentReadBuff)
+    {
+        Fhandler->full_buffers.push(Fhandler->currentReadBuff);
+        Fhandler->currentReadBuff=0;
+    }
+
+    while(!Fhandler->full_buffers.empty())
+    {
+        Fhandler->empty_buffers.push(Fhandler->full_buffers.front());
+        for( int i = 0; i < Fhandler->n*Fhandler->y_blockSize; i++)
+        {
+            ((Fhandler->full_buffers.front())->buff)[i] = 0;
+        }
+        Fhandler->full_buffers.pop();
+    }
+    pthread_mutex_unlock(&(Fhandler->m_buff_upd));
+
+    Fhandler->reset_wait = false;
+
+    pthread_barrier_wait(&(Fhandler->finalize_barrier));
+
+    pthread_mutex_lock(&(Fhandler->m_more));
+    pthread_cond_signal( &(Fhandler->condition_more ));
+    pthread_mutex_unlock(&(Fhandler->m_more));
+
+
+}
+
+void AIOwrapper::reset_AR()
+{
+    void *status;
+
+
+    //cout << "ra" << flush;
+
+    Fhandler->reset_wait = true;
+    pthread_barrier_wait(&(Fhandler->finalize_barrier));
+
+    pthread_mutex_lock(&(Fhandler->m_buff_upd));
+    Fhandler->Ar_to_readSize = Fhandler->Ar_Amount;
+
+    if(Fhandler->Ar_currentReadBuff)
+    {
+        Fhandler->ar_full_buffers.push(Fhandler->Ar_currentReadBuff);
+        Fhandler->Ar_currentReadBuff=0;
+    }
+
+    while(!Fhandler->ar_full_buffers.empty())
+    {
+        Fhandler->ar_empty_buffers.push(Fhandler->ar_full_buffers.front());
+//        for( int i = 0; i < Fhandler->n*Fhandler->r*Fhandler->Ar_blockSize; i++)
+//        {
+//            ((Fhandler->ar_full_buffers.front())->buff)[i] = 0;
+//        }
+        Fhandler->ar_full_buffers.pop();
+    }
+    pthread_mutex_unlock(&(Fhandler->m_buff_upd));
+
+    Fhandler->reset_wait = false;
+
+    pthread_barrier_wait(&(Fhandler->finalize_barrier));
+
+    pthread_mutex_lock(&(Fhandler->m_more));
+    pthread_cond_signal( &(Fhandler->condition_more ));
+    pthread_mutex_unlock(&(Fhandler->m_more));
+
+
+}
+
+void AIOwrapper::finalize_Y()
+{
+
+
+}
+
+void AIOwrapper::prepare_AR( int desired_blockSize, int n, int totalR, int columnsAR)
+{
+
+    Fhandler->Ar = new type_precision[desired_blockSize*columnsAR*n];
+    Fhandler->Ar_blockSize = desired_blockSize;
+    Fhandler->r = columnsAR;
+    Fhandler->Ar_Amount = totalR;
+    Fhandler->Ar_to_readSize = Fhandler->Ar_Amount;
+
+    int buff_count = min(2,(totalR+ desired_blockSize - 1)/desired_blockSize);
+
+    Fhandler->Ar_currentReadBuff = 0;
+    type_buffElement* tmp;
+
+    for(int i = 0; i< buff_count  ; i++)
+    {
+        tmp = new type_buffElement();
+        tmp->buff = new type_precision[n*desired_blockSize*columnsAR];
+        tmp->size = desired_blockSize;
+//        memset(tmp->buff,-9,n*desired_blockSize*columnsAR);
+
+        Fhandler->ar_empty_buffers.push(tmp);
+    }
+    Fhandler->Ar = tmp->buff;
+
+}
+
+
+void AIOwrapper::finalize_AR()
+{
+
+}
+
+
+void AIOwrapper::load_AL(type_precision** AL)
+{
+    if(Fhandler->fakefiles)
+    {
+        re_random_vec(Fhandler->AL,Fhandler->n*Fhandler->l);
+        re_random_vec_nan(Fhandler->AL,Fhandler->n*Fhandler->l);
+        (*AL) = Fhandler->AL;
+    }
+    else
+    {
+        FILE *fp;
+        fp = fopen((Fhandler->fnameAL+".fvd").c_str(), "rb");
+        if(fp == 0)
+        {
+            cout << "Error Reading File " << Fhandler->fnameAL << endl;
+            exit(1);
+        }
+
+        fread (Fhandler->AL,sizeof(type_precision),Fhandler->l*Fhandler->n,fp);
+        fclose(fp);
+    }
+
+
+
+    (*AL) = Fhandler->AL;
+
+    //matlab_print_matrix("AL",Fhandler->n,Fhandler->l,*AL);
+}
+
+void AIOwrapper::prepare_AL( int columnsAL, int n)
+{
+    Fhandler->AL = new type_precision[columnsAL*n];
+    Fhandler->l=columnsAL;
+}
+
+void AIOwrapper::finalize_AL()
+{
+    delete []Fhandler->AL;
+}
+
+void AIOwrapper::free_databel_fvi( struct databel_fvi **fvi )
+{
+	free ((*fvi)->fvi_data);
+	free (*fvi);
+	*fvi = NULL;
+}
+
+FILE * AIOwrapper::fgls_fopen( const char *path, const char *mode )
+{
+	FILE * f;
+	char err[100];
+
+	f = fopen( path, mode );
+	if ( f == NULL )
+	{
+		cout << "\nerror on fgls_fopen\n";
+		exit( 1 );
+	}
+	return f;
+}
+
+void * AIOwrapper::fgls_malloc_impl( const char* file, long line, size_t size )
+{
+    void *buf;
+
+    if ( (buf = malloc( size )) == NULL ) {
+        cout<< "\nCouldn't allocate %ld bytes of memory in %s:%ld\n";
+        exit(1);
+    }
+
+    return buf;
+}
+
+struct databel_fvi * AIOwrapper::load_databel_fvi( const char *path )
+{
+	FILE *f;
+	databel_fvi *fvi;
+	size_t data_size;
+
+	f = fgls_fopen( path, "r" );
+
+	fvi = (databel_fvi*) fgls_malloc( sizeof(databel_fvi) );
+	// Header
+	fread( &fvi->fvi_header, sizeof(databel_fvi_header), 1, f );
+	// Labels
+	data_size = (fvi->fvi_header.numVariables +fvi->fvi_header.numObservations ) *
+		         fvi->fvi_header.namelength * sizeof(char);
+	fvi->fvi_data = (char *) fgls_malloc ( data_size );
+	// Load labels
+	fread( fvi->fvi_data, 1, data_size, f );
+
+	fclose( f );
+
+	return fvi;
+}

Added: pkg/OmicABELnoMM/AIOwrapper.h
===================================================================
--- pkg/OmicABELnoMM/AIOwrapper.h	                        (rev 0)
+++ pkg/OmicABELnoMM/AIOwrapper.h	2013-12-16 08:37:15 UTC (rev 1469)
@@ -0,0 +1,178 @@
+#ifndef AIOWRAPPER_H
+#define AIOWRAPPER_H
+
+#include "Definitions.h"
+#include "Utility.h"
+
+typedef struct BufferElement type_buffElement;
+
+struct BufferElement
+{
+    type_precision* buff;
+    int size;
+};
+
+typedef struct fileh type_fileh;
+
+
+
+struct fileh
+{
+    string fnameAL;
+    string fnameAR;
+    string fnameY;
+    string fnameOutB;
+
+    FILE*  fp_B;
+
+    bool doublefileType;
+    bool fakefiles;
+
+    type_precision* Yb;
+    type_precision* Ar;
+    type_precision* AL;
+    type_buffElement* currentReadBuff;
+    type_buffElement* Ar_currentReadBuff;
+    int buff_count;
+
+    queue<type_buffElement*> empty_buffers;
+    queue<type_buffElement*> full_buffers;
+
+    queue<type_buffElement*> b_empty_buffers;
+    queue<type_buffElement*> b_full_buffers;
+
+    queue<type_buffElement*> ar_empty_buffers;
+    queue<type_buffElement*> ar_full_buffers;
+
+    int index;
+
+    int n;
+    int r;
+    int l;
+
+    int Ar_Amount;
+    int Ar_blockSize;
+    int Ar_to_readSize;
+
+    int Y_Amount;
+    int y_blockSize;
+    int y_to_readSize;
+
+    bool not_done;
+    bool reset_wait;
+
+    int seed;
+    int Aseed;
+
+    pthread_mutex_t m_more     ;
+    pthread_cond_t  condition_more   ;
+    pthread_mutex_t m_read     ;
+
+    pthread_cond_t  condition_read  ;
+    pthread_mutex_t m_buff_upd   ;
+
+    pthread_t iothread;
+    pthread_attr_t attr;
+
+    //barrier
+    pthread_barrier_t finalize_barrier;
+
+
+};
+
+#define fgls_malloc(size) fgls_malloc_impl(__FILE__, __LINE__, size)
+
+enum datatype{ UNSIGNED_SHORT_INT_TYPE = 1,
+               SHORT_INT_TYPE,
+               UNSIGNED_INT_TYPE,
+               INT_TYPE,
+               FLOAT_TYPE,
+               DOUBLE_TYPE,
+               SIGNED_CHAR_TYPE,
+               UNSIGNED_CHAR_TYPE };
+
+#define NAMELENGTH 32
+#define RESERVEDSPACE 5
+
+typedef struct databel_fvi_header
+{
+	unsigned short int type;
+	unsigned int nelements;
+	unsigned int numObservations;
+	unsigned int numVariables;
+	unsigned int bytesPerRecord;
+	unsigned int bitsPerRecord;
+	unsigned int namelength;
+	unsigned int reserved[RESERVEDSPACE];
+} databel_fvi_header;
+
+typedef struct databel_fvi
+{
+	databel_fvi_header  fvi_header;
+	char               *fvi_data;
+} databel_fvi;
+
+class AIOwrapper
+{
+    public:
+        AIOwrapper();
+        ~AIOwrapper();
+
+        void initialize(struct Settings &params);
+        void finalize();
+
+
+        void load_AL(type_precision** AL);
+        void load_ARblock(type_precision** Y, int &blockSize);
+        void load_Yblock(type_precision** Y, int &blockSize);
+        void reset_Y();
+        void reset_AR();
+
+        void write_B(type_precision* B, int p, int blockSize);
+
+        string io_overhead;
+
+
+    protected:
+
+    private:
+
+
+        void prepare_AR( int desired_blockSize, int n, int totalR, int columnsR);
+        void finalize_AR();
+
+
+        void prepare_Y(int desired_blockSize, int n, int totalY);
+        void finalize_Y();
+
+        void prepare_AL( int columns, int n);
+        void finalize_AL();
+
+        void prepare_B();
+        void finalize_B();
+
+
+        static void* async_io(void *ptr );
+
+        struct databel_fvi * load_databel_fvi( const char *path );
+        void free_databel_fvi( struct databel_fvi **fvi );
+        FILE * fgls_fopen( const char *path, const char *mode );
+
+        void * fgls_malloc_impl( const char* file, long line, size_t size );
+
+        type_fileh FHandler;
+        type_fileh* Fhandler;
+
+
+
+
+
+
+
+
+
+
+
+};
+
+#endif // AIOWRAPPER_H

Added: pkg/OmicABELnoMM/Algorithm.cpp
===================================================================
--- pkg/OmicABELnoMM/Algorithm.cpp	                        (rev 0)
+++ pkg/OmicABELnoMM/Algorithm.cpp	2013-12-16 08:37:15 UTC (rev 1469)
@@ -0,0 +1,1714 @@
+#include "Algorithm.h"
+
+Algorithm::Algorithm()
+{
+    //ctor
+}
+
+Algorithm::~Algorithm()
+{
+    //dtor
+}
+
+
+void Algorithm::solve(struct Settings params, struct Outputs &out, int type)
+{
+    switch (type)
+    {
+        case FULL_NEQ:
+                fullNEQ(params,out);
+            break;
+        case P_NEQ:
+                partialNEQ(params,out);
+            break;
+        case P_NEQ_B_OPT:
+                partialNEQ_Blocked_STL(params,out);
+            break;
+        case FULL_QR:
+                fullQR(params,out);
+            break;
+        case P_QR:
+                partialQR(params,out);
+            break;
+        case P_QR_B_OPT:
+                partialQR_Blocked_Rtl(params,out);
+            break;
+        case P_NEQ_B_OPT_MD:
+                partialNEQ_Blocked_STL_MD(params,out);
+            break;
+
+        default:
+            break;
+    }
+
+}
+
+void Algorithm::extract_subMatrix(type_precision* source, type_precision* dest,int dim1_source, int dim2_source,
+                                                            int dim1_ini,int dim1_end,int dim2_ini,int dim2_end)
+{
+
+    int i,j,idx=0;
+    int size, source_ini;
+    for(i = dim2_ini; i<dim2_end; i++)
+    {
+        j = dim1_ini;
+        source_ini = i*dim1_source+j;
+        size = dim1_end-dim1_ini;
+        memcpy( (type_precision*)&dest[idx], (type_precision*)&source[source_ini], size * sizeof(type_precision) );
+//        for(j = dim1_ini; j<dim1_end; j++)
+//        {
+//            dest[idx] = source[i*dim1_source+j];
+//            idx++;
+//        }
+        idx += size;
+    }
+
+}
+
+void Algorithm::prepare_Bfinal(type_precision* bfinal, type_precision* top,type_precision* bot,int dim1_b, int dim2_b,int dim1_b_bot)
+{
+    //memcpy are faster version of the fors
+    int i,k,w,top_idx,bot_idx,max = dim1_b*dim2_b;
+    int size;
+    top_idx = 0;
+    bot_idx = 0;
+    for(k = 0; k < dim2_b; k++)
+    {
+        size = k*dim1_b+(dim1_b-dim1_b_bot)-(k*dim1_b);
+        memcpy( (type_precision*)&bfinal[k*dim1_b], (type_precision*)&top[top_idx], size * sizeof(type_precision) );
+//        for(i = k*dim1_b; i < k*dim1_b+(dim1_b-dim1_b_bot); i++)
+//        {
+//            bfinal[i] = top[top_idx];
+//            top_idx++;
+//        }
+        top_idx += size;
+        i = k*dim1_b + size;
+        w=i;
+
+        size = w+dim1_b_bot - w;
+        memcpy( (type_precision*)&bfinal[w], (type_precision*)&bot[bot_idx], size * sizeof(type_precision) );
+//        for(i = w; i < w+dim1_b_bot; i++)
+//        {
+//            bfinal[i] = bot[bot_idx];
+//            bot_idx++;
+//        }
+        bot_idx += size;
+    }
+
+}
+
+void Algorithm::prepare_QY(type_precision* qy, type_precision* top,type_precision* bot,int dim1_QY, int dim2_QY,int dim1_qy_bot,int bot_blocks )
+{
+
+    int i,k,w,top_idx,bot_idx,max = dim1_QY*dim2_QY;
+    top_idx = 0;
+    bot_idx = 0;
+    for(k = 0; k < dim2_QY; k++)
+    {
+        for(i = k*dim1_QY; i < (k+1)*dim1_QY-dim1_qy_bot; i++)
+        {
+            qy[i] = top[top_idx];
+            top_idx++;
+        }
+        w=i;
+
+        for(i = w; i < w+dim1_qy_bot; i++)
+        {
+            qy[i] = bot[bot_idx];
+            bot_idx++;
+        }
+        bot_idx+=(bot_blocks-1)*dim1_qy_bot;
+    }
+
+}
+
+type_precision* Algorithm::extract_R(type_precision* A,int dim1_A, int dim2_A)
+{
+    type_precision* R = (type_precision*)calloc(dim2_A*dim2_A,sizeof(type_precision));
+    int i,j;
+
+    int R_idx=0;
+
+    for(i = 0; i < dim2_A; i++)
+    {
+        for(j = 0; j <= i; j++)
+        {
+            R[R_idx] = A[j+i*dim1_A];
+            R_idx++;
+        }
+        R_idx = dim2_A*(i+1);
+    }
+    return R;
+}
+
+type_precision* Algorithm::prepare_R(type_precision* RL,int dim1_A, int dim2_AL,int dim2_AR)
+{
+    int R_dim = (dim2_AR+dim2_AL);
+    type_precision* R = new type_precision[R_dim*R_dim];
+
+    int i,j;
+
+    int RL_idx=0;
+    int R_idx=0;
+
+    for(i = 0; i < dim2_AL; i++)
+    {
+        for(j = 0; j <= i; j++)
+        {
+            RL_idx = i*dim1_A + j;
+            R_idx = i*R_dim + j;
+            R[R_idx] = RL[RL_idx];
+
+        }
+
+    }
+    return R;
+}
+
+void Algorithm::update_R(type_precision* R, type_precision* topRr, type_precision* botRr,int dim1, int dim2, int r)
+{
+    int i,j,w;
+    int max = dim1*dim2;
+    int rtr_idx=0;
+    int rbr_idx=0;
+    for( j = r; j > 0 ; j--)
+    {
+        for(i = max-dim1*j; i < max-dim1*j+dim2-r; i++)
+        {
+            R[i] = topRr[rtr_idx];
+            rtr_idx++;
+        }
+        w = i;
+        for(i = w; i < w+r; i++)
+        {
+            R[i] = botRr[rbr_idx];
+            rbr_idx++;
+        }
+
+    }
+
+}
+
+
+void Algorithm::build_S(type_precision* S,type_precision* Stl,type_precision* Str,type_precision* Sbr,int l,int r)
+{
+    int Sidx;
+    int p = l+r;
+    for(int i = 0; i < l; i++)
+    {
+        Sidx = i*p;
+        for(int j= 0; j <= i; j++)
+        {
+            S[Sidx] = Stl[j+i*l];
+            Sidx++;
+        }
+    }
+    for(int i = 0; i < r; i++)
+    {
+        Sidx = l*p+p*i;
+        for(int j= 0; j < l; j++)
+        {
+            S[Sidx] = Str[j+i*l];
+            Sidx++;
+        }
+    }
+
+    for(int i = 0; i < r; i++)
+    {
+        Sidx = l*p+l+p*i;
+        for(int j= 0; j <= i; j++)
+        {
+            S[Sidx] = Sbr[j+i*r];
+            Sidx++;
+        }
+    }
+
+
+
+}
+
+void Algorithm::check_result(type_precision* AL, type_precision* AR,int rowsA,int colsA, int rhs,int colsAR,
+                                                    type_precision* y, type_precision* res)
+{
+    type_precision* A = (type_precision*)malloc(rowsA*colsA*sizeof(type_precision));
+
+    int i,ar_idx=0;
+    for(i = 0; i <rowsA*(colsA-colsAR) ; i++)
+    {
+        A[i] =  AL[i];
+    }
+
+    for(i = rowsA*(colsA-colsAR); i <rowsA*colsA ; i++)
+    {
+        A[i] =  AR[ar_idx];
+        ar_idx++;
+    }
+
+
+
+    type_precision* ynew = replicate_vec(y,rowsA*rhs);
+    type_precision* new_sol = (type_precision*)malloc(colsA*rhs*sizeof(type_precision));
+
+    lapack_int info = LAPACKE_sgels(STORAGE_TYPE,'N',rowsA,colsA,rhs,A,rowsA,ynew,rowsA);
+    assert(info == 0,"Error Check");
+
+
+
+
+
+    int index=0;
+    int index_new = 0;
+    for(i = 0; i < rhs; i++)
+    {
+         copy_vec( &ynew[index], &new_sol[index_new], colsA);
+         index += rowsA;
+         index_new += colsA;
+    }
+
+
+//    if(PRINT)
+//        printf("\n Btop=(Rtl\\(Ql'*Y))-(Rtl\\Rtr)*(Rbr\\(Qr'*Y)); \n [Btop ; Rbr\\Qr'*Y] - bcomputed \n");
+
+    cblas_saxpy(rhs*colsA, -1.0, res, 1,new_sol,1);
+    type_precision u_norm=cblas_snrm2(rhs*colsA,new_sol,1);
+    //
+    if(abs(u_norm) >= 0.0001 || isnan(u_norm))
+    {
+
+
+        fflush(stdout);
+        matlab_print_matrix("AL",rowsA,colsA-colsAR,AL);
+        matlab_print_matrix("AR",rowsA,colsAR,AR);
+        matlab_print_matrix("Y",rowsA,rhs,y);
+        printf("\nA = [AL AR]; [Q,R] = qr(A,0); rr=R\\(Q'*Y)\n");
+        matlab_print_matrix("bcomputed",colsA,rhs,res);
+        matlab_print_matrix("newsol",colsA,rhs,ynew);
+        printf("\n%%\tnrom: %0.2g", u_norm);
+         exit(1);
+    }
+    else
+    {
+        matlab_print_matrix("bcomputed",colsA,rhs,res);
+        matlab_print_matrix("newsol",colsA,rhs,ynew);
+        printf("\n%%\tnrom: %0.2g", u_norm);
+    }
+
+
+
+    //cout << "\t**************";
+
+
+    free(ynew);
+    free(new_sol);
+    free(A);
+
+
+
+
+
+
+}
+
+///////////////////////////////
+
+
+
+void Algorithm::partialNEQ_Blocked_STL_MD(struct Settings params, struct Outputs &out)
+{
+    int max_threads = params.threads;
+
+
+    srand (time(NULL));
+
+    blas_set_num_threads(max_threads);
+
+
+    type_precision *Ytemp;
+    lapack_int info,n,lda,ldy,l,r,k,p;
+
+    int i,j,w;
+    int m;
+
+    cputime_type start_tick,start_tick2, end_tick;
+
+    AIOfile.initialize(params);
+
+    n = params.n; l=params.l; r=params.r; p = l+r;
+
+    int y_amount = params.t;
+    int y_block_size = params.tb;//kk
+
+    int a_amount = params.m;
+    int a_block_size = params.mb;
+
+    int a_iters = (a_amount + a_block_size - 1) / a_block_size;
+
+    int y_iters = (y_amount + y_block_size - 1) / y_block_size;
+
+
+    lda = n; ldy = n;
+    k = p;
+
+    for(int j = 0; j < y_iters; j++)
+    {
+
+        if(y_iters >= 40 && (i%(y_iters/40))==0)
+        {
+            cout << "*" << flush;
+
+        }
[TRUNCATED]

To get the complete diff run:
    svnlook diff /svnroot/genabel -r 1469


More information about the Genabel-commits mailing list