diff --git a/examples/Makefile.am b/examples/Makefile.am
index d91f91d7fb3816dde5f6114c9d8e78677e60803d..2d20cbb1630044ced0bfd7b6d87d05e38696623b 100644
--- a/examples/Makefile.am
+++ b/examples/Makefile.am
@@ -20,13 +20,13 @@
AUTOMAKE_OPTIONS=gnu
# Add the source directory and debug to CFLAGS
-AM_CFLAGS = -g -Wall -Werror -I../src $(OPENMP_CFLAGS) -DCPU_TPS=2.67e9 -DTIMERS \
+AM_CFLAGS = -g -Wall -Werror -I../src $(OPENMP_CFLAGS) -DCPU_TPS=3.1e9 -DTIMERS \
# -fsanitize=address -fno-omit-frame-pointer
-AM_LDFLAGS = -lm # -fsanitize=address
+ AM_LDFLAGS = -lm # -fsanitize=address
# Set-up the library
-bin_PROGRAMS = test test_qr test_bh
+bin_PROGRAMS = test test_qr #test_bh
# Sources for test
test_SOURCES = test.c
@@ -35,11 +35,19 @@ test_LDADD = ../src/.libs/libquicksched.a
# Sources for test_qr
test_qr_SOURCES = test_qr.c
-test_qr_CFLAGS = $(AM_CFLAGS)
-test_qr_LDADD = ../src/.libs/libquicksched.a -llapacke -llapack -lblas
+test_qr_CFLAGS = $(AM_CFLAGS) -I/home/aidan/lapack-3.5.0/lapacke/include/
+test_qr_LDFLAGS = -I/home/aidan/lapack-3.5.0/lapacke/include/
+test_qr_LDADD = ../src/.libs/libquicksched.a /home/aidan/lapack-3.5.0/liblapacke.a /home/aidan/lapack-3.5.0/liblapack.a -lblas
# Sources for test_bh
-test_bh_SOURCES = test_bh.c
-test_bh_CFLAGS = $(AM_CFLAGS)
-test_bh_LDADD = ../src/.libs/libquicksched.a
+#test_bh_SOURCES = test_bh.c
+#test_bh_CFLAGS = $(AM_CFLAGS)
+#test_bh_LDADD = ../src/.libs/libquicksched.a
+
+#if HAVE_CUDA
+#test_cuda_SOURCES = test_gpu_simple.cu
+#test_cuda_CFLAGS = -DWITH_CUDA $(AM_CFLAGS) $(CUDA_CFLAGS) $(CUDA_MYFLAGS)
+#test_cuda_LINK = $(NVCC)
+#test_cuda_LDADD = ../src/.libs/libquicksched_cuda.a $(CUDA_LIBS)
+#endif
diff --git a/examples/test.c b/examples/test.c
index 9d181aa1d4b00ff2e5900d039ca9ab658d07f676..66dfae6af5e7128dcad3f96f952da6d6bea1af82 100644
--- a/examples/test.c
+++ b/examples/test.c
@@ -118,7 +118,7 @@ void test2 ( int m , int n , int k , int nr_threads ) {
/* Build a task for each tile of the matrix c. */
for ( i = 0 ; i < m ; i++ )
for ( j = 0 ; j < n ; j++ ) {
- rid = qsched_addres( &s , qsched_owner_none , qsched_res_none );
+ rid = qsched_addres( &s , qsched_owner_none , qsched_res_none, NULL , 0 );
data[0] = i; data[1] = j;
for ( kk = 0 ; kk < k ; kk++ ) {
data[2] = kk;
@@ -222,7 +222,7 @@ void test1 ( int m , int n , int k , int nr_threads ) {
for ( i = 0 ; i < m ; i++ )
for ( j = 0 ; j < n ; j++ ) {
data[0] = i; data[1] = j;
- rid = qsched_addres( &s , qsched_owner_none , qsched_res_none );
+ rid = qsched_addres( &s , qsched_owner_none , qsched_res_none , NULL , 0);
tid = qsched_addtask( &s , 1 , task_flag_none , data , 2*sizeof(int) , 1 );
qsched_addlock( &s , tid , rid );
}
diff --git a/examples/test_qr.c b/examples/test_qr.c
index 2a754c04d4976e138170aaaf261bfcd8afe77ba5..6f77203afbba2c5b3e1c0a953eb9b89323609187 100644
--- a/examples/test_qr.c
+++ b/examples/test_qr.c
@@ -37,7 +37,6 @@
/* Local includes. */
#include "quicksched.h"
-
/*
* Sam's routines for the tiled QR decomposition.
*/
@@ -47,9 +46,9 @@
Computes the 2-norm by computing the following: \f[\textrm{2-norm}=\sqrt_0^lx(i)^2\f]
*/
-double do2norm(double* x, int l)
+float do2norm(float* x, int l)
{
- double sum = 0, norm;
+ float sum = 0, norm;
int i;
for(i = 0; i < l; i++)
@@ -74,14 +73,14 @@ double do2norm(double* x, int l)
*
* \returns void
*/
-void calcvkDouble (double topDiag,
+void calcvkfloat (float topDiag,
int ma,
- double* xb,
+ float* xb,
int l,
- double* vk)
+ float* vk)
{
int sign, i;
- double norm, div;
+ float norm, div;
//same non-standard normalisation as for single blocks above, but organised without a temporary beta veriable
sign = topDiag >= 0.0 ? 1 : -1;
@@ -106,16 +105,16 @@ void calcvkDouble (double topDiag,
}
-void updateDoubleQ_WY (double* blockA,
- double* blockB,
- double* blockTau,
+void updatefloatQ_WY (float* blockA,
+ float* blockB,
+ float* blockTau,
int k, int ma, int mb, int n,
int ldm,
- double* hhVector)//bottom, essential part.
+ float* hhVector)//bottom, essential part.
{
int i, j;
- double tau = 1.0, beta;
+ float tau = 1.0, beta;
/* Compute tau = 2/v'v */
for(i = 0; i < mb; i ++)
@@ -148,17 +147,17 @@ void updateDoubleQ_WY (double* blockA,
blockTau[k] = tau;
}
-void DTSQRF (double* blockA,
- double* blockB,
- double* blockTau,
+void DTSQRF (float* blockA,
+ float* blockB,
+ float* blockTau,
int ma,
int mb,
int n,
int ldm,
- double* hhVector)
+ float* hhVector)
{
int k;
- double* xVectA, *xVectB;
+ float* xVectA, *xVectB;
xVectA = blockA;
xVectB = blockB;
@@ -167,11 +166,11 @@ void DTSQRF (double* blockA,
{
//vk = sign(x[1])||x||_2e1 + x
//vk = vk/vk[0]
- calcvkDouble(xVectA[0], ma - k, xVectB, (ma + mb) - k, hhVector);//returns essential
+ calcvkfloat(xVectA[0], ma - k, xVectB, (ma + mb) - k, hhVector);//returns essential
//matA(k:ma,k:na) = matA(k:ma,k:na) - (2/(vk.T*vk))*vk*(vk.T*matA(k:ma,k:na)
//update both blocks, preserving the vectors already stored below the diagonal in the top block and treating them as if they were zeros.
- updateDoubleQ_WY (blockA, blockB,
+ updatefloatQ_WY (blockA, blockB,
blockTau,
k, ma, mb, n,
ldm,
@@ -182,14 +181,14 @@ void DTSQRF (double* blockA,
}
}
-void DSSRFT (double* blockV,
- double* blockA, double* blockB,
- double* blockTau,
+void DSSRFT (float* blockV,
+ float* blockA, float* blockB,
+ float* blockTau,
int b, int n, int ldm)
{
int i, j, k;
- double tau, beta;
+ float tau, beta;
/* Compute b_j = b_j - tau*v*v'*b_j for each column j of blocks A & B,
and for each householder vector v of blockV */
@@ -223,7 +222,89 @@ void DSSRFT (double* blockV,
}
}
}
+
+float* randomMatrix(int m, int n)
+{
+ float* Matrix;
+ Matrix = (float*) malloc( sizeof(float) * m*n*32*32);
+ if(Matrix == NULL)
+ error("Failed to allocate Matrix");
+ int r,c;
+ m = m*32;
+ n = n*32;
+for(c = 0; c < n; c++)
+ {
+ for(r = 0; r < m; r++)
+ {
+ //CO(i,j,m) ((m * j) + i)
+ Matrix[(m*c)+r] = ((float)(rand() % 201) - 100.0) / 100.0;
+ }
+ }
+return Matrix;
+
+}
+float* generateMatrix( int m, int n)
+{
+ float* Matrix;
+ Matrix = (float*) malloc( sizeof(float) * m*n*32*32);
+ if(Matrix == NULL)
+ error("Failed to allocate Matrix");
+ int i, j;
+ memset ( Matrix, 0, sizeof(float)*m*n*32*32 );
+
+ for(i = 0; i < n*32; i++)
+ {
+ for(j = 0; j < m*32; j++)
+ {
+ Matrix[i*m*32 + j] = (float)(i+j);
+ }
+ }
+ return Matrix;
+}
+
+float* createIdentity(int m, int n)
+{
+ float* Matrix;
+ Matrix = (float*) malloc( sizeof(float) * m*n*32*32);
+ if(Matrix == NULL)
+ error("Failed to allocate Matrix");
+ int i, j;
+ memset ( Matrix, 0, sizeof(float)*m*n*32*32 );
+
+ for(i = 0; i < n*32; i++)
+ {
+ for(j = 0; j < m*32; j++)
+ {
+ if(i==j)
+ {
+ Matrix[i*m*32 + j] = 1.0;
+ }else
+ {
+ Matrix[i*m*32 + j] = 0.0;
+ }
+ }
+ }
+ return Matrix;
+}
+
+void printMatrix(float* Matrix, int m, int n)
+{
+ int i, j;
+
+ for(i=0; i < m*32; i++)
+ {
+ printf("{ ");
+ for(j = 0; j < n*32; j++)
+ {
+ printf(" %.3f ", Matrix[j*m*32 +i]);
+
+ }
+ printf(" }");
+ printf("\n");
+ }
+
+}
/**
* @brief Computed a tiled QR factorization using QuickSched.
@@ -236,7 +317,7 @@ void DSSRFT (double* blockV,
void test_qr ( int m , int n , int K , int nr_threads , int runs ) {
int k, j, i;
- double *A, *A_orig, *tau;
+ float *A, *A_orig, *tau;
struct qsched s;
qsched_task_t *tid, tid_new;
qsched_res_t *rid;
@@ -251,27 +332,28 @@ void test_qr ( int m , int n , int K , int nr_threads , int runs ) {
/* Decode the task data. */
int *idata = (int *)data;
- int i = idata[0], j = idata[1], k = idata[2];
- double buff[ 2*K*K ];
+ int i = idata[0], j = idata[1];//, k = idata[2];
+ float buff[ 2*K*K ];
/* Decode and execute the task. */
switch ( type ) {
case task_DGEQRF:
- LAPACKE_dgeqrf_work( LAPACK_COL_MAJOR , K, K ,
+ LAPACKE_sgeqrf_work( LAPACK_COL_MAJOR , K, K ,
&A[ j*m*K*K + i*K ] , m*K , &tau[ j*m*K + i*K ] ,
buff , 2*K*K );
+
break;
case task_DLARFT:
- LAPACKE_dlarft_work( LAPACK_COL_MAJOR , 'F' , 'C' ,
+ LAPACKE_slarft_work( LAPACK_COL_MAJOR , 'F' , 'C' ,
K , K , &A[ i*m*K*K + i*K ] ,
m*K , &tau[ i*m*K + i*K ] , &A[ j*m*K*K + i*K ] ,
m*K );
break;
case task_DTSQRF:
- DTSQRF( &A[ j*m*K*K + j*K ] , &A[ j*m*K*K + i*K ] , &tau[ j*m*K + i*K ] , K , K , K , K*m , buff );
+ //DTSQRF( &A[ j*m*K*K + j*K ] , &A[ j*m*K*K + i*K ] , &tau[ j*m*K + i*K ] , K , K , K , K*m , buff );
break;
case task_DSSRFT:
- DSSRFT( &A[ k*m*K + i*K ] , &A[ j*m*K*K + k*K ] , &A[ j*m*K*K + i*K ] , &tau[ k*m*K + i*K ] , K , K , K*m );
+ //DSSRFT( &A[ k*m*K + i*K ] , &A[ j*m*K*K + k*K ] , &A[ j*m*K*K + i*K ] , &tau[ k*m*K + i*K ] , K , K , K*m );
break;
default:
error( "Unknown task type." );
@@ -281,15 +363,28 @@ void test_qr ( int m , int n , int K , int nr_threads , int runs ) {
/* Allocate and fill the original matrix. */
- if ( ( A = (double *)malloc( sizeof(double) * m * n * K * K ) ) == NULL ||
- ( tau = (double *)malloc( sizeof(double) * m * n * K ) ) == NULL ||
- ( A_orig = (double *)malloc( sizeof(double) * m * n * K * K ) ) == NULL )
+ if ( ( A = (float *)malloc( sizeof(float) * m * n * K * K ) ) == NULL ||
+ ( tau = (float *)malloc( sizeof(float) * m * n * K ) ) == NULL ||
+ ( A_orig = (float *)malloc( sizeof(float) * m * n * K * K ) ) == NULL )
error( "Failed to allocate matrices." );
- for ( k = 0 ; k < m * n * K * K ; k++ )
- A_orig[k] = 2*((double)rand()) / RAND_MAX - 1.0;
- memcpy( A , A_orig , sizeof(double) * m * n * K * K );
- bzero( tau , sizeof(double) * m * n * K );
+ free(A_orig);
+// for ( k = 0 ; k < m * n * K * K ; k++ )
+ // A_orig[k] = 2*((float)rand()) / RAND_MAX - 1.0;
+// A_orig = generateMatrix(m, n);
+ srand(5);
+ A_orig = randomMatrix(m,n);
+ printMatrix(A_orig, m, n);
+ printf("\n\n\n");
+ memcpy( A , A_orig , sizeof(float) * m * n * K * K );
+ bzero( tau , sizeof(float) * m * n * K );
+
+ LAPACKE_sgeqrf( LAPACK_COL_MAJOR , K*m, K*n ,
+ A, m*K , tau);
+ printMatrix(A, m , n );
+ printf("\n\n\n");
+ memcpy( A , A_orig , sizeof(float) * m * n * K * K );
+ bzero( tau , sizeof(float) * m * n * K );
/* Dump A_orig. */
/* message( "A_orig = [" );
for ( k = 0 ; k < m*K ; k++ ) {
@@ -309,7 +404,7 @@ void test_qr ( int m , int n , int K , int nr_threads , int runs ) {
error( "Failed to allocate tid/rid matrix." );
for ( k = 0 ; k < m * n ; k++ ) {
tid[k] = qsched_task_none;
- rid[k] = qsched_addres( &s , qsched_owner_none , qsched_res_none );
+ rid[k] = qsched_addres( &s , qsched_owner_none , qsched_res_none , NULL, 0 );
}
/* Build the tasks. */
@@ -386,7 +481,13 @@ void test_qr ( int m , int n , int K , int nr_threads , int runs ) {
tot_run += toc_run - tic;
}
-
+ printf("tau = ");
+ for(k = 0; k < m * n * K ; k++)
+ {
+ printf("%.3f ", tau[k]);
+ }
+ printf("\n");
+ printMatrix(A, m, n);
/* Dump A. */
/* message( "A = [" );