changes cblas version

cmake/.gitignore

 build/
 debug/
 nohup.out
+*.csv

cmake/CMakeLists.txt

@@ -27,7 +27,7 @@ add_definitions(-DNUM_THREADS=64)
 if ("${CMAKE_CXX_COMPILER_ID}" STREQUAL "GNU")
   set(CMAKE_CXX_FLAGS_DEBUG "-std=c++11 -march=knl -mtune=knl -fpic -ffast-math -O0 -fopenmp")
 elseif ("${CMAKE_CXX_COMPILER_ID}" STREQUAL "Intel")
-  set(CMAKE_CXX_FLAGS_DEBUG  "-std=c++11 -fpic -qopenmp -axCOMMON-AVX512 -lmemkind -lmkl_core -lmkl_intel_lp64 -lmkl_intel_thread -liomp5 -lpthread -g -debug all -save-temps -Wl,-no_pie -O0")
+  set(CMAKE_CXX_FLAGS_DEBUG  "-std=c++11 -fpic -qopenmp -axCOMMON-AVX512 -lmemkind -lmkl_core -lmkl_intel_lp64 -lmkl_intel_thread -liomp5 -lpthread -g -debug all -save-temps -Wl, -O0 -fstack-security-check")
 endif()
 
 # adding MKL include directory

cmake/main.cpp

@@ -16,9 +16,10 @@ int main(int argc, char** argv) {
 
     size_t threads = 64;
     size_t N=200, D=20;
+    size_t blocksize=4000;
     double EPS=0.2;
 
-    parsing_args(argc,argv, &N, &EPS, &D, &threads, filename, isBinary);
+    parsing_args(argc,argv, &N, &EPS, &D, &threads, &blocksize, filename, isBinary);
 
     N = N * 1000;
 
@@ -34,9 +35,14 @@ int main(int argc, char** argv) {
         read_file(x, N, D, filename, isBinary);
     }
 
-    elapsed = blasJoin( x, N, D, EPS, threads);
+    elapsed = blasJoin( x, N, D, EPS, threads, blocksize);
 
-    printf("%d; %d; %f; %f\n", N,D,EPS,elapsed);
+#pragma omp parallel
+{
+    if (  omp_get_thread_num() == 0 ){
+        printf("%ld; %ld; %f; %ld; %d; %f \n", N, D, EPS, blocksize, omp_get_num_threads(),elapsed);
+    }
+}
 
     ddr_free(x);
 

cmake/testBlockSize.sh

+# BLOCKSIZE=(1000 2000 3000 4000 5000 6000 7000 8000 9000 10000 11000 12000 13000 14000 15000 18000 20000 )
+BLOCKSIZE=(100 200 300 400 500 600 700 800 900 )
+# THREADS=(96 128 160 192 224 256)
+FILENAME=$1
+REPETITIONS=5
+
+if [ $# -ne 1 ]
+then
+  echo "need filename as parameter!"
+  exit 1
+fi
+
+echo "writing to filename: $1"
+echo "N;D;EPS;BLOCKSIZE;THREADS;TIME;LOOP" > $1
+
+for b in ${BLOCKSIZE[@]}
+do
+  echo "$b"
+  export OMP_NUM_THREADS=64
+  for i in $(seq 1 $REPETITIONS)
+  do
+    ./build/blasJoin -n 200 -e 0.2 -d 20 -s $b >>$1
+#     numactl --membind=1,2,3,4,5,6,7,8
+  done
+done
+
+mail -s "blasJoin done!" -r "root@ivanhoe.dm.univie.ac.at" "martin.perdacher@univie.ac.at"

cmake/util/allocation.cpp

@@ -5,7 +5,7 @@ void * ddr_alloc(size_t bytes){
     void * ptr = _mm_malloc(bytes, ALIGNMENT);
 
     if ( ptr == NULL ){
-        fprintf(stderr, "Error in allocating memory with ddr_alloc!");
+        fprintf(stderr, "Error in allocating memory with ddr_alloc!\n");
         exit(1);
     }
 

octave/joinDot.m

-1; # octave script file

N = 5;
D = 2;
EPS = 0.02;
X = rand(N, D);

Pi = zeros(N,1);

qEPS = EPS^2 / 4;
# pre-compute the values Pi:
for i = 1:rows(X)
  # BLAS level 1 dsdot:
  Pi(i) = qEPS - ( dot( X(1,:), X(1,:) ) / 2);  
endfor

# BLAS level 3 dgemm:
scalarp = X * X';

# loop over lower triangle of scalarp
for i = 2:rows(scalarp)
  for j = 1:(i-1)
    # join partner i and j
    scalarp(i,j) += Pi(i) + Pi(j);
  endfor
endfor

# if scalarp(i,j) greater than 0, then we have a join partner
scalarp
\ No newline at end of file
+1; # octave script file

N = 5;
D = 2;
EPS = 0.02;
X = rand(N, D);

P = zeros(N,1);

qEPS = EPS^2 / 4;
# pre-compute the values Pi:
for i = 1:rows(X)
  # BLAS level 1 ddot:
  P(i) = qEPS - ( dot( X(i,:), X(i,:) ) / 2);
endfor

# BLAS level 3 dgemm:
scalarp = X * X';

# loop over lower triangle of scalarp
for i = 2:rows(scalarp)
  for j = 1:(i-1)
    # join partner i and j
    scalarp(i,j) += P(i) + P(j);
  endfor
endfor

# if scalarp(i,j) greater than 0, then we have a join partner
scalarp
\ No newline at end of file