STEP 5.3: evaluate performance of blocked variant

CMakeLists.txt

@@ -18,6 +18,12 @@ target_link_libraries(multiply matrix)
 add_executable(print src/app/print.cpp)
 target_link_libraries(print matrix)
 
+add_executable(bench_simple src/bench/simple.cpp)
+target_link_libraries(bench_simple matrix)
+
+add_executable(bench_blocked src/bench/blocked.cpp)
+target_link_libraries(bench_blocked matrix)
+
 enable_testing()
 add_executable(testBlocked tests/test.cpp)
 target_link_libraries(testBlocked matrix)

run.blocked.sh

+#!/bin/bash
+#PBS -S /bin/bash
+#PBS -l select=1:ncpus=128:ompthreads=128
+#PBS -q qexp
+#PBS -N blocked
+
+ml icc
+cd $PBS_O_WORKDIR
+
+for BLOCK in {2..6}
+do
+  echo BLOCK=$((2**BLOCK))
+  for AFFINITY in "none" "compact" "scatter" "balanced"
+  do
+    echo ""
+    echo KMP_AFFINITY=$AFFINITY
+    for THREADS in {0..7}
+    do
+      echo OMP_NUM_THREADS=$((2**$THREADS))
+      KMP_AFFINITY=$AFFINITY OMP_NUM_THREADS=$((2**$THREADS)) ./build/bench_blocked tests/large/A.bin tests/large/B.bin C.bin $((2**BLOCK))
+    done
+  done
+done

run.simple.sh

+#!/bin/bash
+#PBS -S /bin/bash
+#PBS -l select=1:ncpus=128:ompthreads=128
+#PBS -q qexp
+#PBS -N simple
+
+ml icc
+cd $PBS_O_WORKDIR
+
+for AFFINITY in "none" "compact" "scatter" "balanced"
+do
+  echo ""
+  echo KMP_AFFINITY=$AFFINITY
+  for THREADS in {0..7}
+  do
+    echo OMP_NUM_THREADS=$((2**$THREADS))
+    KMP_AFFINITY=$AFFINITY OMP_NUM_THREADS=$((2**$THREADS)) ./build/bench_simple tests/large/A.bin tests/large/B.bin C.bin
+  done
+done
+
+

src/bench/blocked.cpp

+
+#include "matrix.h"
+
+#include <omp.h>
+#include <iostream>
+#include <cstdlib>
+
+int main(int argc, char **argv)
+{
+	if (argc != 5) {
+		std::cerr << "use: sequential 'A' 'B' 'output_file' 'block_size'\n";
+		exit(1);
+	}
+
+	double sload = omp_get_wtime();
+	Matrix C, A, B;
+	load(A, argv[1]);
+	load(B, argv[2]);
+	double eload = omp_get_wtime();
+	printf("Matrices loaded: A[%d, %d], B[%d, %d]\n", A.rows, A.cols, B.rows, B.cols);
+
+	initMult(C, A, B);
+	double smult = omp_get_wtime();
+//	mult(C, A, B);
+	multBlocked(C, A, B, std::atoi(argv[4]));
+	double emult = omp_get_wtime();
+	printf("Matrices multiplied: C[%d, %d]\n", C.rows, C.cols);
+
+	double sstore = omp_get_wtime();
+	store(C, argv[3]);
+	double estore = omp_get_wtime();
+
+	printf("Load, Mult, Store; %e; %e; %e;\n", eload - sload, emult - smult, estore - sstore);
+}

src/bench/simple.cpp

+
+#include "matrix.h"
+
+#include <omp.h>
+#include <iostream>
+
+int main(int argc, char **argv)
+{
+	if (argc != 4) {
+		std::cerr << "use: sequential 'A' 'B' 'output_file'\n";
+		exit(1);
+	}
+
+	double sload = omp_get_wtime();
+	Matrix C, A, B;
+	load(A, argv[1]);
+	load(B, argv[2]);
+	double eload = omp_get_wtime();
+	printf("Matrices loaded: A[%d, %d], B[%d, %d]\n", A.rows, A.cols, B.rows, B.cols);
+
+	initMult(C, A, B);
+	double smult = omp_get_wtime();
+	mult(C, A, B);
+//	multBlocked(C, A, B, 8);
+	double emult = omp_get_wtime();
+	printf("Matrices multiplied: C[%d, %d]\n", C.rows, C.cols);
+
+	double sstore = omp_get_wtime();
+	store(C, argv[3]);
+	double estore = omp_get_wtime();
+
+	printf("Load, Mult, Store; %e; %e; %e;\n", eload - sload, emult - smult, estore - sstore);
+}