Merge pull request #143 from jprotze/simd

micro-benchmarks/DRB202-simd-broadcast-yes.c

@@ -0,0 +1,35 @@
+/*
+!!!~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~!!!
+!!! Copyright (c) 2017-20, Lawrence Livermore National Security, LLC
+!!! and DataRaceBench project contributors. See the DataRaceBench/COPYRIGHT file for details.
+!!!
+!!! SPDX-License-Identifier: (BSD-3-Clause)
+!!!~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~!!!
+*/
+
+/*
+Data race in vectorizable code
+Adding a fixed array element to the whole array. Data race present.
+Data Race Pairs, a[i]@30:5:W vs. a[64]@30:19:R
+*/
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <math.h>
+int main(int argc, char *argv[]) {
+  int len = 20000;
+  if (argc > 1)
+    len = atoi(argv[1]);
+  double a[len];
+  for (int i = 0; i < len; i++)
+    a[i] = i;
+  double c = M_PI;
+
+#pragma omp parallel for simd schedule(dynamic, 64)
+  for (int i = 0; i < len; i++)
+    a[i] = a[i] + a[64];
+
+  printf("a[0]=%f, a[%i]=%f, a[%i]=%f\n", a[0], len / 2, a[len / 2], len - 1,
+         a[len - 1]);
+  return 0;
+}

micro-benchmarks/DRB203-simd-broadcast-no.c

@@ -0,0 +1,33 @@
+/*
+!!!~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~!!!
+!!! Copyright (c) 17-20, Lawrence Livermore National Security, LLC
+!!! and DataRaceBench project contributors. See the DataRaceBench/COPYRIGHT file for details.
+!!!
+!!! SPDX-License-Identifier: (BSD-3-Clause)
+!!!~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~!!!
+*/
+
+/*
+No data race present.
+*/
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <math.h>
+int main(int argc, char *argv[]) {
+  int len = 20000;
+  if (argc > 1)
+    len = atoi(argv[1]);
+  double a[len];
+  for (int i = 0; i < len; i++)
+    a[i] = i;
+  double c = M_PI;
+
+#pragma omp parallel for simd schedule(dynamic, 64)
+  for (int i = 0; i < len; i++)
+    a[i] = a[i] + c;
+
+  printf("a[0]=%f, a[%i]=%f, a[%i]=%f\n", a[0], len / 2, a[len / 2], len - 1,
+         a[len - 1]);
+  return 0;
+}

micro-benchmarks/DRB204-simd-gather-yes.c

@@ -0,0 +1,39 @@
+/*
+!!!~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~!!!
+!!! Copyright (c) 2017-20, Lawrence Livermore National Security, LLC
+!!! and DataRaceBench project contributors. See the DataRaceBench/COPYRIGHT file for details.
+!!!
+!!! SPDX-License-Identifier: (BSD-3-Clause)
+!!!~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~!!!
+*/
+
+/*
+Data race in vectorizable code
+Loop depencency with 64 element offset. Data race present.
+Data Race Pairs, a[i + 64]@33:5:W vs. a[i * 2]@33:17:R
+*/
+
+#include <stdio.h>
+#include <stdlib.h>
+
+int main(int argc, char *argv[]) {
+  int len = 20000;
+
+  if (argc > 1)
+    len = atoi(argv[1]);
+  double a[2 * len], b[len];
+
+  for (int i = 0; i < 2 * len; i++)
+    a[i] = i;
+  for (int i = 0; i < len; i++)
+    b[i] = i + 1;
+
+#pragma omp parallel for simd schedule(dynamic, 64)
+  for (int i = 0; i < len; i++)
+    a[i + 64] = a[i * 2] + b[i];
+
+  printf("a[0]=%f, a[%i]=%f, a[%i]=%f\n", a[0], len / 2, a[len / 2], len - 1,
+         a[len - 1]);
+
+  return 0;
+}

micro-benchmarks/DRB205-simd-gatherscatter-no.c

@@ -0,0 +1,37 @@
+/*
+!!!~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~!!!
+!!! Copyright (c) 17-20, Lawrence Livermore National Security, LLC
+!!! and DataRaceBench project contributors. See the DataRaceBench/COPYRIGHT file for details.
+!!!
+!!! SPDX-License-Identifier: (BSD-3-Clause)
+!!!~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~!!!
+*/
+
+/*
+No data race present.
+*/
+
+#include <stdio.h>
+#include <stdlib.h>
+
+int main(int argc, char *argv[]) {
+  int len = 20000;
+
+  if (argc > 1)
+    len = atoi(argv[1]);
+  double a[2 * len], b[len];
+
+  for (int i = 0; i < 2 * len; i++)
+    a[i] = i;
+  for (int i = 0; i < len; i++)
+    b[i] = i + 1;
+
+#pragma omp parallel for simd schedule(dynamic, 64)
+  for (int i = 0; i < len; i++)
+    a[i * 2] = a[i * 2] + b[i];
+
+  printf("a[0]=%f, a[%i]=%f, a[%i]=%f\n", a[0], len / 2, a[len / 2], len - 1,
+         a[len - 1]);
+
+  return 0;
+}

micro-benchmarks/DRB206-simd-scatter-yes.c

@@ -0,0 +1,39 @@
+/*
+!!!~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~!!!
+!!! Copyright (c) 2017-20, Lawrence Livermore National Security, LLC
+!!! and DataRaceBench project contributors. See the DataRaceBench/COPYRIGHT file for details.
+!!!
+!!! SPDX-License-Identifier: (BSD-3-Clause)
+!!!~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~!!!
+*/
+
+/*
+Data race in vectorizable code
+Loop depencency with 64 element offset. Data race present.
+Data Race Pairs, a[i * 2]@33:5:W vs. a[i + 64]@33:17:R
+*/
+
+#include <stdio.h>
+#include <stdlib.h>
+
+int main(int argc, char *argv[]) {
+  int len = 20000;
+
+  if (argc > 1)
+    len = atoi(argv[1]);
+  double a[2 * len], b[len];
+
+  for (int i = 0; i < 2 * len; i++)
+    a[i] = i;
+  for (int i = 0; i < len; i++)
+    b[i] = i + 1;
+
+#pragma omp parallel for simd schedule(dynamic, 64)
+  for (int i = 0; i < len; i++)
+    a[i * 2] = a[i + 64] + b[i];
+
+  printf("a[0]=%f, a[%i]=%f, a[%i]=%f\n", a[0], len / 2, a[len / 2], len - 1,
+         a[len - 1]);
+
+  return 0;
+}

micro-benchmarks/DRB207-simd-loadstore-yes.c

@@ -0,0 +1,39 @@
+/*
+!!!~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~!!!
+!!! Copyright (c) 2017-20, Lawrence Livermore National Security, LLC
+!!! and DataRaceBench project contributors. See the DataRaceBench/COPYRIGHT file for details.
+!!!
+!!! SPDX-License-Identifier: (BSD-3-Clause)
+!!!~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~!!!
+*/
+
+/*
+Data race in vectorizable code
+Loop depencency with 64 element offset. Data race present.
+Data Race Pairs, a[i + 64]@34:5:W vs. a[i]@34:17:R
+*/
+
+#include <stdio.h>
+#include <stdlib.h>
+
+int main(int argc, char *argv[]) {
+  int len = 20000;
+
+  if (argc > 1)
+    len = atoi(argv[1]);
+  double a[len], b[len];
+
+  for (int i = 0; i < len; i++) {
+    a[i] = i;
+    b[i] = i + 1;
+  }
+
+#pragma omp parallel for simd schedule(dynamic, 64)
+  for (int i = 0; i < len - 64; i++)
+    a[i + 64] = a[i] + b[i];
+
+  printf("a[0]=%f, a[%i]=%f, a[%i]=%f\n", a[0], len / 2, a[len / 2], len - 1,
+         a[len - 1]);
+
+  return 0;
+}

micro-benchmarks/DRB208-simd-loadstore-no.c

@@ -0,0 +1,37 @@
+/*
+!!!~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~!!!
+!!! Copyright (c) 17-20, Lawrence Livermore National Security, LLC
+!!! and DataRaceBench project contributors. See the DataRaceBench/COPYRIGHT file for details.
+!!!
+!!! SPDX-License-Identifier: (BSD-3-Clause)
+!!!~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~!!!
+*/
+
+/*
+No data race present.
+*/
+
+#include <stdio.h>
+#include <stdlib.h>
+
+int main(int argc, char *argv[]) {
+  int len = 20000;
+
+  if (argc > 1)
+    len = atoi(argv[1]);
+  double a[len], b[len];
+
+  for (int i = 0; i < len; i++) {
+    a[i] = i;
+    b[i] = i + 1;
+  }
+
+#pragma omp parallel for simd schedule(dynamic, 64)
+  for (int i = 0; i < len; i++)
+    a[i] = a[i] + b[i];
+
+  printf("a[0]=%f, a[%i]=%f, a[%i]=%f\n", a[0], len / 2, a[len / 2], len - 1,
+         a[len - 1]);
+
+  return 0;
+}

scripts/test-harness.sh

@@ -57,14 +57,16 @@ LANGUAGE="default"
 PYTHON=${PYTHON:-"python3"}
 LOGPARSER="scripts/log-parser/logParser.py"
 
+OPTIMIZATION=${OPTIMIZATION:-"-O3 -march=native"}
+
 MEMCHECK=${MEMCHECK:-"/usr/bin/time"}
 TIMEOUTCMD=${TIMEOUTCMD:-"timeout"}
 VALGRIND=${VALGRIND:-"valgrind"}
-VALGRIND_COMPILE_C_FLAGS="-O0 -g -std=c99 -fopenmp"
-VALGRIND_COMPILE_CPP_FLAGS="-O0 -g -fopenmp"
+VALGRIND_COMPILE_C_FLAGS="${OPTIMIZATION} -g -std=c99 -fopenmp"
+VALGRIND_COMPILE_CPP_FLAGS="${OPTIMIZATION} -g -fopenmp"
 
 CLANG=${CLANG:-"clang"}
-TSAN_COMPILE_FLAGS="-O0 -fopenmp -fsanitize=thread -g"
+TSAN_COMPILE_FLAGS="${OPTIMIZATION} -fopenmp -fsanitize=thread -g"
 
 # Path to LLOV is fixed due to it's only available in container
 LLOV_COMPILER="/home/llvm/Work/LLOV"
@@ -77,24 +79,24 @@ LLOV_COMPILE_FLAGS+=" -mllvm -polly-ignore-parameter-bounds"
 LLOV_COMPILE_FLAGS+=" -mllvm -polly-dependences-on-demand"
 LLOV_COMPILE_FLAGS+=" -mllvm -polly-precise-fold-accesses"
 #LLOV_COMPILE_FLAGS+=" -mllvm -openmp-verify-disable-aa"
-LLOV_COMPILE_FLAGS+=" -O0 -g"
+LLOV_COMPILE_FLAGS+=" ${OPTIMIZATION} -g"
 
 # Path to LLOV is fixed due to it's only available in container
 FLANG_PATH=/home/llvm/installs/flang-2020-03-16
 
 ARCHER=${ARCHER:-"clang-archer"}
-ARCHER_COMPILE_FLAGS="-O0 -larcher"
+ARCHER_COMPILE_FLAGS="${OPTIMIZATION} -larcher"
 
 INSPECTOR=${INSPECTOR:-"inspxe-cl"}
-ICC_COMPILE_FLAGS="-O0 -fopenmp -std=c99 -qopenmp-offload=host -g"
-ICPC_COMPILE_FLAGS="-O0 -fopenmp -qopenmp-offload=host -g"
+ICC_COMPILE_FLAGS="${OPTIMIZATION} -fopenmp -std=c99 -qopenmp-offload=host -g"
+ICPC_COMPILE_FLAGS="${OPTIMIZATION} -fopenmp -qopenmp-offload=host -g"
 
-ROMP_CPP_COMPILE_FLAGS="-O0 -g -std=c++11 -fopenmp -lomp"
-ROMP_C_COMPILE_FLAGS="-O0 -g -fopenmp -lomp"
+ROMP_CPP_COMPILE_FLAGS="${OPTIMIZATION} -g -std=c++11 -fopenmp -lomp"
+ROMP_C_COMPILE_FLAGS="${OPTIMIZATION} -g -fopenmp -lomp"
 
 FORTRAN_LINK_FLAGS="-ffree-line-length-none -fopenmp -c -fsanitize=thread"
-FORTRAN_COMPILE_FLAGS="-O0 -fopenmp -fsanitize=thread -lgfortran"
-IFORT_FORTRAN_FLAGS="-g -O0 -free -qopenmp -qopenmp-offload=host -Tf"
+FORTRAN_COMPILE_FLAGS="${OPTIMIZATION} -fopenmp -fsanitize=thread -lgfortran"
+IFORT_FORTRAN_FLAGS="-g ${OPTIMIZATION} -free -qopenmp -qopenmp-offload=host -Tf"
 
 
 POLYFLAG="micro-benchmarks/utilities/polybench.c -I micro-benchmarks -I micro-benchmarks/utilities -DPOLYBENCH_NO_FLUSH_CACHE -DPOLYBENCH_TIME -D_POSIX_C_SOURCE=200112L"
@@ -359,7 +361,7 @@ for tool in "${TOOLS[@]}"; do
         intel)      icpc $ICPC_COMPILE_FLAGS $additional_compile_flags $test -o $exname -lm ;;
         helgrind)   g++ $VALGRIND_COMPILE_CPP_FLAGS $additional_compile_flags $test -o $exname -lm ;;
         archer)     clang-archer++ $ARCHER_COMPILE_FLAGS $additional_compile_flags $test -o $exname -lm ;;
-        coderrect)  coderrect -XbcOnly clang++ -fopenmp -fopenmp-version=45 -g -O0 $additional_compile_flags $test -o $exname -lm > /dev/null 2>&1 ;;
+        coderrect)  coderrect -XbcOnly clang++ -fopenmp -fopenmp-version=45 -g ${OPTIMIZATION} $additional_compile_flags $test -o $exname -lm > /dev/null 2>&1 ;;
         openrace)   clang++ -fopenmp -fopenmp-version=45 -Dmasked=master -g -S -emit-llvm $test -o $exname ;;
         tsan-clang) clang++ $TSAN_COMPILE_FLAGS $additional_compile_flags $test -o $exname -lm ;;
         tsan-gcc)   g++ $TSAN_COMPILE_FLAGS $additional_compile_flags $test -o $exname -lm ;;
@@ -376,7 +378,7 @@ for tool in "${TOOLS[@]}"; do
         intel)      icc $ICC_COMPILE_FLAGS $additional_compile_flags $test -o $exname -lm ;;
         helgrind)   gcc $VALGRIND_COMPILE_C_FLAGS $additional_compile_flags $test -o $exname -lm ;;
         archer)     clang-archer $ARCHER_COMPILE_FLAGS $additional_compile_flags $test -o $exname -lm ;;
-        coderrect)  coderrect -XbcOnly clang -fopenmp -fopenmp-version=45 -g -O0 $additional_compile_flags $test -o $exname -lm  > /dev/null 2>&1 ;;
+        coderrect)  coderrect -XbcOnly clang -fopenmp -fopenmp-version=45 -g ${OPTIMIZATION} $additional_compile_flags $test -o $exname -lm  > /dev/null 2>&1 ;;
         openrace)   clang -fopenmp -fopenmp-version=45 -Dmasked=master -g -S -emit-llvm $test -o $exname ;;
         tsan-clang) clang $TSAN_COMPILE_FLAGS $additional_compile_flags $test -o $exname -lm ;;
         tsan-gcc)   gcc $TSAN_COMPILE_FLAGS $additional_compile_flags $test -o $exname -lm ;;
@@ -564,7 +566,7 @@ for tool in "${TOOLS[@]}"; do
         tsan-gcc)   gfortran -fopenmp -fsanitize=thread $additional_compile_flags  $test -o $exname -lm  ;;
         archer)     gfortran $FORTRAN_LINK_FLAGS $additional_compile_flags $test -o $linkname;
 	            clang-archer $FORTRAN_COMPILE_FLAGS $linkname $linklib -o $exname $ARCHER_COMPILE_FLAGS -lm;;
-        coderrect)  coderrect -XbcOnly gfortran -O0 -g -fopenmp $additional_compile_flags $test -o $exname -lm > /dev/null 2>&1;
+        coderrect)  coderrect -XbcOnly gfortran ${OPTIMIZATION} -g -fopenmp $additional_compile_flags $test -o $exname -lm > /dev/null 2>&1;
                     ls .coderrect/build/$exname.bc ;; # make $? to be 1 if coderrect could not compile the fortran case
         inspector)  ifort $IFORT_FORTRAN_FLAGS  $test -o $exname -lm ;;
         llov)       $FLANG_PATH/bin/flang -fopenmp -S -emit-llvm "$test" -o "$exname.ll";
@@ -579,7 +581,7 @@ for tool in "${TOOLS[@]}"; do
                         -openmp-verify \
                         "$exname.resetbounds.ll" 2> $compilelog;
                     rm -f $exname.ll $exname.ssa.ll $exname.resetbounds.ll;;
-        romp)       gfortran -O0 -g -fopenmp -lomp -ffree-line-length-none $additional_compile_flags $test -o $exname -lm;
+        romp)       gfortran ${OPTIMIZATION} -g -fopenmp -lomp -ffree-line-length-none $additional_compile_flags $test -o $exname -lm;
                     echo $exname
                     InstrumentMain --program=$exname;;
       esac