moebiusband/MD-Bench
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Update script

Browse Source 
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
This commit is contained in:
Rafael Ravedutti 2022-12-21 18:04:18 +01:00
parent 8d5e10f635
commit e5c233e072
3 changed files with 30 additions and 18 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
common/includes/simd/avx512_float.h
View File

@ -7,10 +7,13 @@
#include <stdlib.h>
#include <string.h>
#include <immintrin.h>
#include <zmmintrin.h>
#ifndef NO_ZMM_INTRIN
# include <zmmintrin.h>
#endif
#define MD_SIMD_FLOAT __m512
#define MD_SIMD_MASK __mmask16
#define MD_SIMD_FLOAT __m512
#define MD_SIMD_MASK __mmask16
#define MD_SIMD_INT __m256i
static inline MD_SIMD_FLOAT simd_broadcast(float scalar) { return _mm512_set1_ps(scalar); }
static inline MD_SIMD_FLOAT simd_zero() { return _mm512_set1_ps(0.0f); }
@ -69,7 +72,7 @@ static inline MD_FLOAT simd_h_dual_incr_reduced_sum(float* m, MD_SIMD_FLOAT v0,
return _mm_cvtss_f32(t3);
}
inline void simd_h_decr(MD_FLOAT *m, MD_SIMD_FLOAT a) {
static inline void simd_h_decr(MD_FLOAT *m, MD_SIMD_FLOAT a) {
__m256 t;
a = _mm512_add_ps(a, _mm512_shuffle_f32x4(a, a, 0xee));
t = _mm256_load_ps(m);

5
include_ICX.mk
View File

@ -3,13 +3,14 @@ LINKER = $(CC)
OPENMP = #-qopenmp
PROFILE = #-profile-functions -g -pg
#OPTS = -Ofast -xCORE-AVX512 -qopt-zmm-usage=high $(PROFILE)
OPTS = -Ofast -xCORE-AVX512 -qopt-zmm-usage=high $(PROFILE)
#OPTS = -Ofast -march=cascadelake -xCORE-AVX512 -qopt-zmm-usage=high $(PROFILE)
#OPTS = -Ofast -xCORE-AVX2 $(PROFILE)
#OPTS = -Ofast -xAVX $(PROFILE)
#OPTS = -Ofast -xAVX2 $(PROFILE)
#OPTS = -Ofast -xSSE4.2 $(PROFILE)
#OPTS = -Ofast -no-vec $(PROFILE)
OPTS = -Ofast -xHost $(PROFILE)
#OPTS = -Ofast -xHost $(PROFILE)
CFLAGS = $(PROFILE) $(OPENMP) $(OPTS)
ASFLAGS = #-masm=intel
LFLAGS = $(PROFILE) $(OPTS) $(OPENMP)

32
util/evaluate_latency_and_cfd.sh
View File

@ -1,14 +1,18 @@
#!/bin/bash
TAG=ICC
OPT_SCHEME=lammps
# Adjustable variables
TAG="${TAG:-ICX}"
OPT_SCHEME="${OPT_SCHEME:-gromacs}"
CORE="${CORE:-0}"
FREQ="${FREQ:-2.4}"
NRUNS="${NRUNS:-3}"
LOG="${LOG:-latencies_and_cfds.log}"
STUB_ONLY="${STUB_ONLY:-false}"
# Other useful variables
MDBENCH_BIN=./MDBench-$TAG-$OPT_SCHEME
CORE=0
FREQ=2.4
NRUNS=3
FIXED_PARAMS="--freq $FREQ"
CPU_VENDOR=$(lscpu | grep "Vendor ID" | tr -s ' ' | cut -d ' ' -f3)
LOG=latencies_and_cfds.log
if [ "$CPU_VENDOR" == "GenuineIntel" ]; then
ALL_PREFETCHERS="HW_PREFETCHER,CL_PREFETCHER,DCU_PREFETCHER,IP_PREFETCHER"
@ -45,6 +49,7 @@ echo "Optimization scheme: $OPT_SCHEME" | tee -a $LOG
echo "Binary: $MDBENCH_BIN(-stub)" | tee -a $LOG
echo "Frequency: $FREQ" | tee -a $LOG
echo "Number of runs: $NRUNS" | tee -a $LOG
echo "Run only stubbed cases: $STUB_ONLY" | tee -a $LOG
echo "Fixing frequencies..."
likwid-setFrequencies -f $FREQ -t 0
@ -65,12 +70,15 @@ for p in $PREFETCHERS; do
fi
MSG="$p: "
run_benchmark $MDBENCH_BIN
MSG+="standard=$BEST, "
run_benchmark $MDBENCH_BIN -i data/copper_melting/input_lj_cu_one_atomtype_20x20x20.dmp
MSG+="melt=$BEST, "
run_benchmark $MDBENCH_BIN -p data/argon_1000/mdbench_params.conf -i data/argon_1000/tprout.gro
MSG+="argon=$BEST, "
if [ "$STUB_ONLY" == "false" ]; then
run_benchmark $MDBENCH_BIN
MSG+="standard=$BEST, "
run_benchmark $MDBENCH_BIN -i data/copper_melting/input_lj_cu_one_atomtype_20x20x20.dmp
MSG+="melt=$BEST, "
run_benchmark $MDBENCH_BIN -p data/argon_1000/mdbench_params.conf -i data/argon_1000/tprout.gro
MSG+="argon=$BEST, "
fi
run_benchmark $MDBENCH_BIN-stub $STUB1_PARAMS
MSG+="$STUB1_NAME=$BEST, "
run_benchmark $MDBENCH_BIN-stub $STUB2_PARAMS