From e5c233e072accd611b83c83f3e048777acbfafb1 Mon Sep 17 00:00:00 2001
From: Rafael Ravedutti <rafaelravedutti@gmail.com>
Date: Wed, 21 Dec 2022 18:04:18 +0100
Subject: [PATCH] Update script

Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
---
 common/includes/simd/avx512_float.h | 11 ++++++----
 include_ICX.mk                      |  5 +++--
 util/evaluate_latency_and_cfd.sh    | 32 ++++++++++++++++++-----------
 3 files changed, 30 insertions(+), 18 deletions(-)

diff --git a/common/includes/simd/avx512_float.h b/common/includes/simd/avx512_float.h
index 954e54f..1fe5803 100644
--- a/common/includes/simd/avx512_float.h
+++ b/common/includes/simd/avx512_float.h
@@ -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);
diff --git a/include_ICX.mk b/include_ICX.mk
index 0f21d33..ca45ee7 100644
--- a/include_ICX.mk
+++ b/include_ICX.mk
@@ -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)
diff --git a/util/evaluate_latency_and_cfd.sh b/util/evaluate_latency_and_cfd.sh
index 140fe38..5d02093 100644
--- a/util/evaluate_latency_and_cfd.sh
+++ b/util/evaluate_latency_and_cfd.sh
@@ -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