BS: Support CPU baseline compilation without -march=native

4 files changed, 21 insertions, 10 deletions

diff --git a/BS/baselines/cpu/Makefile b/BS/baselines/cpu/Makefile
index b67602f..735fe84 100644
--- a/BS/baselines/cpu/Makefile
+++ b/BS/baselines/cpu/Makefile
@@ -1,16 +1,23 @@
-NUMA ?= 0
-NUMA_MEMCPY ?= 0
-FLAGS =
+native ?= 1
+numa ?= 0
+numa_memcpy ?= 0
 
-ifeq (${NUMA}, 1)
-	FLAGS += -lnuma
+CFLAGS =
+LDFLAGS =
+
+ifeq (${native}, 1)
+	CFLAGS += -march=native
+endif
+
+ifeq (${numa}, 1)
+	LDFLAGS += -lnuma
 endif
 
 .PHONY: all
 all: bs_omp
 
 bs_omp: bs_omp.c
-	gcc -Wall -Wextra -pedantic -march=native -O2 -DNUMA=${NUMA} -DNUMA_MEMCPY=${NUMA_MEMCPY} bs_omp.c -o bs_omp -fopenmp ${FLAGS}
+	gcc -Wall -Wextra -pedantic -O3 ${CFLAGS} -DNUMA=${numa} -DNUMA_MEMCPY=${numa_memcpy} bs_omp.c -o bs_omp -fopenmp ${LDFLAGS}
 
 bs_omp_O0: bs_omp.c
 	gcc bs_omp.c -o bs_omp_O0 -fopenmp
diff --git a/BS/baselines/cpu/run-perf.sh b/BS/baselines/cpu/run-perf.sh
index a35e0fc..5b671e0 100755
--- a/BS/baselines/cpu/run-perf.sh
+++ b/BS/baselines/cpu/run-perf.sh
@@ -1,6 +1,6 @@
 #!/bin/zsh
 
-make -B NUMA=1
+make -B numa=1
 
 OMP_NUM_THREADS=1 perf stat record -o t1.perf -e ${(j:,:):-$(grep -v '^#' ../../../perf-events.txt | cut -d ' ' -f 1)} ./bs_omp $((2**29)) 16777216 4 4
 OMP_NUM_THREADS=4 perf stat record -o t4.perf -e ${(j:,:):-$(grep -v '^#' ../../../perf-events.txt | cut -d ' ' -f 1)} ./bs_omp $((2**29)) 16777216 4 4
diff --git a/BS/dimes-hetsim-hbm.sh b/BS/dimes-hetsim-hbm.sh
index 4e1500d..4a775ae 100755
--- a/BS/dimes-hetsim-hbm.sh
+++ b/BS/dimes-hetsim-hbm.sh
@@ -1,7 +1,7 @@
 #!/bin/bash
 
 cd baselines/cpu
-make -B NUMA=1
+make -B numa=1
 
 mkdir -p log/$(hostname)
 fn=log/$(hostname)/dimes-hetsim-hbm
diff --git a/BS/dimes-hetsim-nmc.sh b/BS/dimes-hetsim-nmc.sh
index 195334b..fa697bf 100755
--- a/BS/dimes-hetsim-nmc.sh
+++ b/BS/dimes-hetsim-nmc.sh
@@ -3,6 +3,8 @@
 mkdir -p log/$(hostname) baselines/cpu/log/$(hostname)
 fn=log/$(hostname)/dimes-hetsim-nmc
 
+source /opt/upmem/upmem-2024.1.0-Linux-x86_64/upmem_env.sh
+
 # upstream DPU version uses 2048576 * uint64 ≈ 16 MiB (DPU max: 64 MiB)
 # upstream DPU version uses 2 queries
 input_size_upstream=2048576
@@ -11,6 +13,8 @@ num_queries_upstream=2
 input_size_dpu=$(perl -E 'say 2 ** 22')
 num_queries_dpu=1048576
 
+# Make sure that num_queries > input_size!
+
 run_benchmark_nmc() {
 	local "$@"
 	set -e
@@ -69,7 +73,7 @@ cd baselines/cpu
 
 (
 
-make -B NUMA=1 NUMA_MEMCPY=1
+make -B numa=1 numa_memcpy=1
 
 echo "CPU single-node upstream-ref with memcpy, copy node == input node (1/6)" >&2
 
@@ -97,7 +101,7 @@ parallel -j1 --eta --joblog ${fn}.2.joblog --resume --header : \
 	:::+      cpu 0 1 \
 	::: nr_threads 1 2 4 8 12 16
 
-make -B NUMA=1
+make -B numa=1
 
 echo "CPU single-node upstream-ref (3/6)" >&2