Add benchmark workflow (RPi4 / Arm Cortex-A72)

.github/workflows/bench.yml

@@ -33,4 +33,4 @@ jobs:
       - name: Run benchmark
         shell: nix develop .#ci -c bash -e {0}
         run: |
-          tests bench -c PMU -v
+          tests bench -c PMU --cpu cortex-a72 -v

Makefile

@@ -9,23 +9,23 @@ INCLUDE_FIPS202 = -I fips202
 INCLUDE_MLKEM = -I mlkem
 INCLUDE_RANDOM = -I randombytes
 INCLUDE_NISTRANDOM = -I test/nistrng
-CFLAGS += -Wall -Wextra -Wpedantic -Werror -Wmissing-prototypes -Wredundant-decls \
+override CFLAGS += -Wall -Wextra -Wpedantic -Werror -Wmissing-prototypes -Wredundant-decls \
   -Wshadow -Wpointer-arith -Wno-unknown-pragmas -O3 -fomit-frame-pointer -pedantic \
    ${INCLUDE_MLKEM} ${INCLUDE_FIPS202}
 
 HOST_PLATFORM := $(shell uname -s)-$(shell uname -m)
 ifeq ($(HOST_PLATFORM),Linux-x86_64)
-	CFLAGS += -static
+	override CFLAGS += -static
 endif
 
 CYCLES ?= NO
 
 ifeq ($(CYCLES),PMU)
-	CFLAGS += -DPMU_CYCLES
+	override CFLAGS += -DPMU_CYCLES
 endif
 
 ifeq ($(CYCLES),PERF)
-	CFLAGS += -DPERF_CYCLES
+	override CFLAGS += -DPERF_CYCLES
 endif
 
 CFLAGS_RANDOMBYTES = ${CFLAGS} ${INCLUDE_RANDOM}

scripts/tests

@@ -300,7 +300,7 @@ def bench(force_qemu, verbose, cycles, cpu):
         lambda _: True,
         force_qemu,
         verbose,
-        f"CYCLES={cycles}",
+        [f"CYCLES={cycles}", f"CFLAGS=-mcpu={cpu}"],
     )
     for scheme, result in results.items():
         print(scheme)