tests/bench_xtimer: initial import

tests/bench_xtimer/Makefile

@@ -0,0 +1,76 @@
+include ../Makefile.tests_common
+
+USEMODULE += xtimer
+
+# this test uses 1000 timers by default. for boards that boards don't have
+# enough memory, reduce that to 100 or 20, unless NUMOF_TIMERS has been overridden.
+LOW_MEMORY_BOARDS += \
+  airfy-beacon \
+  arduino-mega2560 \
+  b-l072z-lrwan1 \
+  blackpill \
+  blackpill-128kib \
+  bluepill \
+  bluepill-128kib \
+  calliope-mini \
+  cc1352-launchpad \
+  cc2650-launchpad \
+  cc2650stk \
+  chronos \
+  hifive1 \
+  hifive1b \
+  i-nucleo-lrwan1 \
+  lsn50 \
+  maple-mini \
+  microbit \
+  msb-430 \
+  msb-430h \
+  nrf51dongle \
+  nrf6310 \
+  nucleo-f030r8 \
+  nucleo-f042k6 \
+  nucleo-f070rb \
+  nucleo-f072rb \
+  nucleo-f103rb \
+  nucleo-f303k8 \
+  nucleo-f334r8 \
+  nucleo-l031k6 \
+  nucleo-l053r8 \
+  nucleo-l073rz \
+  opencm904 \
+  saml10-xpro \
+  saml11-xpro \
+  spark-core \
+  stm32f0discovery \
+  stm32l0538-disco \
+  telosb \
+  waspmote-pro \
+  wsn430-v1_3b \
+  wsn430-v1_4 \
+  yunjia-nrf51822 \
+  z1 \
+  #
+
+SUPER_LOW_MEMORY_BOARDS += \
+  arduino-duemilanove \
+  arduino-leonardo \
+  arduino-nano \
+  arduino-uno \
+  atmega328p \
+  nucleo-f031k6 \
+  stm32f030f4-demo \
+  #
+
+ifneq (, $(filter $(BOARD), $(LOW_MEMORY_BOARDS)))
+  NUMOF_TIMERS ?= 100
+endif
+
+ifneq (, $(filter $(BOARD), $(SUPER_LOW_MEMORY_BOARDS)))
+  NUMOF_TIMERS ?= 20
+endif
+
+NUMOF_TIMERS ?= 1000
+
+CFLAGS += -DNUMOF_TIMERS=$(NUMOF_TIMERS)
+
+include $(RIOTBASE)/Makefile.include

tests/bench_xtimer/README.md

@@ -0,0 +1,77 @@
+# Introduction
+
+This test executes some benchmarks for xtimer's set() / remove() / now()
+operations.
+
+# Details
+
+This set of benchmarks measures xtimer's list operation efficiency.
+Depending on the available memory, the individual benchmarks that are using
+multiple timers are run with either 1000 (the default), 100 or 20 timers.
+Each benchmark is repeated REPEAT times (default 1000).
+As only the operations are benchmarked, it is asserted that no timer ever
+actually triggers.
+
+### set() one
+
+This repeatedly sets one timer in an otherwise emtpy list.
+All but the first iteration will cause xtimer to implicitly remove the timer
+first.
+All iterations will cause the underlying periph timer to be updated.
+
+### remove() one
+
+This repeatedly removes one timer from the list. In all but the first iteration,
+this list will be empty.
+
+
+### set() + remove() one
+
+This repeatedly first sets, then removes one timer. The list is empty
+before and after each iteration.
+All iterations will cause the underlying periph timer to be updated.
+
+### set() many increasing targets
+
+This adds NUMOF timers with increasing target times. Each iteration will add a
+timer at the end of xtimer's timer list.
+Only the first iteration will cause the underlying periph timer to be updated.
+After this test, the list is populated with NUMOF timers.
+
+### re-set() first
+
+This repatedly re-sets the first timer in the list (to the same target time).
+All iterations will cause the underlying periph timer to be updated.
+
+### re-set() middle
+
+This repatedly re-sets the timer in the middle of the list (to the same target
+time).
+
+### re-set() last
+
+This repatedly re-sets the last timer in the list (to the same target time).
+
+### remove() + set() first, middle, last
+
+Same as the previous three, but does a remove() before set().
+
+### remove() many decreasing
+
+This removes all timers from the list, starting with the last.
+
+### xtimer_now()
+
+This simply calls xtimer_now() in a loop.
+
+
+# How to interpret results
+
+The aim is to measure the time spent in xtimer's list operations.
+Lower values are better.
+The first/middle/last tests give an idea of the best case / average case /
+worst case when running the operation with NUMOF timers.
+Note that every set() on an already set timer will trigger an implicit remove(),
+thus the timer list has to be iterated twice.
+The tests that do a remove() before set() show whether xtimer correctly
+identifies an unset timer.