Browse Source

rp2/CMakeLists: Apply O2 optimisation to map, mpz and vm source code.

Enabling this optimisation increases the RPI_PICO firmware by +3344 bytes.

Performace change is:

diff of scores (higher is better)
N=100 M=100             rpi-base -> rpi-opt      diff      diff% (error%)
bm_chaos.py               196.56 ->  215.26 :  +18.70 =  +9.514% (+/-0.05%)
bm_fannkuch.py             52.47 ->   54.37 :   +1.90 =  +3.621% (+/-0.05%)
bm_fft.py                1476.74 -> 1530.06 :  +53.32 =  +3.611% (+/-0.01%)
bm_float.py              2305.65 -> 2444.11 : +138.46 =  +6.005% (+/-0.08%)
bm_hexiom.py               32.83 ->   35.09 :   +2.26 =  +6.884% (+/-0.05%)
bm_nqueens.py            2335.85 -> 2259.78 :  -76.07 =  -3.257% (+/-0.06%)
bm_pidigits.py            366.23 ->  465.81 :  +99.58 = +27.191% (+/-0.04%)
bm_wordcount.py            41.20 ->   41.87 :   +0.67 =  +1.626% (+/-0.01%)
core_import_mpy_multi.py  327.44 ->  335.24 :   +7.80 =  +2.382% (+/-0.08%)
core_import_mpy_single.py  63.41 ->   64.98 :   +1.57 =  +2.476% (+/-0.21%)
core_locals.py             27.24 ->   29.19 :   +1.95 =  +7.159% (+/-0.01%)
core_qstr.py              137.31 ->  140.84 :   +3.53 =  +2.571% (+/-0.03%)
core_str.py                18.44 ->   18.10 :   -0.34 =  -1.844% (+/-0.03%)
core_yield_from.py        221.69 ->  211.72 :   -9.97 =  -4.497% (+/-0.01%)
misc_aes.py               303.38 ->  308.72 :   +5.34 =  +1.760% (+/-0.02%)
misc_mandel.py           1501.00 -> 1746.60 : +245.60 = +16.362% (+/-0.04%)
misc_pystone.py          1348.22 -> 1456.75 : +108.53 =  +8.050% (+/-0.07%)
misc_raytrace.py          223.81 ->  246.16 :  +22.35 =  +9.986% (+/-0.07%)
viper_call0.py            331.05 ->  331.10 :   +0.05 =  +0.015% (+/-0.00%)
viper_call1a.py           323.34 ->  323.39 :   +0.05 =  +0.015% (+/-0.00%)
viper_call1b.py           241.01 ->  241.04 :   +0.03 =  +0.012% (+/-0.00%)
viper_call1c.py           242.88 ->  242.92 :   +0.04 =  +0.016% (+/-0.00%)
viper_call2a.py           318.41 ->  318.46 :   +0.05 =  +0.016% (+/-0.00%)
viper_call2b.py           211.27 ->  211.30 :   +0.03 =  +0.014% (+/-0.00%)

And the test `tests/basics/builtin_pow3_intbig.py` now passes (it uses a
lot of mpz code and previously took longer than 10 seconds to run on
RPI_PICO, which would lead to a timeout in the test runner).

Signed-off-by: Damien George <damien@micropython.org>
pull/14182/head
Damien George 8 months ago
parent
commit
79edaddf50
  1. 9
      ports/rp2/CMakeLists.txt

9
ports/rp2/CMakeLists.txt

@ -455,6 +455,15 @@ target_link_options(${MICROPY_TARGET} PRIVATE
-Wl,--wrap=dcd_event_handler
)
# Apply optimisations to performance-critical source code.
set_source_files_properties(
${MICROPY_PY_DIR}/map.c
${MICROPY_PY_DIR}/mpz.c
${MICROPY_PY_DIR}/vm.c
PROPERTIES
COMPILE_OPTIONS "-O2"
)
set_source_files_properties(
${PICO_SDK_PATH}/src/rp2_common/pico_double/double_math.c
${PICO_SDK_PATH}/src/rp2_common/pico_float/float_math.c

Loading…
Cancel
Save