This commit enables some significant optimisations for esp32:
- move the VM to iRAM
- move hot parts of the runtime to iRAM (map lookup, load global/name,
mp_obj_get_type)
- enable MICROPY_OPT_LOAD_ATTR_FAST_PATH
- enable MICROPY_OPT_MAP_LOOKUP_CACHE
- disable assertions
- change from -Os to -O2 for compilation
It's hard to measure performance on esp32 due to external flash and
hardware caching. But this set of changes improves performance compared to
master by (on a TinyPICO with the GENERIC build, using IDF 4.2.2, running
at 160MHz):
diff of scores (higher is better)
N=100 M=100 esp32-master -> esp32-perf diff diff% (error%)
bm_chaos.py 71.28 -> 268.08 : +196.80 = +276.094% (+/-0.04%)
bm_fannkuch.py 44.10 -> 69.31 : +25.21 = +57.166% (+/-0.01%)
bm_fft.py 1385.27 -> 2538.23 : +1152.96 = +83.230% (+/-0.01%)
bm_float.py 1060.94 -> 3900.62 : +2839.68 = +267.657% (+/-0.03%)
bm_hexiom.py 10.90 -> 32.79 : +21.89 = +200.826% (+/-0.02%)
bm_nqueens.py 1000.83 -> 2372.87 : +1372.04 = +137.090% (+/-0.01%)
bm_pidigits.py 288.13 -> 664.40 : +376.27 = +130.590% (+/-0.46%)
misc_aes.py 102.45 -> 345.69 : +243.24 = +237.423% (+/-0.01%)
misc_mandel.py 1016.58 -> 2121.92 : +1105.34 = +108.731% (+/-0.01%)
misc_pystone.py 632.91 -> 1801.87 : +1168.96 = +184.696% (+/-0.08%)
misc_raytrace.py 76.66 -> 281.78 : +205.12 = +267.571% (+/-0.05%)
viper_call0.py 210.63 -> 273.17 : +62.54 = +29.692% (+/-0.01%)
viper_call1a.py 208.45 -> 269.51 : +61.06 = +29.292% (+/-0.00%)
viper_call1b.py 185.44 -> 228.25 : +42.81 = +23.086% (+/-0.01%)
viper_call1c.py 185.86 -> 228.90 : +43.04 = +23.157% (+/-0.01%)
viper_call2a.py 207.10 -> 267.25 : +60.15 = +29.044% (+/-0.00%)
viper_call2b.py 173.76 -> 209.42 : +35.66 = +20.523% (+/-0.00%)
Five tests have more than 3x speed up (200%+).
The performance of the tests bm_fft, bm_pidigits and misc_aes now scale
with CPU frequency (eg changing frequency to 240MHz boosts the performance
of these by 50%), which means they are no longer influenced by timing of
external flash access. (The viper_call* tests did previously scale with
CPU frequency, and they still do.)
Turning off assertions reduces code size by about 80k, and going from -Os
to -O2 costs about 100k, so the net change in code size (for the GENERIC
board) is about +20k.
If a board wants to enable assertions, or use -Os instead of -O2, that's
still possible by overriding the sdkconfig parameters.
Signed-off-by: Damien George <damien@micropython.org>
Ensures consistent behaviour and resolves the D-Cache bug (the "exhaustive"
argument being lost due to cache being turned off) when O0 is used.
The changes in this commit are:
- Change -O0 to -Os because "gcc is considered broken at -O0" according to
https://github.com/ARM-software/CMSIS_5/issues/620#issuecomment-550235656
- Use volatile for mem_base so the compiler doesn't optimise away reads or
writes to the SDRAM, which is being tested.
- Use DSB to prevent any other compiler optimisations that would change the
testing logic.
- Use alternating pattern/antipattern in exhaustive test to catch more
hardware/configuration errors.
Implementation adapted by @andrewleech, taken directly from investigation
by @iabdalkader and @dpgeorge.
See #7841 and #7869 for further discussion.
To match network_lan.c and network_ppp.c, and make it clear what code is
specifically for WLAN support.
Also provide a configuration option MICROPY_PY_NETWORK_WLAN which can be
used to fully disable network.WLAN (it's enabled by default).
Signed-off-by: Damien George <damien@micropython.org>
To do this the board must define MICROPY_BOARD_STARTUP, set
MICROPY_SOURCE_BOARD then define the new start-up code.
For example, in mpconfigboard.h:
#define MICROPY_BOARD_STARTUP board_startup
void board_startup(void);
in mpconfigboard.cmake:
set(MICROPY_SOURCE_BOARD
${MICROPY_BOARD_DIR}/board.c
)
and in a new board.c file in the board directory:
#include "py/mpconfig.h"
void board_startup(void) {
boardctrl_startup();
// extra custom startup
}
This follows stm32's boardctrl facilities.
Signed-off-by: Damien George <damien@micropython.org>
Because vPortCleanUpTCB is called by the FreeRTOS idle task, and it checks
thread, but didn't check the thread_mutex.
And if thread is not NULL, but thread_mutex not ready then it will crash
with an error when calling mp_thread_mutex_lock(&thread_mutex, 1).
As suggested by @dpgeorge, move the thread = &thread_entry0 line to the end
of mp_thread_init().
Signed-off-by: leo chung <gewalalb@gmail.com>
This callback allows detecting if there is a USB host connected to the CDC
or not, in which case the stdout_tx should skip CDC TX writing and
flushing or the system will block.
Fixes issue #7820.
This commit allows using all the available PWM timers (up to 8) and
channels (up to 16), without affecting the PWM API.
If a new frequency is set, first it checks if another timer is using the
same frequency. If yes, then it uses this timer, otherwise, it creates a
new one. If all timers are used, the user should set an already used
frequency, or de-init a channel.
This work is based on #6276 and #3608.
The H743 has equal sized pages of 128k, which means the filesystem doesn't
need to be near the beginning. This commit moves the filesystem to the
very end of flash, and extends it to 512k (4 pages).
Signed-off-by: Damien George <damien@micropython.org>
This change adds the OLIMEX H407 support to the STM32 port. The H407
(https://www.olimex.com/Products/ARM/ST/STM32-H407/) is simliar to the
already existing E407
(https://www.olimex.com/Products/ARM/ST/STM32-E407) but does not support
Ethernet and has a full-size USB-A port instead of a Mini-USB socket.
Both boards use the STM32F407ZGT6 CPU.
This port is basically a copy of the E407 but with changed pinmux:
* Removed Ethernet pin definition
* Removed UART1 (pins are used for other functions)
* Removed UART3 flow control pins (pins are used for other functions)
* Removed SD-Card detect pin (since it is not connected on the H407)
A REPL on UART3 is connected to the U3BOOT-header, a 3-pin header with RX,
TX and GND that is intended for the serial terminal.
Tested:
* Micro-SD Card is detected when inserted on RESET
* REPL on UART3 works
* Serial port on the mini USB socket
Signed-off-by: Chris Fiege <cfi@pengutronix.de>
IDF v4.4 does not have an official release so for now use the latest
master. Also remove building GENERIC with no options (all the other boards
are no-option builds), to keep CI time reasonable.
Signed-off-by: Damien George <damien@micropython.org>
This commit removes all parts of code associated with the existing
MICROPY_OPT_CACHE_MAP_LOOKUP_IN_BYTECODE optimisation option, including the
-mcache-lookup-bc option to mpy-cross.
This feature originally provided a significant performance boost for Unix,
but wasn't able to be enabled for MCU targets (due to frozen bytecode), and
added significant extra complexity to generating and distributing .mpy
files.
The equivalent performance gain is now provided by the combination of
MICROPY_OPT_LOAD_ATTR_FAST_PATH and MICROPY_OPT_MAP_LOOKUP_CACHE (which has
been enabled on the unix port in the previous commit).
It's hard to provide precise performance numbers, but tests have been run
on a wide variety of architectures (x86-64, ARM Cortex, Aarch64, RISC-V,
xtensa) and they all generally agree on the qualitative improvements seen
by the combination of MICROPY_OPT_LOAD_ATTR_FAST_PATH and
MICROPY_OPT_MAP_LOOKUP_CACHE.
For example, on a "quiet" Linux x64 environment (i3-5010U @ 2.10GHz) the
change from CACHE_MAP_LOOKUP_IN_BYTECODE, to LOAD_ATTR_FAST_PATH combined
with MAP_LOOKUP_CACHE is:
diff of scores (higher is better)
N=2000 M=2000 bccache -> attrmapcache diff diff% (error%)
bm_chaos.py 13742.56 -> 13905.67 : +163.11 = +1.187% (+/-3.75%)
bm_fannkuch.py 60.13 -> 61.34 : +1.21 = +2.012% (+/-2.11%)
bm_fft.py 113083.20 -> 114793.68 : +1710.48 = +1.513% (+/-1.57%)
bm_float.py 256552.80 -> 243908.29 : -12644.51 = -4.929% (+/-1.90%)
bm_hexiom.py 521.93 -> 625.41 : +103.48 = +19.826% (+/-0.40%)
bm_nqueens.py 197544.25 -> 217713.12 : +20168.87 = +10.210% (+/-3.01%)
bm_pidigits.py 8072.98 -> 8198.75 : +125.77 = +1.558% (+/-3.22%)
misc_aes.py 17283.45 -> 16480.52 : -802.93 = -4.646% (+/-0.82%)
misc_mandel.py 99083.99 -> 128939.84 : +29855.85 = +30.132% (+/-5.88%)
misc_pystone.py 83860.10 -> 82592.56 : -1267.54 = -1.511% (+/-2.27%)
misc_raytrace.py 21490.40 -> 22227.23 : +736.83 = +3.429% (+/-1.88%)
This shows that the new optimisations are at least as good as the existing
inline-bytecode-caching, and are sometimes much better (because the new
ones apply caching to a wider variety of map lookups).
The new optimisations can also benefit code generated by the native
emitter, because they apply to the runtime rather than the generated code.
The improvement for the native emitter when LOAD_ATTR_FAST_PATH and
MAP_LOOKUP_CACHE are enabled is (same Linux environment as above):
diff of scores (higher is better)
N=2000 M=2000 native -> nat-attrmapcache diff diff% (error%)
bm_chaos.py 14130.62 -> 15464.68 : +1334.06 = +9.441% (+/-7.11%)
bm_fannkuch.py 74.96 -> 76.16 : +1.20 = +1.601% (+/-1.80%)
bm_fft.py 166682.99 -> 168221.86 : +1538.87 = +0.923% (+/-4.20%)
bm_float.py 233415.23 -> 265524.90 : +32109.67 = +13.756% (+/-2.57%)
bm_hexiom.py 628.59 -> 734.17 : +105.58 = +16.796% (+/-1.39%)
bm_nqueens.py 225418.44 -> 232926.45 : +7508.01 = +3.331% (+/-3.10%)
bm_pidigits.py 6322.00 -> 6379.52 : +57.52 = +0.910% (+/-5.62%)
misc_aes.py 20670.10 -> 27223.18 : +6553.08 = +31.703% (+/-1.56%)
misc_mandel.py 138221.11 -> 152014.01 : +13792.90 = +9.979% (+/-2.46%)
misc_pystone.py 85032.14 -> 105681.44 : +20649.30 = +24.284% (+/-2.25%)
misc_raytrace.py 19800.01 -> 23350.73 : +3550.72 = +17.933% (+/-2.79%)
In summary, compared to MICROPY_OPT_CACHE_MAP_LOOKUP_IN_BYTECODE, the new
MICROPY_OPT_LOAD_ATTR_FAST_PATH and MICROPY_OPT_MAP_LOOKUP_CACHE options:
- are simpler;
- take less code size;
- are faster (generally);
- work with code generated by the native emitter;
- can be used on embedded targets with a small and constant RAM overhead;
- allow the same .mpy bytecode to run on all targets.
See #7680 for further discussion. And see also #7653 for a discussion
about simplifying mpy-cross options.
Signed-off-by: Jim Mussared <jim.mussared@gmail.com>
The existing inline bytecode caching optimisation, selected by
MICROPY_OPT_CACHE_MAP_LOOKUP_IN_BYTECODE, reserves an extra byte in the
bytecode after certain opcodes, which at runtime stores a map index of the
likely location of this field when looking up the qstr. This scheme is
incompatible with bytecode-in-ROM, and doesn't work with native generated
code. It also stores bytecode in .mpy files which is of a different format
to when the feature is disabled, making generation of .mpy files more
complex.
This commit provides an alternative optimisation via an approach that adds
a global cache for map offsets, then all mp_map_lookup operations use it.
It's less precise than bytecode caching, but allows the cache to be
independent and external to the bytecode that is executing. It also works
for the native emitter and adds a similar performance boost on top of the
gain already provided by the native emitter.
Signed-off-by: Jim Mussared <jim.mussared@gmail.com>
When the LOAD_ATTR opcode is executed there are quite a few different cases
that have to be handled, but the common case is accessing a member on an
instance type. Typically, built-in types provide methods which is why this
is common.
Fortunately, for this specific case, if the member is found in the member
map then there's no further processing.
This optimisation does a relatively cheap check (type is instance) and then
forwards directly to the member map lookup, falling back to the regular
path if necessary.
Signed-off-by: Jim Mussared <jim.mussared@gmail.com>
