2014-05-03 23:27:38 +01:00
|
|
|
/*
|
2017-06-30 08:22:17 +01:00
|
|
|
* This file is part of the MicroPython project, http://micropython.org/
|
2014-05-03 23:27:38 +01:00
|
|
|
*
|
|
|
|
* The MIT License (MIT)
|
|
|
|
*
|
py: Automatically provide weak links from "foo" to "ufoo" module name.
This commit implements automatic module weak links for all built-in
modules, by searching for "ufoo" in the built-in module list if "foo"
cannot be found. This means that all modules named "ufoo" are always
available as "foo". Also, a port can no longer add any other weak links,
which makes strict the definition of a weak link.
It saves some code size (about 100-200 bytes) on ports that previously had
lots of weak links.
Some changes from the previous behaviour:
- It doesn't intern the non-u module names (eg "foo" is not interned),
which saves code size, but will mean that "import foo" creates a new qstr
(namely "foo") in RAM (unless the importing module is frozen).
- help('modules') no longer lists non-u module names, only the u-variants;
this reduces duplication in the help listing.
Weak links are effectively the same as having a set of symbolic links on
the filesystem that is searched last. So an "import foo" will search
built-in modules first, then all paths in sys.path, then weak links last,
importing "ufoo" if it exists. Thus a file called "foo.py" somewhere in
sys.path will still have precedence over the weak link of "foo" to "ufoo".
See issues: #1740, #4449, #5229, #5241.
2019-10-21 15:06:34 +01:00
|
|
|
* Copyright (c) 2013-2019 Damien P. George
|
2014-05-13 06:44:45 +01:00
|
|
|
* Copyright (c) 2014 Paul Sokolovsky
|
2020-02-26 04:24:09 +00:00
|
|
|
* Copyright (c) 2021 Jim Mussared
|
2014-05-03 23:27:38 +01:00
|
|
|
*
|
|
|
|
* Permission is hereby granted, free of charge, to any person obtaining a copy
|
|
|
|
* of this software and associated documentation files (the "Software"), to deal
|
|
|
|
* in the Software without restriction, including without limitation the rights
|
|
|
|
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
|
|
|
* copies of the Software, and to permit persons to whom the Software is
|
|
|
|
* furnished to do so, subject to the following conditions:
|
|
|
|
*
|
|
|
|
* The above copyright notice and this permission notice shall be included in
|
|
|
|
* all copies or substantial portions of the Software.
|
|
|
|
*
|
|
|
|
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
|
|
|
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
|
|
|
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
|
|
|
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
|
|
|
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
|
|
|
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
|
|
|
|
* THE SOFTWARE.
|
|
|
|
*/
|
|
|
|
|
2014-01-03 14:03:48 +00:00
|
|
|
#include <stdio.h>
|
|
|
|
#include <string.h>
|
|
|
|
#include <assert.h>
|
|
|
|
|
2015-01-01 20:27:54 +00:00
|
|
|
#include "py/compile.h"
|
|
|
|
#include "py/objmodule.h"
|
2016-11-16 00:55:41 +00:00
|
|
|
#include "py/persistentcode.h"
|
2015-01-01 20:27:54 +00:00
|
|
|
#include "py/runtime.h"
|
|
|
|
#include "py/builtin.h"
|
2015-01-20 09:52:12 +00:00
|
|
|
#include "py/frozenmod.h"
|
2014-01-03 14:03:48 +00:00
|
|
|
|
2017-07-24 17:55:14 +01:00
|
|
|
#if MICROPY_DEBUG_VERBOSE // print debugging info
|
2014-04-11 21:08:29 +01:00
|
|
|
#define DEBUG_PRINT (1)
|
|
|
|
#define DEBUG_printf DEBUG_printf
|
|
|
|
#else // don't print debugging info
|
2014-11-05 21:16:41 +00:00
|
|
|
#define DEBUG_PRINT (0)
|
2014-04-11 21:08:29 +01:00
|
|
|
#define DEBUG_printf(...) (void)0
|
|
|
|
#endif
|
|
|
|
|
2022-11-01 18:44:23 +00:00
|
|
|
#if MICROPY_MODULE_WEAK_LINKS
|
|
|
|
STATIC qstr make_weak_link_name(vstr_t *buffer, qstr name) {
|
|
|
|
vstr_reset(buffer);
|
|
|
|
vstr_add_char(buffer, 'u');
|
|
|
|
vstr_add_str(buffer, qstr_str(name));
|
|
|
|
return qstr_from_strn(buffer->buf, buffer->len);
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
2018-02-20 07:00:44 +00:00
|
|
|
#if MICROPY_ENABLE_EXTERNAL_IMPORT
|
|
|
|
|
2020-02-26 04:24:09 +00:00
|
|
|
// Must be a string of one byte.
|
|
|
|
#define PATH_SEP_CHAR "/"
|
2014-02-05 23:57:48 +00:00
|
|
|
|
2021-12-11 11:40:21 +00:00
|
|
|
// Virtual sys.path entry that maps to the frozen modules.
|
|
|
|
#define MP_FROZEN_PATH_PREFIX ".frozen/"
|
|
|
|
|
2014-10-25 19:04:13 +01:00
|
|
|
bool mp_obj_is_package(mp_obj_t module) {
|
|
|
|
mp_obj_t dest[2];
|
|
|
|
mp_load_method_maybe(module, MP_QSTR___path__, dest);
|
|
|
|
return dest[0] != MP_OBJ_NULL;
|
|
|
|
}
|
|
|
|
|
2020-02-26 04:24:09 +00:00
|
|
|
// Wrapper for mp_import_stat (which is provided by the port, and typically
|
|
|
|
// uses mp_vfs_import_stat) to also search frozen modules. Given an exact
|
|
|
|
// path to a file or directory (e.g. "foo/bar", foo/bar.py" or "foo/bar.mpy"),
|
|
|
|
// will return whether the path is a file, directory, or doesn't exist.
|
|
|
|
STATIC mp_import_stat_t stat_path_or_frozen(const char *path) {
|
2016-05-23 12:42:23 +01:00
|
|
|
#if MICROPY_MODULE_FROZEN
|
2021-12-11 11:40:21 +00:00
|
|
|
// Only try and load as a frozen module if it starts with .frozen/.
|
|
|
|
const int frozen_path_prefix_len = strlen(MP_FROZEN_PATH_PREFIX);
|
|
|
|
if (strncmp(path, MP_FROZEN_PATH_PREFIX, frozen_path_prefix_len) == 0) {
|
|
|
|
return mp_find_frozen_module(path + frozen_path_prefix_len, NULL, NULL);
|
2016-05-21 19:33:42 +01:00
|
|
|
}
|
2016-05-21 20:23:08 +01:00
|
|
|
#endif
|
2016-05-21 19:33:42 +01:00
|
|
|
return mp_import_stat(path);
|
|
|
|
}
|
|
|
|
|
2020-02-26 04:24:09 +00:00
|
|
|
// Given a path to a .py file, try and find this path as either a .py or .mpy
|
|
|
|
// in either the filesystem or frozen modules.
|
2016-12-13 04:09:48 +00:00
|
|
|
STATIC mp_import_stat_t stat_file_py_or_mpy(vstr_t *path) {
|
2020-02-26 04:24:09 +00:00
|
|
|
mp_import_stat_t stat = stat_path_or_frozen(vstr_null_terminated_str(path));
|
2014-02-05 23:57:48 +00:00
|
|
|
if (stat == MP_IMPORT_STAT_FILE) {
|
|
|
|
return stat;
|
2014-01-19 22:03:34 +00:00
|
|
|
}
|
2015-11-02 21:57:42 +00:00
|
|
|
|
|
|
|
#if MICROPY_PERSISTENT_CODE_LOAD
|
2020-02-26 04:24:09 +00:00
|
|
|
// Didn't find .py -- try the .mpy instead by inserting an 'm' into the '.py'.
|
2015-11-02 21:57:42 +00:00
|
|
|
vstr_ins_byte(path, path->len - 2, 'm');
|
2020-02-26 04:24:09 +00:00
|
|
|
stat = stat_path_or_frozen(vstr_null_terminated_str(path));
|
2015-11-02 21:57:42 +00:00
|
|
|
if (stat == MP_IMPORT_STAT_FILE) {
|
|
|
|
return stat;
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
2014-02-05 23:57:48 +00:00
|
|
|
return MP_IMPORT_STAT_NO_EXIST;
|
|
|
|
}
|
2014-01-19 22:03:34 +00:00
|
|
|
|
2020-02-26 04:24:09 +00:00
|
|
|
// Given an import path (e.g. "foo/bar"), try and find "foo/bar" (a directory)
|
|
|
|
// or "foo/bar.(m)py" in either the filesystem or frozen modules.
|
2016-12-13 04:09:48 +00:00
|
|
|
STATIC mp_import_stat_t stat_dir_or_file(vstr_t *path) {
|
2020-02-26 04:24:09 +00:00
|
|
|
mp_import_stat_t stat = stat_path_or_frozen(vstr_null_terminated_str(path));
|
2016-12-13 04:09:48 +00:00
|
|
|
DEBUG_printf("stat %s: %d\n", vstr_str(path), stat);
|
|
|
|
if (stat == MP_IMPORT_STAT_DIR) {
|
|
|
|
return stat;
|
|
|
|
}
|
|
|
|
|
|
|
|
// not a directory, add .py and try as a file
|
|
|
|
vstr_add_str(path, ".py");
|
|
|
|
return stat_file_py_or_mpy(path);
|
|
|
|
}
|
|
|
|
|
2023-05-10 04:02:09 +01:00
|
|
|
// Given a top-level module name, try and find it in each of the sys.path
|
|
|
|
// entries. Note: On success, the dest argument will be updated to the matching
|
|
|
|
// path (i.e. "<entry>/mod_name(.py)").
|
2020-02-26 04:24:09 +00:00
|
|
|
STATIC mp_import_stat_t stat_top_level_dir_or_file(qstr mod_name, vstr_t *dest) {
|
|
|
|
DEBUG_printf("stat_top_level_dir_or_file: '%s'\n", qstr_str(mod_name));
|
2014-11-05 20:02:33 +00:00
|
|
|
#if MICROPY_PY_SYS
|
2017-03-25 08:35:08 +00:00
|
|
|
size_t path_num;
|
2014-02-04 22:47:06 +00:00
|
|
|
mp_obj_t *path_items;
|
2014-04-13 04:43:18 +01:00
|
|
|
mp_obj_list_get(mp_sys_path, &path_num, &path_items);
|
2014-02-04 22:47:06 +00:00
|
|
|
|
2023-05-10 04:02:09 +01:00
|
|
|
// go through each sys.path entry, trying to import "<entry>/<mod_name>".
|
|
|
|
for (size_t i = 0; i < path_num; i++) {
|
|
|
|
vstr_reset(dest);
|
|
|
|
size_t p_len;
|
|
|
|
const char *p = mp_obj_str_get_data(path_items[i], &p_len);
|
|
|
|
if (p_len > 0) {
|
|
|
|
// Add the path separator (unless the entry is "", i.e. cwd).
|
|
|
|
vstr_add_strn(dest, p, p_len);
|
|
|
|
vstr_add_char(dest, PATH_SEP_CHAR[0]);
|
|
|
|
}
|
|
|
|
vstr_add_str(dest, qstr_str(mod_name));
|
|
|
|
mp_import_stat_t stat = stat_dir_or_file(dest);
|
|
|
|
if (stat != MP_IMPORT_STAT_NO_EXIST) {
|
|
|
|
return stat;
|
2014-02-04 22:47:06 +00:00
|
|
|
}
|
|
|
|
}
|
2020-02-26 04:31:13 +00:00
|
|
|
|
2023-05-10 04:02:09 +01:00
|
|
|
// sys.path was empty or no matches, do not search the filesystem or
|
|
|
|
// frozen code.
|
|
|
|
return MP_IMPORT_STAT_NO_EXIST;
|
|
|
|
|
|
|
|
#else
|
|
|
|
|
|
|
|
// mp_sys_path is not enabled, so just stat the given path directly.
|
2020-02-26 04:24:09 +00:00
|
|
|
vstr_add_str(dest, qstr_str(mod_name));
|
2020-02-26 04:31:13 +00:00
|
|
|
return stat_dir_or_file(dest);
|
2023-05-10 04:02:09 +01:00
|
|
|
|
|
|
|
#endif
|
2014-02-05 23:57:48 +00:00
|
|
|
}
|
|
|
|
|
2018-12-13 12:10:35 +00:00
|
|
|
#if MICROPY_MODULE_FROZEN_STR || MICROPY_ENABLE_COMPILER
|
py: Rework bytecode and .mpy file format to be mostly static data.
Background: .mpy files are precompiled .py files, built using mpy-cross,
that contain compiled bytecode functions (and can also contain machine
code). The benefit of using an .mpy file over a .py file is that they are
faster to import and take less memory when importing. They are also
smaller on disk.
But the real benefit of .mpy files comes when they are frozen into the
firmware. This is done by loading the .mpy file during compilation of the
firmware and turning it into a set of big C data structures (the job of
mpy-tool.py), which are then compiled and downloaded into the ROM of a
device. These C data structures can be executed in-place, ie directly from
ROM. This makes importing even faster because there is very little to do,
and also means such frozen modules take up much less RAM (because their
bytecode stays in ROM).
The downside of frozen code is that it requires recompiling and reflashing
the entire firmware. This can be a big barrier to entry, slows down
development time, and makes it harder to do OTA updates of frozen code
(because the whole firmware must be updated).
This commit attempts to solve this problem by providing a solution that
sits between loading .mpy files into RAM and freezing them into the
firmware. The .mpy file format has been reworked so that it consists of
data and bytecode which is mostly static and ready to run in-place. If
these new .mpy files are located in flash/ROM which is memory addressable,
the .mpy file can be executed (mostly) in-place.
With this approach there is still a small amount of unpacking and linking
of the .mpy file that needs to be done when it's imported, but it's still
much better than loading an .mpy from disk into RAM (although not as good
as freezing .mpy files into the firmware).
The main trick to make static .mpy files is to adjust the bytecode so any
qstrs that it references now go through a lookup table to convert from
local qstr number in the module to global qstr number in the firmware.
That means the bytecode does not need linking/rewriting of qstrs when it's
loaded. Instead only a small qstr table needs to be built (and put in RAM)
at import time. This means the bytecode itself is static/constant and can
be used directly if it's in addressable memory. Also the qstr string data
in the .mpy file, and some constant object data, can be used directly.
Note that the qstr table is global to the module (ie not per function).
In more detail, in the VM what used to be (schematically):
qst = DECODE_QSTR_VALUE;
is now (schematically):
idx = DECODE_QSTR_INDEX;
qst = qstr_table[idx];
That allows the bytecode to be fixed at compile time and not need
relinking/rewriting of the qstr values. Only qstr_table needs to be linked
when the .mpy is loaded.
Incidentally, this helps to reduce the size of bytecode because what used
to be 2-byte qstr values in the bytecode are now (mostly) 1-byte indices.
If the module uses the same qstr more than two times then the bytecode is
smaller than before.
The following changes are measured for this commit compared to the
previous (the baseline):
- average 7%-9% reduction in size of .mpy files
- frozen code size is reduced by about 5%-7%
- importing .py files uses about 5% less RAM in total
- importing .mpy files uses about 4% less RAM in total
- importing .py and .mpy files takes about the same time as before
The qstr indirection in the bytecode has only a small impact on VM
performance. For stm32 on PYBv1.0 the performance change of this commit
is:
diff of scores (higher is better)
N=100 M=100 baseline -> this-commit diff diff% (error%)
bm_chaos.py 371.07 -> 357.39 : -13.68 = -3.687% (+/-0.02%)
bm_fannkuch.py 78.72 -> 77.49 : -1.23 = -1.563% (+/-0.01%)
bm_fft.py 2591.73 -> 2539.28 : -52.45 = -2.024% (+/-0.00%)
bm_float.py 6034.93 -> 5908.30 : -126.63 = -2.098% (+/-0.01%)
bm_hexiom.py 48.96 -> 47.93 : -1.03 = -2.104% (+/-0.00%)
bm_nqueens.py 4510.63 -> 4459.94 : -50.69 = -1.124% (+/-0.00%)
bm_pidigits.py 650.28 -> 644.96 : -5.32 = -0.818% (+/-0.23%)
core_import_mpy_multi.py 564.77 -> 581.49 : +16.72 = +2.960% (+/-0.01%)
core_import_mpy_single.py 68.67 -> 67.16 : -1.51 = -2.199% (+/-0.01%)
core_qstr.py 64.16 -> 64.12 : -0.04 = -0.062% (+/-0.00%)
core_yield_from.py 362.58 -> 354.50 : -8.08 = -2.228% (+/-0.00%)
misc_aes.py 429.69 -> 405.59 : -24.10 = -5.609% (+/-0.01%)
misc_mandel.py 3485.13 -> 3416.51 : -68.62 = -1.969% (+/-0.00%)
misc_pystone.py 2496.53 -> 2405.56 : -90.97 = -3.644% (+/-0.01%)
misc_raytrace.py 381.47 -> 374.01 : -7.46 = -1.956% (+/-0.01%)
viper_call0.py 576.73 -> 572.49 : -4.24 = -0.735% (+/-0.04%)
viper_call1a.py 550.37 -> 546.21 : -4.16 = -0.756% (+/-0.09%)
viper_call1b.py 438.23 -> 435.68 : -2.55 = -0.582% (+/-0.06%)
viper_call1c.py 442.84 -> 440.04 : -2.80 = -0.632% (+/-0.08%)
viper_call2a.py 536.31 -> 532.35 : -3.96 = -0.738% (+/-0.06%)
viper_call2b.py 382.34 -> 377.07 : -5.27 = -1.378% (+/-0.03%)
And for unix on x64:
diff of scores (higher is better)
N=2000 M=2000 baseline -> this-commit diff diff% (error%)
bm_chaos.py 13594.20 -> 13073.84 : -520.36 = -3.828% (+/-5.44%)
bm_fannkuch.py 60.63 -> 59.58 : -1.05 = -1.732% (+/-3.01%)
bm_fft.py 112009.15 -> 111603.32 : -405.83 = -0.362% (+/-4.03%)
bm_float.py 246202.55 -> 247923.81 : +1721.26 = +0.699% (+/-2.79%)
bm_hexiom.py 615.65 -> 617.21 : +1.56 = +0.253% (+/-1.64%)
bm_nqueens.py 215807.95 -> 215600.96 : -206.99 = -0.096% (+/-3.52%)
bm_pidigits.py 8246.74 -> 8422.82 : +176.08 = +2.135% (+/-3.64%)
misc_aes.py 16133.00 -> 16452.74 : +319.74 = +1.982% (+/-1.50%)
misc_mandel.py 128146.69 -> 130796.43 : +2649.74 = +2.068% (+/-3.18%)
misc_pystone.py 83811.49 -> 83124.85 : -686.64 = -0.819% (+/-1.03%)
misc_raytrace.py 21688.02 -> 21385.10 : -302.92 = -1.397% (+/-3.20%)
The code size change is (firmware with a lot of frozen code benefits the
most):
bare-arm: +396 +0.697%
minimal x86: +1595 +0.979% [incl +32(data)]
unix x64: +2408 +0.470% [incl +800(data)]
unix nanbox: +1396 +0.309% [incl -96(data)]
stm32: -1256 -0.318% PYBV10
cc3200: +288 +0.157%
esp8266: -260 -0.037% GENERIC
esp32: -216 -0.014% GENERIC[incl -1072(data)]
nrf: +116 +0.067% pca10040
rp2: -664 -0.135% PICO
samd: +844 +0.607% ADAFRUIT_ITSYBITSY_M4_EXPRESS
As part of this change the .mpy file format version is bumped to version 6.
And mpy-tool.py has been improved to provide a good visualisation of the
contents of .mpy files.
In summary: this commit changes the bytecode to use qstr indirection, and
reworks the .mpy file format to be simpler and allow .mpy files to be
executed in-place. Performance is not impacted too much. Eventually it
will be possible to store such .mpy files in a linear, read-only, memory-
mappable filesystem so they can be executed from flash/ROM. This will
essentially be able to replace frozen code for most applications.
Signed-off-by: Damien George <damien@micropython.org>
2021-10-22 12:22:47 +01:00
|
|
|
STATIC void do_load_from_lexer(mp_module_context_t *context, mp_lexer_t *lex) {
|
2014-07-25 09:00:15 +01:00
|
|
|
#if MICROPY_PY___FILE__
|
2014-12-05 19:35:18 +00:00
|
|
|
qstr source_name = lex->source_name;
|
py: Rework bytecode and .mpy file format to be mostly static data.
Background: .mpy files are precompiled .py files, built using mpy-cross,
that contain compiled bytecode functions (and can also contain machine
code). The benefit of using an .mpy file over a .py file is that they are
faster to import and take less memory when importing. They are also
smaller on disk.
But the real benefit of .mpy files comes when they are frozen into the
firmware. This is done by loading the .mpy file during compilation of the
firmware and turning it into a set of big C data structures (the job of
mpy-tool.py), which are then compiled and downloaded into the ROM of a
device. These C data structures can be executed in-place, ie directly from
ROM. This makes importing even faster because there is very little to do,
and also means such frozen modules take up much less RAM (because their
bytecode stays in ROM).
The downside of frozen code is that it requires recompiling and reflashing
the entire firmware. This can be a big barrier to entry, slows down
development time, and makes it harder to do OTA updates of frozen code
(because the whole firmware must be updated).
This commit attempts to solve this problem by providing a solution that
sits between loading .mpy files into RAM and freezing them into the
firmware. The .mpy file format has been reworked so that it consists of
data and bytecode which is mostly static and ready to run in-place. If
these new .mpy files are located in flash/ROM which is memory addressable,
the .mpy file can be executed (mostly) in-place.
With this approach there is still a small amount of unpacking and linking
of the .mpy file that needs to be done when it's imported, but it's still
much better than loading an .mpy from disk into RAM (although not as good
as freezing .mpy files into the firmware).
The main trick to make static .mpy files is to adjust the bytecode so any
qstrs that it references now go through a lookup table to convert from
local qstr number in the module to global qstr number in the firmware.
That means the bytecode does not need linking/rewriting of qstrs when it's
loaded. Instead only a small qstr table needs to be built (and put in RAM)
at import time. This means the bytecode itself is static/constant and can
be used directly if it's in addressable memory. Also the qstr string data
in the .mpy file, and some constant object data, can be used directly.
Note that the qstr table is global to the module (ie not per function).
In more detail, in the VM what used to be (schematically):
qst = DECODE_QSTR_VALUE;
is now (schematically):
idx = DECODE_QSTR_INDEX;
qst = qstr_table[idx];
That allows the bytecode to be fixed at compile time and not need
relinking/rewriting of the qstr values. Only qstr_table needs to be linked
when the .mpy is loaded.
Incidentally, this helps to reduce the size of bytecode because what used
to be 2-byte qstr values in the bytecode are now (mostly) 1-byte indices.
If the module uses the same qstr more than two times then the bytecode is
smaller than before.
The following changes are measured for this commit compared to the
previous (the baseline):
- average 7%-9% reduction in size of .mpy files
- frozen code size is reduced by about 5%-7%
- importing .py files uses about 5% less RAM in total
- importing .mpy files uses about 4% less RAM in total
- importing .py and .mpy files takes about the same time as before
The qstr indirection in the bytecode has only a small impact on VM
performance. For stm32 on PYBv1.0 the performance change of this commit
is:
diff of scores (higher is better)
N=100 M=100 baseline -> this-commit diff diff% (error%)
bm_chaos.py 371.07 -> 357.39 : -13.68 = -3.687% (+/-0.02%)
bm_fannkuch.py 78.72 -> 77.49 : -1.23 = -1.563% (+/-0.01%)
bm_fft.py 2591.73 -> 2539.28 : -52.45 = -2.024% (+/-0.00%)
bm_float.py 6034.93 -> 5908.30 : -126.63 = -2.098% (+/-0.01%)
bm_hexiom.py 48.96 -> 47.93 : -1.03 = -2.104% (+/-0.00%)
bm_nqueens.py 4510.63 -> 4459.94 : -50.69 = -1.124% (+/-0.00%)
bm_pidigits.py 650.28 -> 644.96 : -5.32 = -0.818% (+/-0.23%)
core_import_mpy_multi.py 564.77 -> 581.49 : +16.72 = +2.960% (+/-0.01%)
core_import_mpy_single.py 68.67 -> 67.16 : -1.51 = -2.199% (+/-0.01%)
core_qstr.py 64.16 -> 64.12 : -0.04 = -0.062% (+/-0.00%)
core_yield_from.py 362.58 -> 354.50 : -8.08 = -2.228% (+/-0.00%)
misc_aes.py 429.69 -> 405.59 : -24.10 = -5.609% (+/-0.01%)
misc_mandel.py 3485.13 -> 3416.51 : -68.62 = -1.969% (+/-0.00%)
misc_pystone.py 2496.53 -> 2405.56 : -90.97 = -3.644% (+/-0.01%)
misc_raytrace.py 381.47 -> 374.01 : -7.46 = -1.956% (+/-0.01%)
viper_call0.py 576.73 -> 572.49 : -4.24 = -0.735% (+/-0.04%)
viper_call1a.py 550.37 -> 546.21 : -4.16 = -0.756% (+/-0.09%)
viper_call1b.py 438.23 -> 435.68 : -2.55 = -0.582% (+/-0.06%)
viper_call1c.py 442.84 -> 440.04 : -2.80 = -0.632% (+/-0.08%)
viper_call2a.py 536.31 -> 532.35 : -3.96 = -0.738% (+/-0.06%)
viper_call2b.py 382.34 -> 377.07 : -5.27 = -1.378% (+/-0.03%)
And for unix on x64:
diff of scores (higher is better)
N=2000 M=2000 baseline -> this-commit diff diff% (error%)
bm_chaos.py 13594.20 -> 13073.84 : -520.36 = -3.828% (+/-5.44%)
bm_fannkuch.py 60.63 -> 59.58 : -1.05 = -1.732% (+/-3.01%)
bm_fft.py 112009.15 -> 111603.32 : -405.83 = -0.362% (+/-4.03%)
bm_float.py 246202.55 -> 247923.81 : +1721.26 = +0.699% (+/-2.79%)
bm_hexiom.py 615.65 -> 617.21 : +1.56 = +0.253% (+/-1.64%)
bm_nqueens.py 215807.95 -> 215600.96 : -206.99 = -0.096% (+/-3.52%)
bm_pidigits.py 8246.74 -> 8422.82 : +176.08 = +2.135% (+/-3.64%)
misc_aes.py 16133.00 -> 16452.74 : +319.74 = +1.982% (+/-1.50%)
misc_mandel.py 128146.69 -> 130796.43 : +2649.74 = +2.068% (+/-3.18%)
misc_pystone.py 83811.49 -> 83124.85 : -686.64 = -0.819% (+/-1.03%)
misc_raytrace.py 21688.02 -> 21385.10 : -302.92 = -1.397% (+/-3.20%)
The code size change is (firmware with a lot of frozen code benefits the
most):
bare-arm: +396 +0.697%
minimal x86: +1595 +0.979% [incl +32(data)]
unix x64: +2408 +0.470% [incl +800(data)]
unix nanbox: +1396 +0.309% [incl -96(data)]
stm32: -1256 -0.318% PYBV10
cc3200: +288 +0.157%
esp8266: -260 -0.037% GENERIC
esp32: -216 -0.014% GENERIC[incl -1072(data)]
nrf: +116 +0.067% pca10040
rp2: -664 -0.135% PICO
samd: +844 +0.607% ADAFRUIT_ITSYBITSY_M4_EXPRESS
As part of this change the .mpy file format version is bumped to version 6.
And mpy-tool.py has been improved to provide a good visualisation of the
contents of .mpy files.
In summary: this commit changes the bytecode to use qstr indirection, and
reworks the .mpy file format to be simpler and allow .mpy files to be
executed in-place. Performance is not impacted too much. Eventually it
will be possible to store such .mpy files in a linear, read-only, memory-
mappable filesystem so they can be executed from flash/ROM. This will
essentially be able to replace frozen code for most applications.
Signed-off-by: Damien George <damien@micropython.org>
2021-10-22 12:22:47 +01:00
|
|
|
mp_store_attr(MP_OBJ_FROM_PTR(&context->module), MP_QSTR___file__, MP_OBJ_NEW_QSTR(source_name));
|
2014-07-25 09:00:15 +01:00
|
|
|
#endif
|
2014-01-03 14:03:48 +00:00
|
|
|
|
2014-10-05 20:13:34 +01:00
|
|
|
// parse, compile and execute the module in its context
|
py: Rework bytecode and .mpy file format to be mostly static data.
Background: .mpy files are precompiled .py files, built using mpy-cross,
that contain compiled bytecode functions (and can also contain machine
code). The benefit of using an .mpy file over a .py file is that they are
faster to import and take less memory when importing. They are also
smaller on disk.
But the real benefit of .mpy files comes when they are frozen into the
firmware. This is done by loading the .mpy file during compilation of the
firmware and turning it into a set of big C data structures (the job of
mpy-tool.py), which are then compiled and downloaded into the ROM of a
device. These C data structures can be executed in-place, ie directly from
ROM. This makes importing even faster because there is very little to do,
and also means such frozen modules take up much less RAM (because their
bytecode stays in ROM).
The downside of frozen code is that it requires recompiling and reflashing
the entire firmware. This can be a big barrier to entry, slows down
development time, and makes it harder to do OTA updates of frozen code
(because the whole firmware must be updated).
This commit attempts to solve this problem by providing a solution that
sits between loading .mpy files into RAM and freezing them into the
firmware. The .mpy file format has been reworked so that it consists of
data and bytecode which is mostly static and ready to run in-place. If
these new .mpy files are located in flash/ROM which is memory addressable,
the .mpy file can be executed (mostly) in-place.
With this approach there is still a small amount of unpacking and linking
of the .mpy file that needs to be done when it's imported, but it's still
much better than loading an .mpy from disk into RAM (although not as good
as freezing .mpy files into the firmware).
The main trick to make static .mpy files is to adjust the bytecode so any
qstrs that it references now go through a lookup table to convert from
local qstr number in the module to global qstr number in the firmware.
That means the bytecode does not need linking/rewriting of qstrs when it's
loaded. Instead only a small qstr table needs to be built (and put in RAM)
at import time. This means the bytecode itself is static/constant and can
be used directly if it's in addressable memory. Also the qstr string data
in the .mpy file, and some constant object data, can be used directly.
Note that the qstr table is global to the module (ie not per function).
In more detail, in the VM what used to be (schematically):
qst = DECODE_QSTR_VALUE;
is now (schematically):
idx = DECODE_QSTR_INDEX;
qst = qstr_table[idx];
That allows the bytecode to be fixed at compile time and not need
relinking/rewriting of the qstr values. Only qstr_table needs to be linked
when the .mpy is loaded.
Incidentally, this helps to reduce the size of bytecode because what used
to be 2-byte qstr values in the bytecode are now (mostly) 1-byte indices.
If the module uses the same qstr more than two times then the bytecode is
smaller than before.
The following changes are measured for this commit compared to the
previous (the baseline):
- average 7%-9% reduction in size of .mpy files
- frozen code size is reduced by about 5%-7%
- importing .py files uses about 5% less RAM in total
- importing .mpy files uses about 4% less RAM in total
- importing .py and .mpy files takes about the same time as before
The qstr indirection in the bytecode has only a small impact on VM
performance. For stm32 on PYBv1.0 the performance change of this commit
is:
diff of scores (higher is better)
N=100 M=100 baseline -> this-commit diff diff% (error%)
bm_chaos.py 371.07 -> 357.39 : -13.68 = -3.687% (+/-0.02%)
bm_fannkuch.py 78.72 -> 77.49 : -1.23 = -1.563% (+/-0.01%)
bm_fft.py 2591.73 -> 2539.28 : -52.45 = -2.024% (+/-0.00%)
bm_float.py 6034.93 -> 5908.30 : -126.63 = -2.098% (+/-0.01%)
bm_hexiom.py 48.96 -> 47.93 : -1.03 = -2.104% (+/-0.00%)
bm_nqueens.py 4510.63 -> 4459.94 : -50.69 = -1.124% (+/-0.00%)
bm_pidigits.py 650.28 -> 644.96 : -5.32 = -0.818% (+/-0.23%)
core_import_mpy_multi.py 564.77 -> 581.49 : +16.72 = +2.960% (+/-0.01%)
core_import_mpy_single.py 68.67 -> 67.16 : -1.51 = -2.199% (+/-0.01%)
core_qstr.py 64.16 -> 64.12 : -0.04 = -0.062% (+/-0.00%)
core_yield_from.py 362.58 -> 354.50 : -8.08 = -2.228% (+/-0.00%)
misc_aes.py 429.69 -> 405.59 : -24.10 = -5.609% (+/-0.01%)
misc_mandel.py 3485.13 -> 3416.51 : -68.62 = -1.969% (+/-0.00%)
misc_pystone.py 2496.53 -> 2405.56 : -90.97 = -3.644% (+/-0.01%)
misc_raytrace.py 381.47 -> 374.01 : -7.46 = -1.956% (+/-0.01%)
viper_call0.py 576.73 -> 572.49 : -4.24 = -0.735% (+/-0.04%)
viper_call1a.py 550.37 -> 546.21 : -4.16 = -0.756% (+/-0.09%)
viper_call1b.py 438.23 -> 435.68 : -2.55 = -0.582% (+/-0.06%)
viper_call1c.py 442.84 -> 440.04 : -2.80 = -0.632% (+/-0.08%)
viper_call2a.py 536.31 -> 532.35 : -3.96 = -0.738% (+/-0.06%)
viper_call2b.py 382.34 -> 377.07 : -5.27 = -1.378% (+/-0.03%)
And for unix on x64:
diff of scores (higher is better)
N=2000 M=2000 baseline -> this-commit diff diff% (error%)
bm_chaos.py 13594.20 -> 13073.84 : -520.36 = -3.828% (+/-5.44%)
bm_fannkuch.py 60.63 -> 59.58 : -1.05 = -1.732% (+/-3.01%)
bm_fft.py 112009.15 -> 111603.32 : -405.83 = -0.362% (+/-4.03%)
bm_float.py 246202.55 -> 247923.81 : +1721.26 = +0.699% (+/-2.79%)
bm_hexiom.py 615.65 -> 617.21 : +1.56 = +0.253% (+/-1.64%)
bm_nqueens.py 215807.95 -> 215600.96 : -206.99 = -0.096% (+/-3.52%)
bm_pidigits.py 8246.74 -> 8422.82 : +176.08 = +2.135% (+/-3.64%)
misc_aes.py 16133.00 -> 16452.74 : +319.74 = +1.982% (+/-1.50%)
misc_mandel.py 128146.69 -> 130796.43 : +2649.74 = +2.068% (+/-3.18%)
misc_pystone.py 83811.49 -> 83124.85 : -686.64 = -0.819% (+/-1.03%)
misc_raytrace.py 21688.02 -> 21385.10 : -302.92 = -1.397% (+/-3.20%)
The code size change is (firmware with a lot of frozen code benefits the
most):
bare-arm: +396 +0.697%
minimal x86: +1595 +0.979% [incl +32(data)]
unix x64: +2408 +0.470% [incl +800(data)]
unix nanbox: +1396 +0.309% [incl -96(data)]
stm32: -1256 -0.318% PYBV10
cc3200: +288 +0.157%
esp8266: -260 -0.037% GENERIC
esp32: -216 -0.014% GENERIC[incl -1072(data)]
nrf: +116 +0.067% pca10040
rp2: -664 -0.135% PICO
samd: +844 +0.607% ADAFRUIT_ITSYBITSY_M4_EXPRESS
As part of this change the .mpy file format version is bumped to version 6.
And mpy-tool.py has been improved to provide a good visualisation of the
contents of .mpy files.
In summary: this commit changes the bytecode to use qstr indirection, and
reworks the .mpy file format to be simpler and allow .mpy files to be
executed in-place. Performance is not impacted too much. Eventually it
will be possible to store such .mpy files in a linear, read-only, memory-
mappable filesystem so they can be executed from flash/ROM. This will
essentially be able to replace frozen code for most applications.
Signed-off-by: Damien George <damien@micropython.org>
2021-10-22 12:22:47 +01:00
|
|
|
mp_obj_dict_t *mod_globals = context->module.globals;
|
2014-10-05 20:13:34 +01:00
|
|
|
mp_parse_compile_execute(lex, MP_PARSE_FILE_INPUT, mod_globals, mod_globals);
|
2014-02-05 23:57:48 +00:00
|
|
|
}
|
2015-12-18 12:35:44 +00:00
|
|
|
#endif
|
2014-02-05 23:57:48 +00:00
|
|
|
|
2021-07-30 16:26:17 +01:00
|
|
|
#if (MICROPY_HAS_FILE_READER && MICROPY_PERSISTENT_CODE_LOAD) || MICROPY_MODULE_FROZEN_MPY
|
2022-07-27 02:37:53 +01:00
|
|
|
STATIC void do_execute_raw_code(const mp_module_context_t *context, const mp_raw_code_t *rc, const char *source_name) {
|
2019-07-08 10:26:20 +01:00
|
|
|
(void)source_name;
|
|
|
|
|
2015-11-02 21:57:42 +00:00
|
|
|
#if MICROPY_PY___FILE__
|
py: Rework bytecode and .mpy file format to be mostly static data.
Background: .mpy files are precompiled .py files, built using mpy-cross,
that contain compiled bytecode functions (and can also contain machine
code). The benefit of using an .mpy file over a .py file is that they are
faster to import and take less memory when importing. They are also
smaller on disk.
But the real benefit of .mpy files comes when they are frozen into the
firmware. This is done by loading the .mpy file during compilation of the
firmware and turning it into a set of big C data structures (the job of
mpy-tool.py), which are then compiled and downloaded into the ROM of a
device. These C data structures can be executed in-place, ie directly from
ROM. This makes importing even faster because there is very little to do,
and also means such frozen modules take up much less RAM (because their
bytecode stays in ROM).
The downside of frozen code is that it requires recompiling and reflashing
the entire firmware. This can be a big barrier to entry, slows down
development time, and makes it harder to do OTA updates of frozen code
(because the whole firmware must be updated).
This commit attempts to solve this problem by providing a solution that
sits between loading .mpy files into RAM and freezing them into the
firmware. The .mpy file format has been reworked so that it consists of
data and bytecode which is mostly static and ready to run in-place. If
these new .mpy files are located in flash/ROM which is memory addressable,
the .mpy file can be executed (mostly) in-place.
With this approach there is still a small amount of unpacking and linking
of the .mpy file that needs to be done when it's imported, but it's still
much better than loading an .mpy from disk into RAM (although not as good
as freezing .mpy files into the firmware).
The main trick to make static .mpy files is to adjust the bytecode so any
qstrs that it references now go through a lookup table to convert from
local qstr number in the module to global qstr number in the firmware.
That means the bytecode does not need linking/rewriting of qstrs when it's
loaded. Instead only a small qstr table needs to be built (and put in RAM)
at import time. This means the bytecode itself is static/constant and can
be used directly if it's in addressable memory. Also the qstr string data
in the .mpy file, and some constant object data, can be used directly.
Note that the qstr table is global to the module (ie not per function).
In more detail, in the VM what used to be (schematically):
qst = DECODE_QSTR_VALUE;
is now (schematically):
idx = DECODE_QSTR_INDEX;
qst = qstr_table[idx];
That allows the bytecode to be fixed at compile time and not need
relinking/rewriting of the qstr values. Only qstr_table needs to be linked
when the .mpy is loaded.
Incidentally, this helps to reduce the size of bytecode because what used
to be 2-byte qstr values in the bytecode are now (mostly) 1-byte indices.
If the module uses the same qstr more than two times then the bytecode is
smaller than before.
The following changes are measured for this commit compared to the
previous (the baseline):
- average 7%-9% reduction in size of .mpy files
- frozen code size is reduced by about 5%-7%
- importing .py files uses about 5% less RAM in total
- importing .mpy files uses about 4% less RAM in total
- importing .py and .mpy files takes about the same time as before
The qstr indirection in the bytecode has only a small impact on VM
performance. For stm32 on PYBv1.0 the performance change of this commit
is:
diff of scores (higher is better)
N=100 M=100 baseline -> this-commit diff diff% (error%)
bm_chaos.py 371.07 -> 357.39 : -13.68 = -3.687% (+/-0.02%)
bm_fannkuch.py 78.72 -> 77.49 : -1.23 = -1.563% (+/-0.01%)
bm_fft.py 2591.73 -> 2539.28 : -52.45 = -2.024% (+/-0.00%)
bm_float.py 6034.93 -> 5908.30 : -126.63 = -2.098% (+/-0.01%)
bm_hexiom.py 48.96 -> 47.93 : -1.03 = -2.104% (+/-0.00%)
bm_nqueens.py 4510.63 -> 4459.94 : -50.69 = -1.124% (+/-0.00%)
bm_pidigits.py 650.28 -> 644.96 : -5.32 = -0.818% (+/-0.23%)
core_import_mpy_multi.py 564.77 -> 581.49 : +16.72 = +2.960% (+/-0.01%)
core_import_mpy_single.py 68.67 -> 67.16 : -1.51 = -2.199% (+/-0.01%)
core_qstr.py 64.16 -> 64.12 : -0.04 = -0.062% (+/-0.00%)
core_yield_from.py 362.58 -> 354.50 : -8.08 = -2.228% (+/-0.00%)
misc_aes.py 429.69 -> 405.59 : -24.10 = -5.609% (+/-0.01%)
misc_mandel.py 3485.13 -> 3416.51 : -68.62 = -1.969% (+/-0.00%)
misc_pystone.py 2496.53 -> 2405.56 : -90.97 = -3.644% (+/-0.01%)
misc_raytrace.py 381.47 -> 374.01 : -7.46 = -1.956% (+/-0.01%)
viper_call0.py 576.73 -> 572.49 : -4.24 = -0.735% (+/-0.04%)
viper_call1a.py 550.37 -> 546.21 : -4.16 = -0.756% (+/-0.09%)
viper_call1b.py 438.23 -> 435.68 : -2.55 = -0.582% (+/-0.06%)
viper_call1c.py 442.84 -> 440.04 : -2.80 = -0.632% (+/-0.08%)
viper_call2a.py 536.31 -> 532.35 : -3.96 = -0.738% (+/-0.06%)
viper_call2b.py 382.34 -> 377.07 : -5.27 = -1.378% (+/-0.03%)
And for unix on x64:
diff of scores (higher is better)
N=2000 M=2000 baseline -> this-commit diff diff% (error%)
bm_chaos.py 13594.20 -> 13073.84 : -520.36 = -3.828% (+/-5.44%)
bm_fannkuch.py 60.63 -> 59.58 : -1.05 = -1.732% (+/-3.01%)
bm_fft.py 112009.15 -> 111603.32 : -405.83 = -0.362% (+/-4.03%)
bm_float.py 246202.55 -> 247923.81 : +1721.26 = +0.699% (+/-2.79%)
bm_hexiom.py 615.65 -> 617.21 : +1.56 = +0.253% (+/-1.64%)
bm_nqueens.py 215807.95 -> 215600.96 : -206.99 = -0.096% (+/-3.52%)
bm_pidigits.py 8246.74 -> 8422.82 : +176.08 = +2.135% (+/-3.64%)
misc_aes.py 16133.00 -> 16452.74 : +319.74 = +1.982% (+/-1.50%)
misc_mandel.py 128146.69 -> 130796.43 : +2649.74 = +2.068% (+/-3.18%)
misc_pystone.py 83811.49 -> 83124.85 : -686.64 = -0.819% (+/-1.03%)
misc_raytrace.py 21688.02 -> 21385.10 : -302.92 = -1.397% (+/-3.20%)
The code size change is (firmware with a lot of frozen code benefits the
most):
bare-arm: +396 +0.697%
minimal x86: +1595 +0.979% [incl +32(data)]
unix x64: +2408 +0.470% [incl +800(data)]
unix nanbox: +1396 +0.309% [incl -96(data)]
stm32: -1256 -0.318% PYBV10
cc3200: +288 +0.157%
esp8266: -260 -0.037% GENERIC
esp32: -216 -0.014% GENERIC[incl -1072(data)]
nrf: +116 +0.067% pca10040
rp2: -664 -0.135% PICO
samd: +844 +0.607% ADAFRUIT_ITSYBITSY_M4_EXPRESS
As part of this change the .mpy file format version is bumped to version 6.
And mpy-tool.py has been improved to provide a good visualisation of the
contents of .mpy files.
In summary: this commit changes the bytecode to use qstr indirection, and
reworks the .mpy file format to be simpler and allow .mpy files to be
executed in-place. Performance is not impacted too much. Eventually it
will be possible to store such .mpy files in a linear, read-only, memory-
mappable filesystem so they can be executed from flash/ROM. This will
essentially be able to replace frozen code for most applications.
Signed-off-by: Damien George <damien@micropython.org>
2021-10-22 12:22:47 +01:00
|
|
|
mp_store_attr(MP_OBJ_FROM_PTR(&context->module), MP_QSTR___file__, MP_OBJ_NEW_QSTR(qstr_from_str(source_name)));
|
2015-11-02 21:57:42 +00:00
|
|
|
#endif
|
|
|
|
|
|
|
|
// execute the module in its context
|
py: Rework bytecode and .mpy file format to be mostly static data.
Background: .mpy files are precompiled .py files, built using mpy-cross,
that contain compiled bytecode functions (and can also contain machine
code). The benefit of using an .mpy file over a .py file is that they are
faster to import and take less memory when importing. They are also
smaller on disk.
But the real benefit of .mpy files comes when they are frozen into the
firmware. This is done by loading the .mpy file during compilation of the
firmware and turning it into a set of big C data structures (the job of
mpy-tool.py), which are then compiled and downloaded into the ROM of a
device. These C data structures can be executed in-place, ie directly from
ROM. This makes importing even faster because there is very little to do,
and also means such frozen modules take up much less RAM (because their
bytecode stays in ROM).
The downside of frozen code is that it requires recompiling and reflashing
the entire firmware. This can be a big barrier to entry, slows down
development time, and makes it harder to do OTA updates of frozen code
(because the whole firmware must be updated).
This commit attempts to solve this problem by providing a solution that
sits between loading .mpy files into RAM and freezing them into the
firmware. The .mpy file format has been reworked so that it consists of
data and bytecode which is mostly static and ready to run in-place. If
these new .mpy files are located in flash/ROM which is memory addressable,
the .mpy file can be executed (mostly) in-place.
With this approach there is still a small amount of unpacking and linking
of the .mpy file that needs to be done when it's imported, but it's still
much better than loading an .mpy from disk into RAM (although not as good
as freezing .mpy files into the firmware).
The main trick to make static .mpy files is to adjust the bytecode so any
qstrs that it references now go through a lookup table to convert from
local qstr number in the module to global qstr number in the firmware.
That means the bytecode does not need linking/rewriting of qstrs when it's
loaded. Instead only a small qstr table needs to be built (and put in RAM)
at import time. This means the bytecode itself is static/constant and can
be used directly if it's in addressable memory. Also the qstr string data
in the .mpy file, and some constant object data, can be used directly.
Note that the qstr table is global to the module (ie not per function).
In more detail, in the VM what used to be (schematically):
qst = DECODE_QSTR_VALUE;
is now (schematically):
idx = DECODE_QSTR_INDEX;
qst = qstr_table[idx];
That allows the bytecode to be fixed at compile time and not need
relinking/rewriting of the qstr values. Only qstr_table needs to be linked
when the .mpy is loaded.
Incidentally, this helps to reduce the size of bytecode because what used
to be 2-byte qstr values in the bytecode are now (mostly) 1-byte indices.
If the module uses the same qstr more than two times then the bytecode is
smaller than before.
The following changes are measured for this commit compared to the
previous (the baseline):
- average 7%-9% reduction in size of .mpy files
- frozen code size is reduced by about 5%-7%
- importing .py files uses about 5% less RAM in total
- importing .mpy files uses about 4% less RAM in total
- importing .py and .mpy files takes about the same time as before
The qstr indirection in the bytecode has only a small impact on VM
performance. For stm32 on PYBv1.0 the performance change of this commit
is:
diff of scores (higher is better)
N=100 M=100 baseline -> this-commit diff diff% (error%)
bm_chaos.py 371.07 -> 357.39 : -13.68 = -3.687% (+/-0.02%)
bm_fannkuch.py 78.72 -> 77.49 : -1.23 = -1.563% (+/-0.01%)
bm_fft.py 2591.73 -> 2539.28 : -52.45 = -2.024% (+/-0.00%)
bm_float.py 6034.93 -> 5908.30 : -126.63 = -2.098% (+/-0.01%)
bm_hexiom.py 48.96 -> 47.93 : -1.03 = -2.104% (+/-0.00%)
bm_nqueens.py 4510.63 -> 4459.94 : -50.69 = -1.124% (+/-0.00%)
bm_pidigits.py 650.28 -> 644.96 : -5.32 = -0.818% (+/-0.23%)
core_import_mpy_multi.py 564.77 -> 581.49 : +16.72 = +2.960% (+/-0.01%)
core_import_mpy_single.py 68.67 -> 67.16 : -1.51 = -2.199% (+/-0.01%)
core_qstr.py 64.16 -> 64.12 : -0.04 = -0.062% (+/-0.00%)
core_yield_from.py 362.58 -> 354.50 : -8.08 = -2.228% (+/-0.00%)
misc_aes.py 429.69 -> 405.59 : -24.10 = -5.609% (+/-0.01%)
misc_mandel.py 3485.13 -> 3416.51 : -68.62 = -1.969% (+/-0.00%)
misc_pystone.py 2496.53 -> 2405.56 : -90.97 = -3.644% (+/-0.01%)
misc_raytrace.py 381.47 -> 374.01 : -7.46 = -1.956% (+/-0.01%)
viper_call0.py 576.73 -> 572.49 : -4.24 = -0.735% (+/-0.04%)
viper_call1a.py 550.37 -> 546.21 : -4.16 = -0.756% (+/-0.09%)
viper_call1b.py 438.23 -> 435.68 : -2.55 = -0.582% (+/-0.06%)
viper_call1c.py 442.84 -> 440.04 : -2.80 = -0.632% (+/-0.08%)
viper_call2a.py 536.31 -> 532.35 : -3.96 = -0.738% (+/-0.06%)
viper_call2b.py 382.34 -> 377.07 : -5.27 = -1.378% (+/-0.03%)
And for unix on x64:
diff of scores (higher is better)
N=2000 M=2000 baseline -> this-commit diff diff% (error%)
bm_chaos.py 13594.20 -> 13073.84 : -520.36 = -3.828% (+/-5.44%)
bm_fannkuch.py 60.63 -> 59.58 : -1.05 = -1.732% (+/-3.01%)
bm_fft.py 112009.15 -> 111603.32 : -405.83 = -0.362% (+/-4.03%)
bm_float.py 246202.55 -> 247923.81 : +1721.26 = +0.699% (+/-2.79%)
bm_hexiom.py 615.65 -> 617.21 : +1.56 = +0.253% (+/-1.64%)
bm_nqueens.py 215807.95 -> 215600.96 : -206.99 = -0.096% (+/-3.52%)
bm_pidigits.py 8246.74 -> 8422.82 : +176.08 = +2.135% (+/-3.64%)
misc_aes.py 16133.00 -> 16452.74 : +319.74 = +1.982% (+/-1.50%)
misc_mandel.py 128146.69 -> 130796.43 : +2649.74 = +2.068% (+/-3.18%)
misc_pystone.py 83811.49 -> 83124.85 : -686.64 = -0.819% (+/-1.03%)
misc_raytrace.py 21688.02 -> 21385.10 : -302.92 = -1.397% (+/-3.20%)
The code size change is (firmware with a lot of frozen code benefits the
most):
bare-arm: +396 +0.697%
minimal x86: +1595 +0.979% [incl +32(data)]
unix x64: +2408 +0.470% [incl +800(data)]
unix nanbox: +1396 +0.309% [incl -96(data)]
stm32: -1256 -0.318% PYBV10
cc3200: +288 +0.157%
esp8266: -260 -0.037% GENERIC
esp32: -216 -0.014% GENERIC[incl -1072(data)]
nrf: +116 +0.067% pca10040
rp2: -664 -0.135% PICO
samd: +844 +0.607% ADAFRUIT_ITSYBITSY_M4_EXPRESS
As part of this change the .mpy file format version is bumped to version 6.
And mpy-tool.py has been improved to provide a good visualisation of the
contents of .mpy files.
In summary: this commit changes the bytecode to use qstr indirection, and
reworks the .mpy file format to be simpler and allow .mpy files to be
executed in-place. Performance is not impacted too much. Eventually it
will be possible to store such .mpy files in a linear, read-only, memory-
mappable filesystem so they can be executed from flash/ROM. This will
essentially be able to replace frozen code for most applications.
Signed-off-by: Damien George <damien@micropython.org>
2021-10-22 12:22:47 +01:00
|
|
|
mp_obj_dict_t *mod_globals = context->module.globals;
|
2015-11-02 21:57:42 +00:00
|
|
|
|
|
|
|
// save context
|
|
|
|
mp_obj_dict_t *volatile old_globals = mp_globals_get();
|
|
|
|
mp_obj_dict_t *volatile old_locals = mp_locals_get();
|
|
|
|
|
|
|
|
// set new context
|
|
|
|
mp_globals_set(mod_globals);
|
|
|
|
mp_locals_set(mod_globals);
|
|
|
|
|
|
|
|
nlr_buf_t nlr;
|
|
|
|
if (nlr_push(&nlr) == 0) {
|
2022-07-27 02:37:53 +01:00
|
|
|
mp_obj_t module_fun = mp_make_function_from_raw_code(rc, context, NULL);
|
2015-11-02 21:57:42 +00:00
|
|
|
mp_call_function_0(module_fun);
|
|
|
|
|
|
|
|
// finish nlr block, restore context
|
|
|
|
nlr_pop();
|
|
|
|
mp_globals_set(old_globals);
|
|
|
|
mp_locals_set(old_locals);
|
|
|
|
} else {
|
|
|
|
// exception; restore context and re-raise same exception
|
|
|
|
mp_globals_set(old_globals);
|
|
|
|
mp_locals_set(old_locals);
|
2015-11-27 17:01:44 +00:00
|
|
|
nlr_jump(nlr.ret_val);
|
2015-11-02 21:57:42 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
py: Rework bytecode and .mpy file format to be mostly static data.
Background: .mpy files are precompiled .py files, built using mpy-cross,
that contain compiled bytecode functions (and can also contain machine
code). The benefit of using an .mpy file over a .py file is that they are
faster to import and take less memory when importing. They are also
smaller on disk.
But the real benefit of .mpy files comes when they are frozen into the
firmware. This is done by loading the .mpy file during compilation of the
firmware and turning it into a set of big C data structures (the job of
mpy-tool.py), which are then compiled and downloaded into the ROM of a
device. These C data structures can be executed in-place, ie directly from
ROM. This makes importing even faster because there is very little to do,
and also means such frozen modules take up much less RAM (because their
bytecode stays in ROM).
The downside of frozen code is that it requires recompiling and reflashing
the entire firmware. This can be a big barrier to entry, slows down
development time, and makes it harder to do OTA updates of frozen code
(because the whole firmware must be updated).
This commit attempts to solve this problem by providing a solution that
sits between loading .mpy files into RAM and freezing them into the
firmware. The .mpy file format has been reworked so that it consists of
data and bytecode which is mostly static and ready to run in-place. If
these new .mpy files are located in flash/ROM which is memory addressable,
the .mpy file can be executed (mostly) in-place.
With this approach there is still a small amount of unpacking and linking
of the .mpy file that needs to be done when it's imported, but it's still
much better than loading an .mpy from disk into RAM (although not as good
as freezing .mpy files into the firmware).
The main trick to make static .mpy files is to adjust the bytecode so any
qstrs that it references now go through a lookup table to convert from
local qstr number in the module to global qstr number in the firmware.
That means the bytecode does not need linking/rewriting of qstrs when it's
loaded. Instead only a small qstr table needs to be built (and put in RAM)
at import time. This means the bytecode itself is static/constant and can
be used directly if it's in addressable memory. Also the qstr string data
in the .mpy file, and some constant object data, can be used directly.
Note that the qstr table is global to the module (ie not per function).
In more detail, in the VM what used to be (schematically):
qst = DECODE_QSTR_VALUE;
is now (schematically):
idx = DECODE_QSTR_INDEX;
qst = qstr_table[idx];
That allows the bytecode to be fixed at compile time and not need
relinking/rewriting of the qstr values. Only qstr_table needs to be linked
when the .mpy is loaded.
Incidentally, this helps to reduce the size of bytecode because what used
to be 2-byte qstr values in the bytecode are now (mostly) 1-byte indices.
If the module uses the same qstr more than two times then the bytecode is
smaller than before.
The following changes are measured for this commit compared to the
previous (the baseline):
- average 7%-9% reduction in size of .mpy files
- frozen code size is reduced by about 5%-7%
- importing .py files uses about 5% less RAM in total
- importing .mpy files uses about 4% less RAM in total
- importing .py and .mpy files takes about the same time as before
The qstr indirection in the bytecode has only a small impact on VM
performance. For stm32 on PYBv1.0 the performance change of this commit
is:
diff of scores (higher is better)
N=100 M=100 baseline -> this-commit diff diff% (error%)
bm_chaos.py 371.07 -> 357.39 : -13.68 = -3.687% (+/-0.02%)
bm_fannkuch.py 78.72 -> 77.49 : -1.23 = -1.563% (+/-0.01%)
bm_fft.py 2591.73 -> 2539.28 : -52.45 = -2.024% (+/-0.00%)
bm_float.py 6034.93 -> 5908.30 : -126.63 = -2.098% (+/-0.01%)
bm_hexiom.py 48.96 -> 47.93 : -1.03 = -2.104% (+/-0.00%)
bm_nqueens.py 4510.63 -> 4459.94 : -50.69 = -1.124% (+/-0.00%)
bm_pidigits.py 650.28 -> 644.96 : -5.32 = -0.818% (+/-0.23%)
core_import_mpy_multi.py 564.77 -> 581.49 : +16.72 = +2.960% (+/-0.01%)
core_import_mpy_single.py 68.67 -> 67.16 : -1.51 = -2.199% (+/-0.01%)
core_qstr.py 64.16 -> 64.12 : -0.04 = -0.062% (+/-0.00%)
core_yield_from.py 362.58 -> 354.50 : -8.08 = -2.228% (+/-0.00%)
misc_aes.py 429.69 -> 405.59 : -24.10 = -5.609% (+/-0.01%)
misc_mandel.py 3485.13 -> 3416.51 : -68.62 = -1.969% (+/-0.00%)
misc_pystone.py 2496.53 -> 2405.56 : -90.97 = -3.644% (+/-0.01%)
misc_raytrace.py 381.47 -> 374.01 : -7.46 = -1.956% (+/-0.01%)
viper_call0.py 576.73 -> 572.49 : -4.24 = -0.735% (+/-0.04%)
viper_call1a.py 550.37 -> 546.21 : -4.16 = -0.756% (+/-0.09%)
viper_call1b.py 438.23 -> 435.68 : -2.55 = -0.582% (+/-0.06%)
viper_call1c.py 442.84 -> 440.04 : -2.80 = -0.632% (+/-0.08%)
viper_call2a.py 536.31 -> 532.35 : -3.96 = -0.738% (+/-0.06%)
viper_call2b.py 382.34 -> 377.07 : -5.27 = -1.378% (+/-0.03%)
And for unix on x64:
diff of scores (higher is better)
N=2000 M=2000 baseline -> this-commit diff diff% (error%)
bm_chaos.py 13594.20 -> 13073.84 : -520.36 = -3.828% (+/-5.44%)
bm_fannkuch.py 60.63 -> 59.58 : -1.05 = -1.732% (+/-3.01%)
bm_fft.py 112009.15 -> 111603.32 : -405.83 = -0.362% (+/-4.03%)
bm_float.py 246202.55 -> 247923.81 : +1721.26 = +0.699% (+/-2.79%)
bm_hexiom.py 615.65 -> 617.21 : +1.56 = +0.253% (+/-1.64%)
bm_nqueens.py 215807.95 -> 215600.96 : -206.99 = -0.096% (+/-3.52%)
bm_pidigits.py 8246.74 -> 8422.82 : +176.08 = +2.135% (+/-3.64%)
misc_aes.py 16133.00 -> 16452.74 : +319.74 = +1.982% (+/-1.50%)
misc_mandel.py 128146.69 -> 130796.43 : +2649.74 = +2.068% (+/-3.18%)
misc_pystone.py 83811.49 -> 83124.85 : -686.64 = -0.819% (+/-1.03%)
misc_raytrace.py 21688.02 -> 21385.10 : -302.92 = -1.397% (+/-3.20%)
The code size change is (firmware with a lot of frozen code benefits the
most):
bare-arm: +396 +0.697%
minimal x86: +1595 +0.979% [incl +32(data)]
unix x64: +2408 +0.470% [incl +800(data)]
unix nanbox: +1396 +0.309% [incl -96(data)]
stm32: -1256 -0.318% PYBV10
cc3200: +288 +0.157%
esp8266: -260 -0.037% GENERIC
esp32: -216 -0.014% GENERIC[incl -1072(data)]
nrf: +116 +0.067% pca10040
rp2: -664 -0.135% PICO
samd: +844 +0.607% ADAFRUIT_ITSYBITSY_M4_EXPRESS
As part of this change the .mpy file format version is bumped to version 6.
And mpy-tool.py has been improved to provide a good visualisation of the
contents of .mpy files.
In summary: this commit changes the bytecode to use qstr indirection, and
reworks the .mpy file format to be simpler and allow .mpy files to be
executed in-place. Performance is not impacted too much. Eventually it
will be possible to store such .mpy files in a linear, read-only, memory-
mappable filesystem so they can be executed from flash/ROM. This will
essentially be able to replace frozen code for most applications.
Signed-off-by: Damien George <damien@micropython.org>
2021-10-22 12:22:47 +01:00
|
|
|
STATIC void do_load(mp_module_context_t *module_obj, vstr_t *file) {
|
2018-12-13 12:10:35 +00:00
|
|
|
#if MICROPY_MODULE_FROZEN || MICROPY_ENABLE_COMPILER || (MICROPY_PERSISTENT_CODE_LOAD && MICROPY_HAS_FILE_READER)
|
2021-12-11 11:40:21 +00:00
|
|
|
const char *file_str = vstr_null_terminated_str(file);
|
2016-01-31 22:24:16 +00:00
|
|
|
#endif
|
2015-12-18 12:35:44 +00:00
|
|
|
|
2016-05-23 12:42:23 +01:00
|
|
|
// If we support frozen modules (either as str or mpy) then try to find the
|
|
|
|
// requested filename in the list of frozen module filenames.
|
|
|
|
#if MICROPY_MODULE_FROZEN
|
|
|
|
void *modref;
|
2021-12-11 11:40:21 +00:00
|
|
|
int frozen_type;
|
|
|
|
const int frozen_path_prefix_len = strlen(MP_FROZEN_PATH_PREFIX);
|
|
|
|
if (strncmp(file_str, MP_FROZEN_PATH_PREFIX, frozen_path_prefix_len) == 0) {
|
|
|
|
mp_find_frozen_module(file_str + frozen_path_prefix_len, &frozen_type, &modref);
|
|
|
|
|
|
|
|
// If we support frozen str modules and the compiler is enabled, and we
|
|
|
|
// found the filename in the list of frozen files, then load and execute it.
|
|
|
|
#if MICROPY_MODULE_FROZEN_STR
|
|
|
|
if (frozen_type == MP_FROZEN_STR) {
|
|
|
|
do_load_from_lexer(module_obj, modref);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
|
|
|
// If we support frozen mpy modules and we found a corresponding file (and
|
|
|
|
// its data) in the list of frozen files, execute it.
|
|
|
|
#if MICROPY_MODULE_FROZEN_MPY
|
|
|
|
if (frozen_type == MP_FROZEN_MPY) {
|
py: Rework bytecode and .mpy file format to be mostly static data.
Background: .mpy files are precompiled .py files, built using mpy-cross,
that contain compiled bytecode functions (and can also contain machine
code). The benefit of using an .mpy file over a .py file is that they are
faster to import and take less memory when importing. They are also
smaller on disk.
But the real benefit of .mpy files comes when they are frozen into the
firmware. This is done by loading the .mpy file during compilation of the
firmware and turning it into a set of big C data structures (the job of
mpy-tool.py), which are then compiled and downloaded into the ROM of a
device. These C data structures can be executed in-place, ie directly from
ROM. This makes importing even faster because there is very little to do,
and also means such frozen modules take up much less RAM (because their
bytecode stays in ROM).
The downside of frozen code is that it requires recompiling and reflashing
the entire firmware. This can be a big barrier to entry, slows down
development time, and makes it harder to do OTA updates of frozen code
(because the whole firmware must be updated).
This commit attempts to solve this problem by providing a solution that
sits between loading .mpy files into RAM and freezing them into the
firmware. The .mpy file format has been reworked so that it consists of
data and bytecode which is mostly static and ready to run in-place. If
these new .mpy files are located in flash/ROM which is memory addressable,
the .mpy file can be executed (mostly) in-place.
With this approach there is still a small amount of unpacking and linking
of the .mpy file that needs to be done when it's imported, but it's still
much better than loading an .mpy from disk into RAM (although not as good
as freezing .mpy files into the firmware).
The main trick to make static .mpy files is to adjust the bytecode so any
qstrs that it references now go through a lookup table to convert from
local qstr number in the module to global qstr number in the firmware.
That means the bytecode does not need linking/rewriting of qstrs when it's
loaded. Instead only a small qstr table needs to be built (and put in RAM)
at import time. This means the bytecode itself is static/constant and can
be used directly if it's in addressable memory. Also the qstr string data
in the .mpy file, and some constant object data, can be used directly.
Note that the qstr table is global to the module (ie not per function).
In more detail, in the VM what used to be (schematically):
qst = DECODE_QSTR_VALUE;
is now (schematically):
idx = DECODE_QSTR_INDEX;
qst = qstr_table[idx];
That allows the bytecode to be fixed at compile time and not need
relinking/rewriting of the qstr values. Only qstr_table needs to be linked
when the .mpy is loaded.
Incidentally, this helps to reduce the size of bytecode because what used
to be 2-byte qstr values in the bytecode are now (mostly) 1-byte indices.
If the module uses the same qstr more than two times then the bytecode is
smaller than before.
The following changes are measured for this commit compared to the
previous (the baseline):
- average 7%-9% reduction in size of .mpy files
- frozen code size is reduced by about 5%-7%
- importing .py files uses about 5% less RAM in total
- importing .mpy files uses about 4% less RAM in total
- importing .py and .mpy files takes about the same time as before
The qstr indirection in the bytecode has only a small impact on VM
performance. For stm32 on PYBv1.0 the performance change of this commit
is:
diff of scores (higher is better)
N=100 M=100 baseline -> this-commit diff diff% (error%)
bm_chaos.py 371.07 -> 357.39 : -13.68 = -3.687% (+/-0.02%)
bm_fannkuch.py 78.72 -> 77.49 : -1.23 = -1.563% (+/-0.01%)
bm_fft.py 2591.73 -> 2539.28 : -52.45 = -2.024% (+/-0.00%)
bm_float.py 6034.93 -> 5908.30 : -126.63 = -2.098% (+/-0.01%)
bm_hexiom.py 48.96 -> 47.93 : -1.03 = -2.104% (+/-0.00%)
bm_nqueens.py 4510.63 -> 4459.94 : -50.69 = -1.124% (+/-0.00%)
bm_pidigits.py 650.28 -> 644.96 : -5.32 = -0.818% (+/-0.23%)
core_import_mpy_multi.py 564.77 -> 581.49 : +16.72 = +2.960% (+/-0.01%)
core_import_mpy_single.py 68.67 -> 67.16 : -1.51 = -2.199% (+/-0.01%)
core_qstr.py 64.16 -> 64.12 : -0.04 = -0.062% (+/-0.00%)
core_yield_from.py 362.58 -> 354.50 : -8.08 = -2.228% (+/-0.00%)
misc_aes.py 429.69 -> 405.59 : -24.10 = -5.609% (+/-0.01%)
misc_mandel.py 3485.13 -> 3416.51 : -68.62 = -1.969% (+/-0.00%)
misc_pystone.py 2496.53 -> 2405.56 : -90.97 = -3.644% (+/-0.01%)
misc_raytrace.py 381.47 -> 374.01 : -7.46 = -1.956% (+/-0.01%)
viper_call0.py 576.73 -> 572.49 : -4.24 = -0.735% (+/-0.04%)
viper_call1a.py 550.37 -> 546.21 : -4.16 = -0.756% (+/-0.09%)
viper_call1b.py 438.23 -> 435.68 : -2.55 = -0.582% (+/-0.06%)
viper_call1c.py 442.84 -> 440.04 : -2.80 = -0.632% (+/-0.08%)
viper_call2a.py 536.31 -> 532.35 : -3.96 = -0.738% (+/-0.06%)
viper_call2b.py 382.34 -> 377.07 : -5.27 = -1.378% (+/-0.03%)
And for unix on x64:
diff of scores (higher is better)
N=2000 M=2000 baseline -> this-commit diff diff% (error%)
bm_chaos.py 13594.20 -> 13073.84 : -520.36 = -3.828% (+/-5.44%)
bm_fannkuch.py 60.63 -> 59.58 : -1.05 = -1.732% (+/-3.01%)
bm_fft.py 112009.15 -> 111603.32 : -405.83 = -0.362% (+/-4.03%)
bm_float.py 246202.55 -> 247923.81 : +1721.26 = +0.699% (+/-2.79%)
bm_hexiom.py 615.65 -> 617.21 : +1.56 = +0.253% (+/-1.64%)
bm_nqueens.py 215807.95 -> 215600.96 : -206.99 = -0.096% (+/-3.52%)
bm_pidigits.py 8246.74 -> 8422.82 : +176.08 = +2.135% (+/-3.64%)
misc_aes.py 16133.00 -> 16452.74 : +319.74 = +1.982% (+/-1.50%)
misc_mandel.py 128146.69 -> 130796.43 : +2649.74 = +2.068% (+/-3.18%)
misc_pystone.py 83811.49 -> 83124.85 : -686.64 = -0.819% (+/-1.03%)
misc_raytrace.py 21688.02 -> 21385.10 : -302.92 = -1.397% (+/-3.20%)
The code size change is (firmware with a lot of frozen code benefits the
most):
bare-arm: +396 +0.697%
minimal x86: +1595 +0.979% [incl +32(data)]
unix x64: +2408 +0.470% [incl +800(data)]
unix nanbox: +1396 +0.309% [incl -96(data)]
stm32: -1256 -0.318% PYBV10
cc3200: +288 +0.157%
esp8266: -260 -0.037% GENERIC
esp32: -216 -0.014% GENERIC[incl -1072(data)]
nrf: +116 +0.067% pca10040
rp2: -664 -0.135% PICO
samd: +844 +0.607% ADAFRUIT_ITSYBITSY_M4_EXPRESS
As part of this change the .mpy file format version is bumped to version 6.
And mpy-tool.py has been improved to provide a good visualisation of the
contents of .mpy files.
In summary: this commit changes the bytecode to use qstr indirection, and
reworks the .mpy file format to be simpler and allow .mpy files to be
executed in-place. Performance is not impacted too much. Eventually it
will be possible to store such .mpy files in a linear, read-only, memory-
mappable filesystem so they can be executed from flash/ROM. This will
essentially be able to replace frozen code for most applications.
Signed-off-by: Damien George <damien@micropython.org>
2021-10-22 12:22:47 +01:00
|
|
|
const mp_frozen_module_t *frozen = modref;
|
|
|
|
module_obj->constants = frozen->constants;
|
2022-07-27 02:37:53 +01:00
|
|
|
do_execute_raw_code(module_obj, frozen->rc, file_str + frozen_path_prefix_len);
|
2021-12-11 11:40:21 +00:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
#endif
|
2016-05-23 12:42:23 +01:00
|
|
|
}
|
|
|
|
|
2020-02-26 04:24:09 +00:00
|
|
|
#endif // MICROPY_MODULE_FROZEN
|
|
|
|
|
2016-05-23 12:42:23 +01:00
|
|
|
// If we support loading .mpy files then check if the file extension is of
|
|
|
|
// the correct format and, if so, load and execute the file.
|
2018-12-13 12:10:35 +00:00
|
|
|
#if MICROPY_HAS_FILE_READER && MICROPY_PERSISTENT_CODE_LOAD
|
2015-11-02 21:57:42 +00:00
|
|
|
if (file_str[file->len - 3] == 'm') {
|
2022-12-07 03:42:35 +00:00
|
|
|
mp_compiled_module_t cm;
|
|
|
|
cm.context = module_obj;
|
|
|
|
mp_raw_code_load_file(file_str, &cm);
|
2022-07-27 02:37:53 +01:00
|
|
|
do_execute_raw_code(cm.context, cm.rc, file_str);
|
2015-12-18 12:35:44 +00:00
|
|
|
return;
|
|
|
|
}
|
2015-11-02 21:57:42 +00:00
|
|
|
#endif
|
2015-12-18 12:35:44 +00:00
|
|
|
|
2016-05-23 12:42:23 +01:00
|
|
|
// If we can compile scripts then load the file and compile and execute it.
|
2015-12-18 12:35:44 +00:00
|
|
|
#if MICROPY_ENABLE_COMPILER
|
2015-11-02 21:57:42 +00:00
|
|
|
{
|
2016-05-23 12:42:23 +01:00
|
|
|
mp_lexer_t *lex = mp_lexer_new_from_file(file_str);
|
2017-03-14 00:16:31 +00:00
|
|
|
do_load_from_lexer(module_obj, lex);
|
2016-05-23 12:42:23 +01:00
|
|
|
return;
|
2015-11-02 21:57:42 +00:00
|
|
|
}
|
2017-07-01 00:23:29 +01:00
|
|
|
#else
|
2016-05-23 12:42:23 +01:00
|
|
|
// If we get here then the file was not frozen and we can't compile scripts.
|
2020-03-02 11:35:22 +00:00
|
|
|
mp_raise_msg(&mp_type_ImportError, MP_ERROR_TEXT("script compilation not supported"));
|
2017-07-01 00:23:29 +01:00
|
|
|
#endif
|
2015-01-20 09:52:12 +00:00
|
|
|
}
|
|
|
|
|
2020-02-26 04:24:09 +00:00
|
|
|
// Convert a relative (to the current module) import, going up "level" levels,
|
|
|
|
// into an absolute import.
|
|
|
|
STATIC void evaluate_relative_import(mp_int_t level, const char **module_name, size_t *module_name_len) {
|
|
|
|
// What we want to do here is to take the name of the current module,
|
|
|
|
// remove <level> trailing components, and concatenate the passed-in
|
|
|
|
// module name.
|
|
|
|
// For example, level=3, module_name="foo.bar", __name__="a.b.c.d" --> "a.foo.bar"
|
|
|
|
// "Relative imports use a module's __name__ attribute to determine that
|
|
|
|
// module's position in the package hierarchy."
|
|
|
|
// http://legacy.python.org/dev/peps/pep-0328/#relative-imports-and-name
|
|
|
|
|
|
|
|
mp_obj_t current_module_name_obj = mp_obj_dict_get(MP_OBJ_FROM_PTR(mp_globals_get()), MP_OBJ_NEW_QSTR(MP_QSTR___name__));
|
|
|
|
assert(current_module_name_obj != MP_OBJ_NULL);
|
|
|
|
|
|
|
|
#if MICROPY_MODULE_OVERRIDE_MAIN_IMPORT && MICROPY_CPYTHON_COMPAT
|
|
|
|
if (MP_OBJ_QSTR_VALUE(current_module_name_obj) == MP_QSTR___main__) {
|
|
|
|
// This is a module loaded by -m command-line switch (e.g. unix port),
|
|
|
|
// and so its __name__ has been set to "__main__". Get its real name
|
|
|
|
// that we stored during import in the __main__ attribute.
|
|
|
|
current_module_name_obj = mp_obj_dict_get(MP_OBJ_FROM_PTR(mp_globals_get()), MP_OBJ_NEW_QSTR(MP_QSTR___main__));
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
|
|
|
// If we have a __path__ in the globals dict, then we're a package.
|
|
|
|
bool is_pkg = mp_map_lookup(&mp_globals_get()->map, MP_OBJ_NEW_QSTR(MP_QSTR___path__), MP_MAP_LOOKUP);
|
|
|
|
|
|
|
|
#if DEBUG_PRINT
|
|
|
|
DEBUG_printf("Current module/package: ");
|
|
|
|
mp_obj_print_helper(MICROPY_DEBUG_PRINTER, current_module_name_obj, PRINT_REPR);
|
|
|
|
DEBUG_printf(", is_package: %d", is_pkg);
|
|
|
|
DEBUG_printf("\n");
|
|
|
|
#endif
|
|
|
|
|
|
|
|
size_t current_module_name_len;
|
|
|
|
const char *current_module_name = mp_obj_str_get_data(current_module_name_obj, ¤t_module_name_len);
|
|
|
|
|
|
|
|
const char *p = current_module_name + current_module_name_len;
|
|
|
|
if (is_pkg) {
|
|
|
|
// If we're evaluating relative to a package, then take off one fewer
|
|
|
|
// level (i.e. the relative search starts inside the package, rather
|
|
|
|
// than as a sibling of the package).
|
|
|
|
--level;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Walk back 'level' dots (or run out of path).
|
|
|
|
while (level && p > current_module_name) {
|
2015-02-16 10:10:13 +00:00
|
|
|
if (*--p == '.') {
|
2020-02-26 04:24:09 +00:00
|
|
|
--level;
|
2015-02-16 10:10:13 +00:00
|
|
|
}
|
|
|
|
}
|
2020-02-26 04:24:09 +00:00
|
|
|
|
|
|
|
// We must have some component left over to import from.
|
|
|
|
if (p == current_module_name) {
|
|
|
|
mp_raise_msg(&mp_type_ImportError, MP_ERROR_TEXT("can't perform relative import"));
|
|
|
|
}
|
|
|
|
|
|
|
|
// New length is len("<chopped path>.<module_name>"). Note: might be one byte
|
|
|
|
// more than we need if module_name is empty (for the extra . we will
|
|
|
|
// append).
|
|
|
|
uint new_module_name_len = (size_t)(p - current_module_name) + 1 + *module_name_len;
|
|
|
|
char *new_mod = mp_local_alloc(new_module_name_len);
|
|
|
|
memcpy(new_mod, current_module_name, p - current_module_name);
|
|
|
|
|
|
|
|
// Only append ".<module_name>" if there was one).
|
|
|
|
if (*module_name_len != 0) {
|
|
|
|
new_mod[p - current_module_name] = '.';
|
|
|
|
memcpy(new_mod + (p - current_module_name) + 1, *module_name, *module_name_len);
|
|
|
|
} else {
|
|
|
|
--new_module_name_len;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Copy into a QSTR.
|
|
|
|
qstr new_mod_q = qstr_from_strn(new_mod, new_module_name_len);
|
|
|
|
mp_local_free(new_mod);
|
|
|
|
|
|
|
|
DEBUG_printf("Resolved base name for relative import: '%s'\n", qstr_str(new_mod_q));
|
|
|
|
*module_name = qstr_str(new_mod_q);
|
|
|
|
*module_name_len = new_module_name_len;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Load a module at the specified absolute path, possibly as a submodule of the given outer module.
|
|
|
|
// full_mod_name: The full absolute path to this module (e.g. "foo.bar.baz").
|
|
|
|
// level_mod_name: The final component of the path (e.g. "baz").
|
|
|
|
// outer_module_obj: The parent module (we need to store this module as an
|
|
|
|
// attribute on it) (or MP_OBJ_NULL for top-level).
|
|
|
|
// path: The filesystem path where we found the parent module
|
|
|
|
// (or empty for a top level module).
|
|
|
|
// override_main: Whether to set the __name__ to "__main__" (and use __main__
|
|
|
|
// for the actual path).
|
|
|
|
STATIC mp_obj_t process_import_at_level(qstr full_mod_name, qstr level_mod_name, mp_obj_t outer_module_obj, vstr_t *path, bool override_main) {
|
|
|
|
mp_import_stat_t stat = MP_IMPORT_STAT_NO_EXIST;
|
|
|
|
|
|
|
|
// Exact-match of built-in (or already-loaded) takes priority.
|
|
|
|
mp_obj_t module_obj = mp_module_get_loaded_or_builtin(full_mod_name);
|
|
|
|
|
|
|
|
// Even if we find the module, go through the motions of searching for it
|
|
|
|
// because we may actually be in the process of importing a sub-module.
|
|
|
|
// So we need to (re-)find the correct path to be finding the sub-module
|
|
|
|
// on the next iteration of process_import_at_level.
|
|
|
|
|
|
|
|
if (outer_module_obj == MP_OBJ_NULL) {
|
|
|
|
DEBUG_printf("Searching for top-level module\n");
|
|
|
|
|
|
|
|
// First module in the dotted-name; search for a directory or file
|
|
|
|
// relative to all the locations in sys.path.
|
|
|
|
stat = stat_top_level_dir_or_file(full_mod_name, path);
|
|
|
|
|
|
|
|
// If the module "foo" doesn't exist on the filesystem, and it's not a
|
|
|
|
// builtin, try and find "ufoo" as a built-in. (This feature was
|
|
|
|
// formerly known as "weak links").
|
|
|
|
#if MICROPY_MODULE_WEAK_LINKS
|
|
|
|
if (stat == MP_IMPORT_STAT_NO_EXIST && module_obj == MP_OBJ_NULL) {
|
2022-11-01 18:44:23 +00:00
|
|
|
qstr umodule_name = make_weak_link_name(path, level_mod_name);
|
2020-02-26 04:24:09 +00:00
|
|
|
module_obj = mp_module_get_builtin(umodule_name);
|
|
|
|
}
|
2022-03-25 00:57:59 +00:00
|
|
|
#elif MICROPY_PY_SYS
|
|
|
|
if (stat == MP_IMPORT_STAT_NO_EXIST && module_obj == MP_OBJ_NULL && level_mod_name == MP_QSTR_sys) {
|
|
|
|
module_obj = MP_OBJ_FROM_PTR(&mp_module_sys);
|
|
|
|
}
|
2020-02-26 04:24:09 +00:00
|
|
|
#endif
|
|
|
|
} else {
|
|
|
|
DEBUG_printf("Searching for sub-module\n");
|
|
|
|
|
|
|
|
// Add the current part of the module name to the path.
|
|
|
|
vstr_add_char(path, PATH_SEP_CHAR[0]);
|
|
|
|
vstr_add_str(path, qstr_str(level_mod_name));
|
|
|
|
|
|
|
|
// Because it's not top level, we already know which path the parent was found in.
|
|
|
|
stat = stat_dir_or_file(path);
|
|
|
|
}
|
|
|
|
DEBUG_printf("Current path: %.*s\n", (int)vstr_len(path), vstr_str(path));
|
|
|
|
|
|
|
|
if (module_obj == MP_OBJ_NULL) {
|
|
|
|
// Not a built-in and not already-loaded.
|
|
|
|
|
|
|
|
if (stat == MP_IMPORT_STAT_NO_EXIST) {
|
|
|
|
// And the file wasn't found -- fail.
|
|
|
|
#if MICROPY_ERROR_REPORTING <= MICROPY_ERROR_REPORTING_TERSE
|
|
|
|
mp_raise_msg(&mp_type_ImportError, MP_ERROR_TEXT("module not found"));
|
|
|
|
#else
|
|
|
|
mp_raise_msg_varg(&mp_type_ImportError, MP_ERROR_TEXT("no module named '%q'"), full_mod_name);
|
|
|
|
#endif
|
|
|
|
}
|
|
|
|
|
|
|
|
// Not a built-in but found on the filesystem, try and load it.
|
|
|
|
|
|
|
|
DEBUG_printf("Found path: %.*s\n", (int)vstr_len(path), vstr_str(path));
|
|
|
|
|
|
|
|
// Prepare for loading from the filesystem. Create a new shell module.
|
|
|
|
module_obj = mp_obj_new_module(full_mod_name);
|
|
|
|
|
|
|
|
#if MICROPY_MODULE_OVERRIDE_MAIN_IMPORT
|
|
|
|
// If this module is being loaded via -m on unix, then
|
|
|
|
// override __name__ to "__main__". Do this only for *modules*
|
|
|
|
// however - packages never have their names replaced, instead
|
|
|
|
// they're -m'ed using a special __main__ submodule in them. (This all
|
|
|
|
// apparently is done to not touch the package name itself, which is
|
|
|
|
// important for future imports).
|
|
|
|
if (override_main && stat != MP_IMPORT_STAT_DIR) {
|
|
|
|
mp_obj_module_t *o = MP_OBJ_TO_PTR(module_obj);
|
|
|
|
mp_obj_dict_store(MP_OBJ_FROM_PTR(o->globals), MP_OBJ_NEW_QSTR(MP_QSTR___name__), MP_OBJ_NEW_QSTR(MP_QSTR___main__));
|
|
|
|
#if MICROPY_CPYTHON_COMPAT
|
|
|
|
// Store module as "__main__" in the dictionary of loaded modules (returned by sys.modules).
|
|
|
|
mp_obj_dict_store(MP_OBJ_FROM_PTR(&MP_STATE_VM(mp_loaded_modules_dict)), MP_OBJ_NEW_QSTR(MP_QSTR___main__), module_obj);
|
|
|
|
// Store real name in "__main__" attribute. Need this for
|
|
|
|
// resolving relative imports later. "__main__ was chosen
|
|
|
|
// semi-randonly, to reuse existing qstr's.
|
|
|
|
mp_obj_dict_store(MP_OBJ_FROM_PTR(o->globals), MP_OBJ_NEW_QSTR(MP_QSTR___main__), MP_OBJ_NEW_QSTR(full_mod_name));
|
|
|
|
#endif
|
|
|
|
}
|
|
|
|
#endif // MICROPY_MODULE_OVERRIDE_MAIN_IMPORT
|
|
|
|
|
|
|
|
if (stat == MP_IMPORT_STAT_DIR) {
|
|
|
|
// Directory -- execute "path/__init__.py".
|
|
|
|
DEBUG_printf("%.*s is dir\n", (int)vstr_len(path), vstr_str(path));
|
|
|
|
// Store the __path__ attribute onto this module.
|
|
|
|
// https://docs.python.org/3/reference/import.html
|
|
|
|
// "Specifically, any module that contains a __path__ attribute is considered a package."
|
|
|
|
mp_store_attr(module_obj, MP_QSTR___path__, mp_obj_new_str(vstr_str(path), vstr_len(path)));
|
|
|
|
size_t orig_path_len = path->len;
|
|
|
|
vstr_add_str(path, PATH_SEP_CHAR "__init__.py");
|
|
|
|
if (stat_file_py_or_mpy(path) == MP_IMPORT_STAT_FILE) {
|
py: Rework bytecode and .mpy file format to be mostly static data.
Background: .mpy files are precompiled .py files, built using mpy-cross,
that contain compiled bytecode functions (and can also contain machine
code). The benefit of using an .mpy file over a .py file is that they are
faster to import and take less memory when importing. They are also
smaller on disk.
But the real benefit of .mpy files comes when they are frozen into the
firmware. This is done by loading the .mpy file during compilation of the
firmware and turning it into a set of big C data structures (the job of
mpy-tool.py), which are then compiled and downloaded into the ROM of a
device. These C data structures can be executed in-place, ie directly from
ROM. This makes importing even faster because there is very little to do,
and also means such frozen modules take up much less RAM (because their
bytecode stays in ROM).
The downside of frozen code is that it requires recompiling and reflashing
the entire firmware. This can be a big barrier to entry, slows down
development time, and makes it harder to do OTA updates of frozen code
(because the whole firmware must be updated).
This commit attempts to solve this problem by providing a solution that
sits between loading .mpy files into RAM and freezing them into the
firmware. The .mpy file format has been reworked so that it consists of
data and bytecode which is mostly static and ready to run in-place. If
these new .mpy files are located in flash/ROM which is memory addressable,
the .mpy file can be executed (mostly) in-place.
With this approach there is still a small amount of unpacking and linking
of the .mpy file that needs to be done when it's imported, but it's still
much better than loading an .mpy from disk into RAM (although not as good
as freezing .mpy files into the firmware).
The main trick to make static .mpy files is to adjust the bytecode so any
qstrs that it references now go through a lookup table to convert from
local qstr number in the module to global qstr number in the firmware.
That means the bytecode does not need linking/rewriting of qstrs when it's
loaded. Instead only a small qstr table needs to be built (and put in RAM)
at import time. This means the bytecode itself is static/constant and can
be used directly if it's in addressable memory. Also the qstr string data
in the .mpy file, and some constant object data, can be used directly.
Note that the qstr table is global to the module (ie not per function).
In more detail, in the VM what used to be (schematically):
qst = DECODE_QSTR_VALUE;
is now (schematically):
idx = DECODE_QSTR_INDEX;
qst = qstr_table[idx];
That allows the bytecode to be fixed at compile time and not need
relinking/rewriting of the qstr values. Only qstr_table needs to be linked
when the .mpy is loaded.
Incidentally, this helps to reduce the size of bytecode because what used
to be 2-byte qstr values in the bytecode are now (mostly) 1-byte indices.
If the module uses the same qstr more than two times then the bytecode is
smaller than before.
The following changes are measured for this commit compared to the
previous (the baseline):
- average 7%-9% reduction in size of .mpy files
- frozen code size is reduced by about 5%-7%
- importing .py files uses about 5% less RAM in total
- importing .mpy files uses about 4% less RAM in total
- importing .py and .mpy files takes about the same time as before
The qstr indirection in the bytecode has only a small impact on VM
performance. For stm32 on PYBv1.0 the performance change of this commit
is:
diff of scores (higher is better)
N=100 M=100 baseline -> this-commit diff diff% (error%)
bm_chaos.py 371.07 -> 357.39 : -13.68 = -3.687% (+/-0.02%)
bm_fannkuch.py 78.72 -> 77.49 : -1.23 = -1.563% (+/-0.01%)
bm_fft.py 2591.73 -> 2539.28 : -52.45 = -2.024% (+/-0.00%)
bm_float.py 6034.93 -> 5908.30 : -126.63 = -2.098% (+/-0.01%)
bm_hexiom.py 48.96 -> 47.93 : -1.03 = -2.104% (+/-0.00%)
bm_nqueens.py 4510.63 -> 4459.94 : -50.69 = -1.124% (+/-0.00%)
bm_pidigits.py 650.28 -> 644.96 : -5.32 = -0.818% (+/-0.23%)
core_import_mpy_multi.py 564.77 -> 581.49 : +16.72 = +2.960% (+/-0.01%)
core_import_mpy_single.py 68.67 -> 67.16 : -1.51 = -2.199% (+/-0.01%)
core_qstr.py 64.16 -> 64.12 : -0.04 = -0.062% (+/-0.00%)
core_yield_from.py 362.58 -> 354.50 : -8.08 = -2.228% (+/-0.00%)
misc_aes.py 429.69 -> 405.59 : -24.10 = -5.609% (+/-0.01%)
misc_mandel.py 3485.13 -> 3416.51 : -68.62 = -1.969% (+/-0.00%)
misc_pystone.py 2496.53 -> 2405.56 : -90.97 = -3.644% (+/-0.01%)
misc_raytrace.py 381.47 -> 374.01 : -7.46 = -1.956% (+/-0.01%)
viper_call0.py 576.73 -> 572.49 : -4.24 = -0.735% (+/-0.04%)
viper_call1a.py 550.37 -> 546.21 : -4.16 = -0.756% (+/-0.09%)
viper_call1b.py 438.23 -> 435.68 : -2.55 = -0.582% (+/-0.06%)
viper_call1c.py 442.84 -> 440.04 : -2.80 = -0.632% (+/-0.08%)
viper_call2a.py 536.31 -> 532.35 : -3.96 = -0.738% (+/-0.06%)
viper_call2b.py 382.34 -> 377.07 : -5.27 = -1.378% (+/-0.03%)
And for unix on x64:
diff of scores (higher is better)
N=2000 M=2000 baseline -> this-commit diff diff% (error%)
bm_chaos.py 13594.20 -> 13073.84 : -520.36 = -3.828% (+/-5.44%)
bm_fannkuch.py 60.63 -> 59.58 : -1.05 = -1.732% (+/-3.01%)
bm_fft.py 112009.15 -> 111603.32 : -405.83 = -0.362% (+/-4.03%)
bm_float.py 246202.55 -> 247923.81 : +1721.26 = +0.699% (+/-2.79%)
bm_hexiom.py 615.65 -> 617.21 : +1.56 = +0.253% (+/-1.64%)
bm_nqueens.py 215807.95 -> 215600.96 : -206.99 = -0.096% (+/-3.52%)
bm_pidigits.py 8246.74 -> 8422.82 : +176.08 = +2.135% (+/-3.64%)
misc_aes.py 16133.00 -> 16452.74 : +319.74 = +1.982% (+/-1.50%)
misc_mandel.py 128146.69 -> 130796.43 : +2649.74 = +2.068% (+/-3.18%)
misc_pystone.py 83811.49 -> 83124.85 : -686.64 = -0.819% (+/-1.03%)
misc_raytrace.py 21688.02 -> 21385.10 : -302.92 = -1.397% (+/-3.20%)
The code size change is (firmware with a lot of frozen code benefits the
most):
bare-arm: +396 +0.697%
minimal x86: +1595 +0.979% [incl +32(data)]
unix x64: +2408 +0.470% [incl +800(data)]
unix nanbox: +1396 +0.309% [incl -96(data)]
stm32: -1256 -0.318% PYBV10
cc3200: +288 +0.157%
esp8266: -260 -0.037% GENERIC
esp32: -216 -0.014% GENERIC[incl -1072(data)]
nrf: +116 +0.067% pca10040
rp2: -664 -0.135% PICO
samd: +844 +0.607% ADAFRUIT_ITSYBITSY_M4_EXPRESS
As part of this change the .mpy file format version is bumped to version 6.
And mpy-tool.py has been improved to provide a good visualisation of the
contents of .mpy files.
In summary: this commit changes the bytecode to use qstr indirection, and
reworks the .mpy file format to be simpler and allow .mpy files to be
executed in-place. Performance is not impacted too much. Eventually it
will be possible to store such .mpy files in a linear, read-only, memory-
mappable filesystem so they can be executed from flash/ROM. This will
essentially be able to replace frozen code for most applications.
Signed-off-by: Damien George <damien@micropython.org>
2021-10-22 12:22:47 +01:00
|
|
|
do_load(MP_OBJ_TO_PTR(module_obj), path);
|
2020-02-26 04:24:09 +00:00
|
|
|
} else {
|
|
|
|
// No-op. Nothing to load.
|
|
|
|
// mp_warning("%s is imported as namespace package", vstr_str(&path));
|
|
|
|
}
|
|
|
|
// Remove /__init__.py suffix.
|
|
|
|
path->len = orig_path_len;
|
|
|
|
} else { // MP_IMPORT_STAT_FILE
|
|
|
|
// File -- execute "path.(m)py".
|
py: Rework bytecode and .mpy file format to be mostly static data.
Background: .mpy files are precompiled .py files, built using mpy-cross,
that contain compiled bytecode functions (and can also contain machine
code). The benefit of using an .mpy file over a .py file is that they are
faster to import and take less memory when importing. They are also
smaller on disk.
But the real benefit of .mpy files comes when they are frozen into the
firmware. This is done by loading the .mpy file during compilation of the
firmware and turning it into a set of big C data structures (the job of
mpy-tool.py), which are then compiled and downloaded into the ROM of a
device. These C data structures can be executed in-place, ie directly from
ROM. This makes importing even faster because there is very little to do,
and also means such frozen modules take up much less RAM (because their
bytecode stays in ROM).
The downside of frozen code is that it requires recompiling and reflashing
the entire firmware. This can be a big barrier to entry, slows down
development time, and makes it harder to do OTA updates of frozen code
(because the whole firmware must be updated).
This commit attempts to solve this problem by providing a solution that
sits between loading .mpy files into RAM and freezing them into the
firmware. The .mpy file format has been reworked so that it consists of
data and bytecode which is mostly static and ready to run in-place. If
these new .mpy files are located in flash/ROM which is memory addressable,
the .mpy file can be executed (mostly) in-place.
With this approach there is still a small amount of unpacking and linking
of the .mpy file that needs to be done when it's imported, but it's still
much better than loading an .mpy from disk into RAM (although not as good
as freezing .mpy files into the firmware).
The main trick to make static .mpy files is to adjust the bytecode so any
qstrs that it references now go through a lookup table to convert from
local qstr number in the module to global qstr number in the firmware.
That means the bytecode does not need linking/rewriting of qstrs when it's
loaded. Instead only a small qstr table needs to be built (and put in RAM)
at import time. This means the bytecode itself is static/constant and can
be used directly if it's in addressable memory. Also the qstr string data
in the .mpy file, and some constant object data, can be used directly.
Note that the qstr table is global to the module (ie not per function).
In more detail, in the VM what used to be (schematically):
qst = DECODE_QSTR_VALUE;
is now (schematically):
idx = DECODE_QSTR_INDEX;
qst = qstr_table[idx];
That allows the bytecode to be fixed at compile time and not need
relinking/rewriting of the qstr values. Only qstr_table needs to be linked
when the .mpy is loaded.
Incidentally, this helps to reduce the size of bytecode because what used
to be 2-byte qstr values in the bytecode are now (mostly) 1-byte indices.
If the module uses the same qstr more than two times then the bytecode is
smaller than before.
The following changes are measured for this commit compared to the
previous (the baseline):
- average 7%-9% reduction in size of .mpy files
- frozen code size is reduced by about 5%-7%
- importing .py files uses about 5% less RAM in total
- importing .mpy files uses about 4% less RAM in total
- importing .py and .mpy files takes about the same time as before
The qstr indirection in the bytecode has only a small impact on VM
performance. For stm32 on PYBv1.0 the performance change of this commit
is:
diff of scores (higher is better)
N=100 M=100 baseline -> this-commit diff diff% (error%)
bm_chaos.py 371.07 -> 357.39 : -13.68 = -3.687% (+/-0.02%)
bm_fannkuch.py 78.72 -> 77.49 : -1.23 = -1.563% (+/-0.01%)
bm_fft.py 2591.73 -> 2539.28 : -52.45 = -2.024% (+/-0.00%)
bm_float.py 6034.93 -> 5908.30 : -126.63 = -2.098% (+/-0.01%)
bm_hexiom.py 48.96 -> 47.93 : -1.03 = -2.104% (+/-0.00%)
bm_nqueens.py 4510.63 -> 4459.94 : -50.69 = -1.124% (+/-0.00%)
bm_pidigits.py 650.28 -> 644.96 : -5.32 = -0.818% (+/-0.23%)
core_import_mpy_multi.py 564.77 -> 581.49 : +16.72 = +2.960% (+/-0.01%)
core_import_mpy_single.py 68.67 -> 67.16 : -1.51 = -2.199% (+/-0.01%)
core_qstr.py 64.16 -> 64.12 : -0.04 = -0.062% (+/-0.00%)
core_yield_from.py 362.58 -> 354.50 : -8.08 = -2.228% (+/-0.00%)
misc_aes.py 429.69 -> 405.59 : -24.10 = -5.609% (+/-0.01%)
misc_mandel.py 3485.13 -> 3416.51 : -68.62 = -1.969% (+/-0.00%)
misc_pystone.py 2496.53 -> 2405.56 : -90.97 = -3.644% (+/-0.01%)
misc_raytrace.py 381.47 -> 374.01 : -7.46 = -1.956% (+/-0.01%)
viper_call0.py 576.73 -> 572.49 : -4.24 = -0.735% (+/-0.04%)
viper_call1a.py 550.37 -> 546.21 : -4.16 = -0.756% (+/-0.09%)
viper_call1b.py 438.23 -> 435.68 : -2.55 = -0.582% (+/-0.06%)
viper_call1c.py 442.84 -> 440.04 : -2.80 = -0.632% (+/-0.08%)
viper_call2a.py 536.31 -> 532.35 : -3.96 = -0.738% (+/-0.06%)
viper_call2b.py 382.34 -> 377.07 : -5.27 = -1.378% (+/-0.03%)
And for unix on x64:
diff of scores (higher is better)
N=2000 M=2000 baseline -> this-commit diff diff% (error%)
bm_chaos.py 13594.20 -> 13073.84 : -520.36 = -3.828% (+/-5.44%)
bm_fannkuch.py 60.63 -> 59.58 : -1.05 = -1.732% (+/-3.01%)
bm_fft.py 112009.15 -> 111603.32 : -405.83 = -0.362% (+/-4.03%)
bm_float.py 246202.55 -> 247923.81 : +1721.26 = +0.699% (+/-2.79%)
bm_hexiom.py 615.65 -> 617.21 : +1.56 = +0.253% (+/-1.64%)
bm_nqueens.py 215807.95 -> 215600.96 : -206.99 = -0.096% (+/-3.52%)
bm_pidigits.py 8246.74 -> 8422.82 : +176.08 = +2.135% (+/-3.64%)
misc_aes.py 16133.00 -> 16452.74 : +319.74 = +1.982% (+/-1.50%)
misc_mandel.py 128146.69 -> 130796.43 : +2649.74 = +2.068% (+/-3.18%)
misc_pystone.py 83811.49 -> 83124.85 : -686.64 = -0.819% (+/-1.03%)
misc_raytrace.py 21688.02 -> 21385.10 : -302.92 = -1.397% (+/-3.20%)
The code size change is (firmware with a lot of frozen code benefits the
most):
bare-arm: +396 +0.697%
minimal x86: +1595 +0.979% [incl +32(data)]
unix x64: +2408 +0.470% [incl +800(data)]
unix nanbox: +1396 +0.309% [incl -96(data)]
stm32: -1256 -0.318% PYBV10
cc3200: +288 +0.157%
esp8266: -260 -0.037% GENERIC
esp32: -216 -0.014% GENERIC[incl -1072(data)]
nrf: +116 +0.067% pca10040
rp2: -664 -0.135% PICO
samd: +844 +0.607% ADAFRUIT_ITSYBITSY_M4_EXPRESS
As part of this change the .mpy file format version is bumped to version 6.
And mpy-tool.py has been improved to provide a good visualisation of the
contents of .mpy files.
In summary: this commit changes the bytecode to use qstr indirection, and
reworks the .mpy file format to be simpler and allow .mpy files to be
executed in-place. Performance is not impacted too much. Eventually it
will be possible to store such .mpy files in a linear, read-only, memory-
mappable filesystem so they can be executed from flash/ROM. This will
essentially be able to replace frozen code for most applications.
Signed-off-by: Damien George <damien@micropython.org>
2021-10-22 12:22:47 +01:00
|
|
|
do_load(MP_OBJ_TO_PTR(module_obj), path);
|
2020-02-26 04:24:09 +00:00
|
|
|
// Note: This should be the last component in the import path. If
|
|
|
|
// there are remaining components then it's an ImportError
|
|
|
|
// because the current path(the module that was just loaded) is
|
|
|
|
// not a package. This will be caught on the next iteration
|
|
|
|
// because the file will not exist.
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (outer_module_obj != MP_OBJ_NULL) {
|
|
|
|
// If it's a sub-module (not a built-in one), then make it available on
|
|
|
|
// the parent module.
|
|
|
|
mp_store_attr(outer_module_obj, level_mod_name, module_obj);
|
|
|
|
}
|
|
|
|
|
|
|
|
return module_obj;
|
2015-02-16 10:10:13 +00:00
|
|
|
}
|
|
|
|
|
2022-08-19 15:40:28 +01:00
|
|
|
mp_obj_t mp_builtin___import___default(size_t n_args, const mp_obj_t *args) {
|
2014-04-11 21:08:29 +01:00
|
|
|
#if DEBUG_PRINT
|
2014-09-23 10:59:05 +01:00
|
|
|
DEBUG_printf("__import__:\n");
|
2017-07-04 14:44:22 +01:00
|
|
|
for (size_t i = 0; i < n_args; i++) {
|
2014-09-23 10:59:05 +01:00
|
|
|
DEBUG_printf(" ");
|
2021-09-21 11:00:01 +01:00
|
|
|
mp_obj_print_helper(MICROPY_DEBUG_PRINTER, args[i], PRINT_REPR);
|
2014-09-23 10:59:05 +01:00
|
|
|
DEBUG_printf("\n");
|
2014-02-05 23:57:48 +00:00
|
|
|
}
|
2014-04-11 21:08:29 +01:00
|
|
|
#endif
|
2014-02-05 23:57:48 +00:00
|
|
|
|
2020-02-26 04:24:09 +00:00
|
|
|
// This is the import path, with any leading dots stripped.
|
|
|
|
// "import foo.bar" --> module_name="foo.bar"
|
|
|
|
// "from foo.bar import baz" --> module_name="foo.bar"
|
|
|
|
// "from . import foo" --> module_name=""
|
|
|
|
// "from ...foo.bar import baz" --> module_name="foo.bar"
|
|
|
|
mp_obj_t module_name_obj = args[0];
|
|
|
|
|
|
|
|
// These are the imported names.
|
|
|
|
// i.e. "from foo.bar import baz, zap" --> fromtuple=("baz", "zap",)
|
|
|
|
// Note: There's a special case on the Unix port, where this is set to mp_const_false which means that it's __main__.
|
2014-02-19 22:29:54 +00:00
|
|
|
mp_obj_t fromtuple = mp_const_none;
|
2020-02-26 04:24:09 +00:00
|
|
|
|
|
|
|
// Level is the number of leading dots in a relative import.
|
|
|
|
// i.e. "from . import foo" --> level=1
|
|
|
|
// i.e. "from ...foo.bar import baz" --> level=3
|
2014-10-03 18:44:14 +01:00
|
|
|
mp_int_t level = 0;
|
2020-02-26 04:24:09 +00:00
|
|
|
|
2014-02-19 22:29:54 +00:00
|
|
|
if (n_args >= 4) {
|
|
|
|
fromtuple = args[3];
|
2014-02-20 23:15:20 +00:00
|
|
|
if (n_args >= 5) {
|
|
|
|
level = MP_OBJ_SMALL_INT_VALUE(args[4]);
|
2017-01-16 05:54:56 +00:00
|
|
|
if (level < 0) {
|
|
|
|
mp_raise_ValueError(NULL);
|
|
|
|
}
|
2014-02-20 23:15:20 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2020-02-26 04:24:09 +00:00
|
|
|
size_t module_name_len;
|
|
|
|
const char *module_name = mp_obj_str_get_data(module_name_obj, &module_name_len);
|
2014-04-12 15:46:54 +01:00
|
|
|
|
2014-02-20 23:15:20 +00:00
|
|
|
if (level != 0) {
|
2020-02-26 04:24:09 +00:00
|
|
|
// Turn "foo.bar" into "<current module minus 3 components>.foo.bar".
|
|
|
|
evaluate_relative_import(level, &module_name, &module_name_len);
|
2014-04-12 15:46:54 +01:00
|
|
|
}
|
2014-02-16 00:53:44 +00:00
|
|
|
|
2020-02-26 04:24:09 +00:00
|
|
|
if (module_name_len == 0) {
|
2019-11-22 13:06:49 +00:00
|
|
|
mp_raise_ValueError(NULL);
|
|
|
|
}
|
|
|
|
|
2020-02-26 04:24:09 +00:00
|
|
|
DEBUG_printf("Starting module search for '%s'\n", module_name);
|
2014-02-05 23:57:48 +00:00
|
|
|
|
2014-05-21 20:32:59 +01:00
|
|
|
VSTR_FIXED(path, MICROPY_ALLOC_PATH_MAX)
|
2014-02-16 00:53:44 +00:00
|
|
|
mp_obj_t top_module_obj = MP_OBJ_NULL;
|
|
|
|
mp_obj_t outer_module_obj = MP_OBJ_NULL;
|
2014-02-05 23:57:48 +00:00
|
|
|
|
2020-02-26 04:24:09 +00:00
|
|
|
// Search for the end of each component.
|
|
|
|
size_t current_component_start = 0;
|
|
|
|
for (size_t i = 1; i <= module_name_len; i++) {
|
|
|
|
if (i == module_name_len || module_name[i] == '.') {
|
|
|
|
// The module name up to this depth (e.g. foo.bar.baz).
|
|
|
|
qstr full_mod_name = qstr_from_strn(module_name, i);
|
|
|
|
// The current level name (e.g. baz).
|
|
|
|
qstr level_mod_name = qstr_from_strn(module_name + current_component_start, i - current_component_start);
|
|
|
|
|
|
|
|
DEBUG_printf("Processing module: '%s' at level '%s'\n", qstr_str(full_mod_name), qstr_str(level_mod_name));
|
|
|
|
DEBUG_printf("Previous path: =%.*s=\n", (int)vstr_len(&path), vstr_str(&path));
|
|
|
|
|
|
|
|
#if MICROPY_MODULE_OVERRIDE_MAIN_IMPORT
|
|
|
|
// On unix, if this is being loaded via -m (magic mp_const_false),
|
|
|
|
// then handle that if it's the final component.
|
|
|
|
bool override_main = (i == module_name_len && fromtuple == mp_const_false);
|
|
|
|
#else
|
|
|
|
bool override_main = false;
|
|
|
|
#endif
|
|
|
|
|
|
|
|
// Import this module.
|
|
|
|
mp_obj_t module_obj = process_import_at_level(full_mod_name, level_mod_name, outer_module_obj, &path, override_main);
|
|
|
|
|
|
|
|
// Set this as the parent module, and remember the top-level module if it's the first.
|
2014-02-16 00:53:44 +00:00
|
|
|
outer_module_obj = module_obj;
|
|
|
|
if (top_module_obj == MP_OBJ_NULL) {
|
|
|
|
top_module_obj = module_obj;
|
|
|
|
}
|
2020-02-26 04:24:09 +00:00
|
|
|
|
|
|
|
current_component_start = i + 1;
|
2014-02-05 23:57:48 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2014-02-19 22:29:54 +00:00
|
|
|
if (fromtuple != mp_const_none) {
|
2020-02-26 04:24:09 +00:00
|
|
|
// If fromtuple is not empty, return leaf module
|
|
|
|
return outer_module_obj;
|
|
|
|
} else {
|
|
|
|
// Otherwise, we need to return top-level package
|
|
|
|
return top_module_obj;
|
2014-02-19 22:29:54 +00:00
|
|
|
}
|
2014-01-03 14:03:48 +00:00
|
|
|
}
|
2018-02-20 07:00:44 +00:00
|
|
|
|
|
|
|
#else // MICROPY_ENABLE_EXTERNAL_IMPORT
|
|
|
|
|
2023-03-09 14:52:51 +00:00
|
|
|
bool mp_obj_is_package(mp_obj_t module) {
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
2022-08-19 15:40:28 +01:00
|
|
|
mp_obj_t mp_builtin___import___default(size_t n_args, const mp_obj_t *args) {
|
2018-02-20 07:00:44 +00:00
|
|
|
// Check that it's not a relative import
|
|
|
|
if (n_args >= 5 && MP_OBJ_SMALL_INT_VALUE(args[4]) != 0) {
|
2020-03-02 11:35:22 +00:00
|
|
|
mp_raise_NotImplementedError(MP_ERROR_TEXT("relative import"));
|
2018-02-20 07:00:44 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
// Check if module already exists, and return it if it does
|
|
|
|
qstr module_name_qstr = mp_obj_str_get_qstr(args[0]);
|
2020-02-26 04:24:09 +00:00
|
|
|
mp_obj_t module_obj = mp_module_get_loaded_or_builtin(module_name_qstr);
|
2018-02-20 07:00:44 +00:00
|
|
|
if (module_obj != MP_OBJ_NULL) {
|
|
|
|
return module_obj;
|
|
|
|
}
|
|
|
|
|
|
|
|
#if MICROPY_MODULE_WEAK_LINKS
|
|
|
|
// Check if there is a weak link to this module
|
2022-11-01 18:44:23 +00:00
|
|
|
VSTR_FIXED(umodule_path, MICROPY_ALLOC_PATH_MAX);
|
|
|
|
qstr umodule_name_qstr = make_weak_link_name(&umodule_path, module_name_qstr);
|
2020-02-26 04:24:09 +00:00
|
|
|
module_obj = mp_module_get_loaded_or_builtin(umodule_name_qstr);
|
py: Automatically provide weak links from "foo" to "ufoo" module name.
This commit implements automatic module weak links for all built-in
modules, by searching for "ufoo" in the built-in module list if "foo"
cannot be found. This means that all modules named "ufoo" are always
available as "foo". Also, a port can no longer add any other weak links,
which makes strict the definition of a weak link.
It saves some code size (about 100-200 bytes) on ports that previously had
lots of weak links.
Some changes from the previous behaviour:
- It doesn't intern the non-u module names (eg "foo" is not interned),
which saves code size, but will mean that "import foo" creates a new qstr
(namely "foo") in RAM (unless the importing module is frozen).
- help('modules') no longer lists non-u module names, only the u-variants;
this reduces duplication in the help listing.
Weak links are effectively the same as having a set of symbolic links on
the filesystem that is searched last. So an "import foo" will search
built-in modules first, then all paths in sys.path, then weak links last,
importing "ufoo" if it exists. Thus a file called "foo.py" somewhere in
sys.path will still have precedence over the weak link of "foo" to "ufoo".
See issues: #1740, #4449, #5229, #5241.
2019-10-21 15:06:34 +01:00
|
|
|
if (module_obj != MP_OBJ_NULL) {
|
|
|
|
return module_obj;
|
2018-02-20 07:00:44 +00:00
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
|
|
|
// Couldn't find the module, so fail
|
2021-04-22 03:13:58 +01:00
|
|
|
#if MICROPY_ERROR_REPORTING <= MICROPY_ERROR_REPORTING_TERSE
|
2020-03-02 11:35:22 +00:00
|
|
|
mp_raise_msg(&mp_type_ImportError, MP_ERROR_TEXT("module not found"));
|
2019-09-26 13:52:04 +01:00
|
|
|
#else
|
2020-03-02 11:35:22 +00:00
|
|
|
mp_raise_msg_varg(&mp_type_ImportError, MP_ERROR_TEXT("no module named '%q'"), module_name_qstr);
|
2019-09-26 13:52:04 +01:00
|
|
|
#endif
|
2018-02-20 07:00:44 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
#endif // MICROPY_ENABLE_EXTERNAL_IMPORT
|
|
|
|
|
2014-02-03 22:46:17 +00:00
|
|
|
MP_DEFINE_CONST_FUN_OBJ_VAR_BETWEEN(mp_builtin___import___obj, 1, 5, mp_builtin___import__);
|