2014-05-03 18:27:38 -04:00
|
|
|
/*
|
2017-06-30 03:22:17 -04:00
|
|
|
* This file is part of the MicroPython project, http://micropython.org/
|
2014-05-03 18:27:38 -04:00
|
|
|
*
|
|
|
|
* The MIT License (MIT)
|
|
|
|
*
|
|
|
|
* Copyright (c) 2013, 2014 Damien P. George
|
|
|
|
*
|
|
|
|
* 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.
|
|
|
|
*/
|
|
|
|
|
2013-11-17 08:19:33 -05:00
|
|
|
#include <stdio.h>
|
|
|
|
#include <assert.h>
|
|
|
|
|
2015-01-01 15:27:54 -05:00
|
|
|
#include "py/bc0.h"
|
2022-03-15 20:39:27 -04:00
|
|
|
#include "py/emitglue.h"
|
2015-01-01 15:27:54 -05:00
|
|
|
|
2014-01-19 06:48:48 -05:00
|
|
|
#if MICROPY_DEBUG_PRINTERS
|
2014-01-07 10:20:33 -05:00
|
|
|
|
2014-02-20 19:31:05 -05:00
|
|
|
#define DECODE_UINT { \
|
2020-02-26 23:36:53 -05:00
|
|
|
unum = 0; \
|
|
|
|
do { \
|
|
|
|
unum = (unum << 7) + (*ip & 0x7f); \
|
|
|
|
} while ((*ip++ & 0x80) != 0); \
|
2014-02-20 19:31:05 -05:00
|
|
|
}
|
2022-03-15 18:37:58 -04:00
|
|
|
|
|
|
|
#define DECODE_ULABEL \
|
|
|
|
do { \
|
|
|
|
if (ip[0] & 0x80) { \
|
|
|
|
unum = ((ip[0] & 0x7f) | (ip[1] << 7)); \
|
|
|
|
ip += 2; \
|
|
|
|
} else { \
|
|
|
|
unum = ip[0]; \
|
|
|
|
ip += 1; \
|
|
|
|
} \
|
|
|
|
} while (0)
|
|
|
|
|
|
|
|
#define DECODE_SLABEL \
|
|
|
|
do { \
|
|
|
|
if (ip[0] & 0x80) { \
|
|
|
|
unum = ((ip[0] & 0x7f) | (ip[1] << 7)) - 0x4000; \
|
|
|
|
ip += 2; \
|
|
|
|
} else { \
|
|
|
|
unum = ip[0] - 0x40; \
|
|
|
|
ip += 1; \
|
|
|
|
} \
|
|
|
|
} while (0)
|
2015-11-02 12:27:18 -05:00
|
|
|
|
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 07:22:47 -04:00
|
|
|
#if MICROPY_EMIT_BYTECODE_USES_QSTR_TABLE
|
2015-11-02 12:27:18 -05:00
|
|
|
|
|
|
|
#define DECODE_QSTR \
|
2015-11-27 12:01:44 -05:00
|
|
|
DECODE_UINT; \
|
2022-03-15 20:39:27 -04:00
|
|
|
qst = qstr_table[unum]
|
2015-11-02 12:27:18 -05:00
|
|
|
|
|
|
|
#else
|
|
|
|
|
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 07:22:47 -04:00
|
|
|
#define DECODE_QSTR \
|
|
|
|
DECODE_UINT; \
|
|
|
|
qst = unum;
|
2013-11-17 08:19:33 -05:00
|
|
|
|
2015-11-02 12:27:18 -05:00
|
|
|
#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 07:22:47 -04:00
|
|
|
#define DECODE_PTR \
|
2022-03-15 20:39:27 -04:00
|
|
|
DECODE_UINT; \
|
|
|
|
unum = (mp_uint_t)(uintptr_t)child_table[unum]
|
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 07:22:47 -04:00
|
|
|
|
|
|
|
#define DECODE_OBJ \
|
|
|
|
DECODE_UINT; \
|
2022-03-15 20:39:27 -04:00
|
|
|
unum = (mp_uint_t)obj_table[unum]
|
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 07:22:47 -04:00
|
|
|
|
2022-03-15 20:39:27 -04:00
|
|
|
void mp_bytecode_print(const mp_print_t *print, const mp_raw_code_t *rc, const mp_module_constants_t *cm) {
|
|
|
|
const byte *ip_start = rc->fun_data;
|
|
|
|
const byte *ip = rc->fun_data;
|
2013-12-30 17:32:17 -05:00
|
|
|
|
2019-09-16 08:12:59 -04:00
|
|
|
// Decode prelude
|
|
|
|
MP_BC_PRELUDE_SIG_DECODE(ip);
|
2019-09-25 01:45:47 -04:00
|
|
|
MP_BC_PRELUDE_SIZE_DECODE(ip);
|
2014-04-13 09:51:56 -04:00
|
|
|
const byte *code_info = ip;
|
2014-01-18 18:24:36 -05:00
|
|
|
|
2014-09-04 09:44:01 -04:00
|
|
|
qstr block_name = mp_decode_uint(&code_info);
|
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 07:22:47 -04:00
|
|
|
#if MICROPY_EMIT_BYTECODE_USES_QSTR_TABLE
|
|
|
|
block_name = cm->qstr_table[block_name];
|
|
|
|
qstr source_file = cm->qstr_table[0];
|
|
|
|
#else
|
|
|
|
qstr source_file = cm->source_file;
|
2015-11-02 12:27:18 -05:00
|
|
|
#endif
|
2022-03-15 20:39:27 -04:00
|
|
|
mp_printf(print, "File %s, code block '%s' (descriptor: %p, bytecode @%p %u bytes)\n",
|
|
|
|
qstr_str(source_file), qstr_str(block_name), rc, ip_start, (unsigned)rc->fun_data_len);
|
2014-06-02 09:11:16 -04:00
|
|
|
|
2014-10-24 10:42:50 -04:00
|
|
|
// raw bytecode dump
|
2022-03-15 20:39:27 -04:00
|
|
|
size_t prelude_size = ip - ip_start + n_info + n_cell;
|
2021-12-15 00:54:47 -05:00
|
|
|
mp_printf(print, "Raw bytecode (code_info_size=%u, bytecode_size=%u):\n",
|
2022-03-15 20:39:27 -04:00
|
|
|
(unsigned)prelude_size, (unsigned)(rc->fun_data_len - prelude_size));
|
|
|
|
for (size_t i = 0; i < rc->fun_data_len; i++) {
|
2014-10-24 10:42:50 -04:00
|
|
|
if (i > 0 && i % 16 == 0) {
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "\n");
|
2014-10-24 10:42:50 -04:00
|
|
|
}
|
2022-03-15 20:39:27 -04:00
|
|
|
mp_printf(print, " %02x", ip_start[i]);
|
2014-10-24 10:42:50 -04:00
|
|
|
}
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "\n");
|
2014-10-24 10:42:50 -04:00
|
|
|
|
2014-10-25 10:07:02 -04:00
|
|
|
// bytecode prelude: arg names (as qstr objects)
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "arg names:");
|
2015-10-22 18:45:37 -04:00
|
|
|
for (mp_uint_t i = 0; i < n_pos_args + n_kwonly_args; i++) {
|
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 07:22:47 -04:00
|
|
|
qstr qst = mp_decode_uint(&code_info);
|
|
|
|
#if MICROPY_EMIT_BYTECODE_USES_QSTR_TABLE
|
|
|
|
qst = cm->qstr_table[qst];
|
|
|
|
#endif
|
|
|
|
mp_printf(print, " %s", qstr_str(qst));
|
2014-10-25 10:07:02 -04:00
|
|
|
}
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "\n");
|
2014-10-25 10:07:02 -04:00
|
|
|
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "(N_STATE %u)\n", (unsigned)n_state);
|
|
|
|
mp_printf(print, "(N_EXC_STACK %u)\n", (unsigned)n_exc_stack);
|
2015-03-18 13:47:47 -04:00
|
|
|
|
2019-09-25 01:45:47 -04:00
|
|
|
// skip over code_info
|
|
|
|
ip += n_info;
|
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 07:22:47 -04:00
|
|
|
const byte *line_info_top = ip;
|
2014-03-28 14:38:20 -04:00
|
|
|
|
|
|
|
// bytecode prelude: initialise closed over variables
|
2019-09-25 01:45:47 -04:00
|
|
|
for (size_t i = 0; i < n_cell; ++i) {
|
|
|
|
uint local_num = *ip++;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "(INIT_CELL %u)\n", local_num);
|
2013-12-30 17:32:17 -05:00
|
|
|
}
|
|
|
|
|
2014-04-13 09:51:56 -04:00
|
|
|
// print out line number info
|
|
|
|
{
|
2019-09-25 01:45:47 -04:00
|
|
|
mp_int_t bc = 0;
|
2014-07-03 08:25:24 -04:00
|
|
|
mp_uint_t source_line = 1;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, " bc=" INT_FMT " line=" UINT_FMT "\n", bc, source_line);
|
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 07:22:47 -04:00
|
|
|
for (const byte *ci = code_info; ci < line_info_top;) {
|
2014-07-31 12:12:01 -04:00
|
|
|
if ((ci[0] & 0x80) == 0) {
|
|
|
|
// 0b0LLBBBBB encoding
|
|
|
|
bc += ci[0] & 0x1f;
|
|
|
|
source_line += ci[0] >> 5;
|
|
|
|
ci += 1;
|
|
|
|
} else {
|
|
|
|
// 0b1LLLBBBB 0bLLLLLLLL encoding (l's LSB in second byte)
|
|
|
|
bc += ci[0] & 0xf;
|
|
|
|
source_line += ((ci[0] << 4) & 0x700) | ci[1];
|
|
|
|
ci += 2;
|
|
|
|
}
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, " bc=" INT_FMT " line=" UINT_FMT "\n", bc, source_line);
|
2014-04-13 09:51:56 -04:00
|
|
|
}
|
|
|
|
}
|
2022-03-15 20:39:27 -04:00
|
|
|
mp_bytecode_print2(print, ip, rc->fun_data_len - prelude_size, rc->children, cm);
|
2014-04-22 20:40:24 -04:00
|
|
|
}
|
2014-04-13 09:51:56 -04:00
|
|
|
|
2022-03-15 20:39:27 -04:00
|
|
|
const byte *mp_bytecode_print_str(const mp_print_t *print, const byte *ip_start, const byte *ip, mp_raw_code_t *const *child_table, const mp_module_constants_t *cm) {
|
|
|
|
#if MICROPY_EMIT_BYTECODE_USES_QSTR_TABLE
|
|
|
|
const qstr_short_t *qstr_table = cm->qstr_table;
|
|
|
|
#endif
|
|
|
|
const mp_obj_t *obj_table = cm->obj_table;
|
2014-07-03 08:25:24 -04:00
|
|
|
mp_uint_t unum;
|
2015-01-20 06:55:10 -05:00
|
|
|
qstr qst;
|
2014-12-26 22:00:08 -05:00
|
|
|
|
|
|
|
switch (*ip++) {
|
|
|
|
case MP_BC_LOAD_CONST_FALSE:
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "LOAD_CONST_FALSE");
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_LOAD_CONST_NONE:
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "LOAD_CONST_NONE");
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_LOAD_CONST_TRUE:
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "LOAD_CONST_TRUE");
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_LOAD_CONST_SMALL_INT: {
|
|
|
|
mp_int_t num = 0;
|
|
|
|
if ((ip[0] & 0x40) != 0) {
|
|
|
|
// Number is negative
|
|
|
|
num--;
|
2014-02-19 08:47:59 -05:00
|
|
|
}
|
2014-12-26 22:00:08 -05:00
|
|
|
do {
|
2021-06-08 08:45:56 -04:00
|
|
|
num = ((mp_uint_t)num << 7) | (*ip & 0x7f);
|
2014-12-26 22:00:08 -05:00
|
|
|
} while ((*ip++ & 0x80) != 0);
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "LOAD_CONST_SMALL_INT " INT_FMT, num);
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
}
|
2013-11-17 08:19:33 -05:00
|
|
|
|
2014-12-26 22:00:08 -05:00
|
|
|
case MP_BC_LOAD_CONST_STRING:
|
|
|
|
DECODE_QSTR;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "LOAD_CONST_STRING '%s'", qstr_str(qst));
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
2015-01-13 18:08:47 -05:00
|
|
|
case MP_BC_LOAD_CONST_OBJ:
|
2015-11-27 12:01:44 -05:00
|
|
|
DECODE_OBJ;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "LOAD_CONST_OBJ %p=", MP_OBJ_TO_PTR(unum));
|
|
|
|
mp_obj_print_helper(print, (mp_obj_t)unum, PRINT_REPR);
|
2015-01-13 18:08:47 -05:00
|
|
|
break;
|
|
|
|
|
2014-12-26 22:00:08 -05:00
|
|
|
case MP_BC_LOAD_NULL:
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "LOAD_NULL");
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_LOAD_FAST_N:
|
|
|
|
DECODE_UINT;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "LOAD_FAST_N " UINT_FMT, unum);
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_LOAD_DEREF:
|
|
|
|
DECODE_UINT;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "LOAD_DEREF " UINT_FMT, unum);
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_LOAD_NAME:
|
|
|
|
DECODE_QSTR;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "LOAD_NAME %s", qstr_str(qst));
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_LOAD_GLOBAL:
|
|
|
|
DECODE_QSTR;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "LOAD_GLOBAL %s", qstr_str(qst));
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_LOAD_ATTR:
|
|
|
|
DECODE_QSTR;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "LOAD_ATTR %s", qstr_str(qst));
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_LOAD_METHOD:
|
|
|
|
DECODE_QSTR;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "LOAD_METHOD %s", qstr_str(qst));
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
2017-04-18 19:45:59 -04:00
|
|
|
case MP_BC_LOAD_SUPER_METHOD:
|
|
|
|
DECODE_QSTR;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "LOAD_SUPER_METHOD %s", qstr_str(qst));
|
2017-04-18 19:45:59 -04:00
|
|
|
break;
|
|
|
|
|
2014-12-26 22:00:08 -05:00
|
|
|
case MP_BC_LOAD_BUILD_CLASS:
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "LOAD_BUILD_CLASS");
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_LOAD_SUBSCR:
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "LOAD_SUBSCR");
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_STORE_FAST_N:
|
|
|
|
DECODE_UINT;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "STORE_FAST_N " UINT_FMT, unum);
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_STORE_DEREF:
|
|
|
|
DECODE_UINT;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "STORE_DEREF " UINT_FMT, unum);
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_STORE_NAME:
|
|
|
|
DECODE_QSTR;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "STORE_NAME %s", qstr_str(qst));
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_STORE_GLOBAL:
|
|
|
|
DECODE_QSTR;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "STORE_GLOBAL %s", qstr_str(qst));
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_STORE_ATTR:
|
|
|
|
DECODE_QSTR;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "STORE_ATTR %s", qstr_str(qst));
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_STORE_SUBSCR:
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "STORE_SUBSCR");
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_DELETE_FAST:
|
|
|
|
DECODE_UINT;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "DELETE_FAST " UINT_FMT, unum);
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_DELETE_DEREF:
|
|
|
|
DECODE_UINT;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "DELETE_DEREF " UINT_FMT, unum);
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_DELETE_NAME:
|
|
|
|
DECODE_QSTR;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "DELETE_NAME %s", qstr_str(qst));
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
2015-03-20 13:12:09 -04:00
|
|
|
case MP_BC_DELETE_GLOBAL:
|
|
|
|
DECODE_QSTR;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "DELETE_GLOBAL %s", qstr_str(qst));
|
2015-03-20 13:12:09 -04:00
|
|
|
break;
|
|
|
|
|
2014-12-26 22:00:08 -05:00
|
|
|
case MP_BC_DUP_TOP:
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "DUP_TOP");
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_DUP_TOP_TWO:
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "DUP_TOP_TWO");
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_POP_TOP:
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "POP_TOP");
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_ROT_TWO:
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "ROT_TWO");
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_ROT_THREE:
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "ROT_THREE");
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_JUMP:
|
|
|
|
DECODE_SLABEL;
|
2022-03-15 20:39:27 -04:00
|
|
|
mp_printf(print, "JUMP " UINT_FMT, (mp_uint_t)(ip + unum - ip_start));
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_POP_JUMP_IF_TRUE:
|
|
|
|
DECODE_SLABEL;
|
2022-03-15 20:39:27 -04:00
|
|
|
mp_printf(print, "POP_JUMP_IF_TRUE " UINT_FMT, (mp_uint_t)(ip + unum - ip_start));
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_POP_JUMP_IF_FALSE:
|
|
|
|
DECODE_SLABEL;
|
2022-03-15 20:39:27 -04:00
|
|
|
mp_printf(print, "POP_JUMP_IF_FALSE " UINT_FMT, (mp_uint_t)(ip + unum - ip_start));
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_JUMP_IF_TRUE_OR_POP:
|
2022-03-21 01:36:13 -04:00
|
|
|
DECODE_ULABEL;
|
2022-03-15 20:39:27 -04:00
|
|
|
mp_printf(print, "JUMP_IF_TRUE_OR_POP " UINT_FMT, (mp_uint_t)(ip + unum - ip_start));
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_JUMP_IF_FALSE_OR_POP:
|
2022-03-21 01:36:13 -04:00
|
|
|
DECODE_ULABEL;
|
2022-03-15 20:39:27 -04:00
|
|
|
mp_printf(print, "JUMP_IF_FALSE_OR_POP " UINT_FMT, (mp_uint_t)(ip + unum - ip_start));
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_SETUP_WITH:
|
|
|
|
DECODE_ULABEL; // loop-like labels are always forward
|
2022-03-15 20:39:27 -04:00
|
|
|
mp_printf(print, "SETUP_WITH " UINT_FMT, (mp_uint_t)(ip + unum - ip_start));
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_WITH_CLEANUP:
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "WITH_CLEANUP");
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_UNWIND_JUMP:
|
|
|
|
DECODE_SLABEL;
|
2022-03-15 20:39:27 -04:00
|
|
|
mp_printf(print, "UNWIND_JUMP " UINT_FMT " %d", (mp_uint_t)(ip + unum - ip_start), *ip);
|
2014-12-26 22:00:08 -05:00
|
|
|
ip += 1;
|
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_SETUP_EXCEPT:
|
|
|
|
DECODE_ULABEL; // except labels are always forward
|
2022-03-15 20:39:27 -04:00
|
|
|
mp_printf(print, "SETUP_EXCEPT " UINT_FMT, (mp_uint_t)(ip + unum - ip_start));
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_SETUP_FINALLY:
|
|
|
|
DECODE_ULABEL; // except labels are always forward
|
2022-03-15 20:39:27 -04:00
|
|
|
mp_printf(print, "SETUP_FINALLY " UINT_FMT, (mp_uint_t)(ip + unum - ip_start));
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_END_FINALLY:
|
|
|
|
// if TOS is an exception, reraises the exception (3 values on TOS)
|
|
|
|
// if TOS is an integer, does something else
|
|
|
|
// if TOS is None, just pops it and continues
|
|
|
|
// else error
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "END_FINALLY");
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_GET_ITER:
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "GET_ITER");
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
2016-01-09 18:59:52 -05:00
|
|
|
case MP_BC_GET_ITER_STACK:
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "GET_ITER_STACK");
|
2016-01-09 18:59:52 -05:00
|
|
|
break;
|
|
|
|
|
2014-12-26 22:00:08 -05:00
|
|
|
case MP_BC_FOR_ITER:
|
|
|
|
DECODE_ULABEL; // the jump offset if iteration finishes; for labels are always forward
|
2022-03-15 20:39:27 -04:00
|
|
|
mp_printf(print, "FOR_ITER " UINT_FMT, (mp_uint_t)(ip + unum - ip_start));
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
2019-02-14 20:18:59 -05:00
|
|
|
case MP_BC_POP_EXCEPT_JUMP:
|
|
|
|
DECODE_ULABEL; // these labels are always forward
|
2022-03-15 20:39:27 -04:00
|
|
|
mp_printf(print, "POP_EXCEPT_JUMP " UINT_FMT, (mp_uint_t)(ip + unum - ip_start));
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_BUILD_TUPLE:
|
|
|
|
DECODE_UINT;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "BUILD_TUPLE " UINT_FMT, unum);
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_BUILD_LIST:
|
|
|
|
DECODE_UINT;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "BUILD_LIST " UINT_FMT, unum);
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_BUILD_MAP:
|
|
|
|
DECODE_UINT;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "BUILD_MAP " UINT_FMT, unum);
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_STORE_MAP:
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "STORE_MAP");
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_BUILD_SET:
|
|
|
|
DECODE_UINT;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "BUILD_SET " UINT_FMT, unum);
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
2020-02-26 23:36:53 -05:00
|
|
|
#if MICROPY_PY_BUILTINS_SLICE
|
2014-12-26 22:00:08 -05:00
|
|
|
case MP_BC_BUILD_SLICE:
|
|
|
|
DECODE_UINT;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "BUILD_SLICE " UINT_FMT, unum);
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
2020-02-26 23:36:53 -05:00
|
|
|
#endif
|
2014-01-18 09:10:48 -05:00
|
|
|
|
2016-09-18 18:46:01 -04:00
|
|
|
case MP_BC_STORE_COMP:
|
|
|
|
DECODE_UINT;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "STORE_COMP " UINT_FMT, unum);
|
2016-09-18 18:46:01 -04:00
|
|
|
break;
|
|
|
|
|
2014-12-26 22:00:08 -05:00
|
|
|
case MP_BC_UNPACK_SEQUENCE:
|
|
|
|
DECODE_UINT;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "UNPACK_SEQUENCE " UINT_FMT, unum);
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
2015-06-18 11:12:17 -04:00
|
|
|
case MP_BC_UNPACK_EX:
|
|
|
|
DECODE_UINT;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "UNPACK_EX " UINT_FMT, unum);
|
2015-06-18 11:12:17 -04:00
|
|
|
break;
|
|
|
|
|
2014-12-26 22:00:08 -05:00
|
|
|
case MP_BC_MAKE_FUNCTION:
|
|
|
|
DECODE_PTR;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "MAKE_FUNCTION %p", (void *)(uintptr_t)unum);
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_MAKE_FUNCTION_DEFARGS:
|
|
|
|
DECODE_PTR;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "MAKE_FUNCTION_DEFARGS %p", (void *)(uintptr_t)unum);
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_MAKE_CLOSURE: {
|
|
|
|
DECODE_PTR;
|
|
|
|
mp_uint_t n_closed_over = *ip++;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "MAKE_CLOSURE %p " UINT_FMT, (void *)(uintptr_t)unum, n_closed_over);
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
case MP_BC_MAKE_CLOSURE_DEFARGS: {
|
|
|
|
DECODE_PTR;
|
|
|
|
mp_uint_t n_closed_over = *ip++;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "MAKE_CLOSURE_DEFARGS %p " UINT_FMT, (void *)(uintptr_t)unum, n_closed_over);
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
}
|
2013-12-30 17:32:17 -05:00
|
|
|
|
2014-12-26 22:00:08 -05:00
|
|
|
case MP_BC_CALL_FUNCTION:
|
|
|
|
DECODE_UINT;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "CALL_FUNCTION n=" UINT_FMT " nkw=" UINT_FMT, unum & 0xff, (unum >> 8) & 0xff);
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_CALL_FUNCTION_VAR_KW:
|
|
|
|
DECODE_UINT;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "CALL_FUNCTION_VAR_KW n=" UINT_FMT " nkw=" UINT_FMT, unum & 0xff, (unum >> 8) & 0xff);
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_CALL_METHOD:
|
|
|
|
DECODE_UINT;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "CALL_METHOD n=" UINT_FMT " nkw=" UINT_FMT, unum & 0xff, (unum >> 8) & 0xff);
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_CALL_METHOD_VAR_KW:
|
|
|
|
DECODE_UINT;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "CALL_METHOD_VAR_KW n=" UINT_FMT " nkw=" UINT_FMT, unum & 0xff, (unum >> 8) & 0xff);
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_RETURN_VALUE:
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "RETURN_VALUE");
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
2019-08-21 22:39:07 -04:00
|
|
|
case MP_BC_RAISE_LAST:
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "RAISE_LAST");
|
2019-08-21 22:39:07 -04:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_RAISE_OBJ:
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "RAISE_OBJ");
|
2019-08-21 22:39:07 -04:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_RAISE_FROM:
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "RAISE_FROM");
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_YIELD_VALUE:
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "YIELD_VALUE");
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_YIELD_FROM:
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "YIELD_FROM");
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_IMPORT_NAME:
|
|
|
|
DECODE_QSTR;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "IMPORT_NAME '%s'", qstr_str(qst));
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_IMPORT_FROM:
|
|
|
|
DECODE_QSTR;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "IMPORT_FROM '%s'", qstr_str(qst));
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
case MP_BC_IMPORT_STAR:
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "IMPORT_STAR");
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
|
|
|
|
default:
|
|
|
|
if (ip[-1] < MP_BC_LOAD_CONST_SMALL_INT_MULTI + 64) {
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "LOAD_CONST_SMALL_INT " INT_FMT, (mp_int_t)ip[-1] - MP_BC_LOAD_CONST_SMALL_INT_MULTI - 16);
|
2014-12-26 22:00:08 -05:00
|
|
|
} else if (ip[-1] < MP_BC_LOAD_FAST_MULTI + 16) {
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "LOAD_FAST " UINT_FMT, (mp_uint_t)ip[-1] - MP_BC_LOAD_FAST_MULTI);
|
2014-12-26 22:00:08 -05:00
|
|
|
} else if (ip[-1] < MP_BC_STORE_FAST_MULTI + 16) {
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "STORE_FAST " UINT_FMT, (mp_uint_t)ip[-1] - MP_BC_STORE_FAST_MULTI);
|
2017-10-03 08:34:28 -04:00
|
|
|
} else if (ip[-1] < MP_BC_UNARY_OP_MULTI + MP_UNARY_OP_NUM_BYTECODE) {
|
2021-11-19 01:05:40 -05:00
|
|
|
mp_uint_t op = ip[-1] - MP_BC_UNARY_OP_MULTI;
|
|
|
|
mp_printf(print, "UNARY_OP " UINT_FMT " %s", op, qstr_str(mp_unary_op_method_name[op]));
|
2017-10-03 08:34:28 -04:00
|
|
|
} else if (ip[-1] < MP_BC_BINARY_OP_MULTI + MP_BINARY_OP_NUM_BYTECODE) {
|
2014-12-28 14:41:58 -05:00
|
|
|
mp_uint_t op = ip[-1] - MP_BC_BINARY_OP_MULTI;
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "BINARY_OP " UINT_FMT " %s", op, qstr_str(mp_binary_op_method_name[op]));
|
2014-12-26 22:00:08 -05:00
|
|
|
} else {
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "code %p, byte code 0x%02x not implemented\n", ip - 1, ip[-1]);
|
2014-12-26 22:00:08 -05:00
|
|
|
assert(0);
|
|
|
|
return ip;
|
2014-04-22 20:22:10 -04:00
|
|
|
}
|
2014-12-26 22:00:08 -05:00
|
|
|
break;
|
|
|
|
}
|
2014-03-26 17:14:59 -04:00
|
|
|
|
2014-12-26 22:00:08 -05:00
|
|
|
return ip;
|
|
|
|
}
|
|
|
|
|
2022-03-15 20:39:27 -04:00
|
|
|
void mp_bytecode_print2(const mp_print_t *print, const byte *ip, size_t len, mp_raw_code_t *const *child_table, const mp_module_constants_t *cm) {
|
|
|
|
const byte *ip_start = ip;
|
|
|
|
while (ip < ip_start + len) {
|
|
|
|
mp_printf(print, "%02u ", (uint)(ip - ip_start));
|
|
|
|
ip = mp_bytecode_print_str(print, ip_start, ip, child_table, cm);
|
2020-09-11 03:22:28 -04:00
|
|
|
mp_printf(print, "\n");
|
2013-11-17 08:19:33 -05:00
|
|
|
}
|
|
|
|
}
|
2014-01-07 10:20:33 -05:00
|
|
|
|
2014-01-19 06:48:48 -05:00
|
|
|
#endif // MICROPY_DEBUG_PRINTERS
|