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>
POP_BLOCK and POP_EXCEPT are now the same, and are always followed by a
JUMP. So this optimisation reduces code size, and RAM usage of bytecode by
two bytes for each try-except handler.
Taking the address of a local variable is mildly expensive, in code size
and stack usage. So optimise scope_find_or_add_id() to not need to take a
pointer to the "added" variable, and instead take the kind to use for newly
added identifiers.
Prior to this patch, native code would use a full nlr_buf_t for each
exception handler (try-except, try-finally, with). For nested exception
handlers this would use a lot of C stack and be rather inefficient.
This patch changes how exceptions are handled in native code by setting up
only a single nlr_buf_t context for the entire function, and then manages a
state machine (using the PC) to work out which exception handler to run
when an exception is raised by an nlr_jump. This keeps the C stack usage
at a constant level regardless of the depth of Python exception blocks.
The patch also fixes an existing bug when local variables are written to
within an exception handler, then their value was incorrectly restored if
an exception was raised (since the nlr_jump would restore register values,
back to the point of the nlr_push).
And it also gets nested try-finally+with working with the viper emitter.
Broadly speaking, efficiency of executing native code that doesn't use
any exception blocks is unchanged, and emitted code size is only slightly
increased for such function. C stack usage of all native functions is
either equal or less than before. Emitted code size for native functions
that use exception blocks is increased by roughly 10% (due in part to
fixing of above-mentioned bugs).
But, most importantly, this patch allows to implement more Python features
in native code, like unwind jumps and yielding from within nested exception
blocks.
Header files that are considered internal to the py core and should not
normally be included directly are:
py/nlr.h - internal nlr configuration and declarations
py/bc0.h - contains bytecode macro definitions
py/runtime0.h - contains basic runtime enums
Instead, the top-level header files to include are one of:
py/obj.h - includes runtime0.h and defines everything to use the
mp_obj_t type
py/runtime.h - includes mpstate.h and hence nlr.h, obj.h, runtime0.h,
and defines everything to use the general runtime support functions
Additional, specific headers (eg py/objlist.h) can be included if needed.
The code conventions suggest using header guards, but do not define how
those should look like and instead point to existing files. However, not
all existing files follow the same scheme, sometimes omitting header guards
altogether, sometimes using non-standard names, making it easy to
accidentally pick a "wrong" example.
This commit ensures that all header files of the MicroPython project (that
were not simply copied from somewhere else) follow the same pattern, that
was already present in the majority of files, especially in the py folder.
The rules are as follows.
Naming convention:
* start with the words MICROPY_INCLUDED
* contain the full path to the file
* replace special characters with _
In addition, there are no empty lines before #ifndef, between #ifndef and
one empty line before #endif. #endif is followed by a comment containing
the name of the guard macro.
py/grammar.h cannot use header guards by design, since it has to be
included multiple times in a single C file. Several other files also do not
need header guards as they are only used internally and guaranteed to be
included only once:
* MICROPY_MPHALPORT_H
* mpconfigboard.h
* mpconfigport.h
* mpthreadport.h
* pin_defs_*.h
* qstrdefs*.h
This patch allows the following code to run without allocating on the heap:
super().foo(...)
Before this patch such a call would allocate a super object on the heap and
then load the foo method and call it right away. The super object is only
needed to perform the lookup of the method and not needed after that. This
patch makes an optimisation to allocate the super object on the C stack and
discard it right after use.
Changes in code size due to this patch are:
bare-arm: +128
minimal: +232
unix x64: +416
unix nanbox: +364
stmhal: +184
esp8266: +340
cc3200: +128
This patch moves some common code from the individual inline assemblers to
the compiler, the code that calls the emit-glue to assign the machine code
to the functions scope.
This patch adds the MICROPY_EMIT_INLINE_XTENSA option, which, when
enabled, allows the @micropython.asm_xtensa decorator to be used.
The following opcodes are currently supported (ax is a register, a0-a15):
ret_n()
callx0(ax)
j(label)
jx(ax)
beqz(ax, label)
bnez(ax, label)
mov(ax, ay)
movi(ax, imm) # imm can be full 32-bit, uses l32r if needed
and_(ax, ay, az)
or_(ax, ay, az)
xor(ax, ay, az)
add(ax, ay, az)
sub(ax, ay, az)
mull(ax, ay, az)
l8ui(ax, ay, imm)
l16ui(ax, ay, imm)
l32i(ax, ay, imm)
s8i(ax, ay, imm)
s16i(ax, ay, imm)
s32i(ax, ay, imm)
l16si(ax, ay, imm)
addi(ax, ay, imm)
ball(ax, ay, label)
bany(ax, ay, label)
bbc(ax, ay, label)
bbs(ax, ay, label)
beq(ax, ay, label)
bge(ax, ay, label)
bgeu(ax, ay, label)
blt(ax, ay, label)
bnall(ax, ay, label)
bne(ax, ay, label)
bnone(ax, ay, label)
Upon entry to the assembly function the registers a0, a12, a13, a14 are
pushed to the stack and the stack pointer (a1) decreased by 16. Upon
exit, these registers and the stack pointer are restored, and ret.n is
executed to return to the caller (caller address is in a0).
Note that the ABI for the Xtensa emitters is non-windowing.
The 3 kinds of comprehensions are similar enough that merging their emit
functions reduces code size. Decreases in code size in bytes are:
bare-arm:24, minimal:96, unix(NDEBUG,x86-64):328, stmhal:80, esp8266:76.
unix-cpy was originally written to get semantic equivalent with CPython
without writing functional tests. When writing the initial
implementation of uPy it was a long way between lexer and functional
tests, so the half-way test was to make sure that the bytecode was
correct. The idea was that if the uPy bytecode matched CPython 1-1 then
uPy would be proper Python if the bytecodes acted correctly. And having
matching bytecode meant that it was less likely to miss some deep
subtlety in the Python semantics that would require an architectural
change later on.
But that is all history and it no longer makes sense to retain the
ability to output CPython bytecode, because:
1. It outputs CPython 3.3 compatible bytecode. CPython's bytecode
changes from version to version, and seems to have changed quite a bit
in 3.5. There's no point in changing the bytecode output to match
CPython anymore.
2. uPy and CPy do different optimisations to the bytecode which makes it
harder to match.
3. The bytecode tests are not run. They were never part of Travis and
are not run locally anymore.
4. The EMIT_CPYTHON option needs a lot of extra source code which adds
heaps of noise, especially in compile.c.
5. Now that there is an extensive test suite (which tests functionality)
there is no need to match the bytecode. Some very subtle behaviour is
tested with the test suite and passing these tests is a much better
way to stay Python-language compliant, rather than trying to match
CPy bytecode.
Previous to this patch each time a bytes object was referenced a new
instance (with the same data) was created. With this patch a single
bytes object is created in the compiler and is loaded directly at execute
time as a true constant (similar to loading bignum and float objects).
This saves on allocating RAM and means that bytes objects can now be
used when the memory manager is locked (eg in interrupts).
The MP_BC_LOAD_CONST_BYTES bytecode was removed as part of this.
Generated bytecode is slightly larger due to storing a pointer to the
bytes object instead of the qstr identifier.
Code size is reduced by about 60 bytes on Thumb2 architectures.
This fixes a long standing problem that viper code generation gave
terrible error messages, and actually no errors on pyboard where
assertions are disabled.
Now all compile-time errors are raised as proper Python exceptions, and
are of type ViperTypeError.
Addresses issue #940.
When just the bytecode emitter is needed there is no need to have a
dynamic method table for the emitter back-end, and we can instead
directly call the mp_emit_bc_XXX functions. This gives a significant
reduction in code size and a very slight performance boost for the
compiler.
This patch saves 1160 bytes code on Thumb2 and 972 bytes on x86, when
native emitters are disabled.
Overall savings in code over the last 3 commits are:
bare-arm: 1664 bytes.
minimal: 2136 bytes.
stmhal: 584 bytes (it has native emitter enabled).
cc3200: 1736 bytes.
First pass for the compiler is computing the scope (eg if an identifier
is local or not) and originally had an entire table of methods dedicated
to this, most of which did nothing. With changes from previous commit,
this set of methods can be removed and the methods from the bytecode
emitter used instead, with very little modification -- this is what is
done in this commit.
This factoring has little to no impact on the speed of the compiler
(tested by compiling 3763 Python scripts and timing it).
This factoring reduces code size by about 270-300 bytes on Thumb2 archs,
and 400 bytes on x86.
Previous to this patch, a big-int, float or imag constant was interned
(made into a qstr) and then parsed at runtime to create an object each
time it was needed. This is wasteful in RAM and not efficient. Now,
these constants are parsed straight away in the parser and turned into
objects. This allows constants with large numbers of digits (so
addresses issue #1103) and takes us a step closer to #722.
This cleans up vstr so that it's a pure "variable buffer", and the user
can decide whether they need to add a terminating null byte. In most
places where vstr is used, the vstr did not need to be null terminated
and so this patch saves code size, a tiny bit of RAM, and makes vstr
usage more efficient. When null termination is needed it must be
done explicitly using vstr_null_terminate.
