This patch combines the compiler optimisation code for double and triple
tuple-to-tuple assignment, taking it from two separate if-blocks to one
combined if-block. This can be done because the code for both of these
optimisations has a lot in common. Combining them together reduces code
size for ports that have the triple-tuple optimisation enabled (and doesn't
change code size for ports that have it disabled).
The technique of using alloca is how dotted import names are composed in
mp_import_from and mp_builtin___import__, so use the same technique in the
compiler. This puts less pressure on the heap (only the stack is used if
the qstr already exists, and if it doesn't exist then the standard qstr
block memory is used for the new qstr rather than a separate chunk of the
heap) and reduces overall code size.
Previous to this patch, a label with value "0" was used to indicate an
invalid label, but that meant a wasted word (at slot 0) in the array of
label offsets. This patch adjusts the label indices so the first one
starts at 0, and the maximum value indicates an invalid label.
This patch fixes a bug whereby the Python stack was not correctly reset if
there was a break/continue statement in the else black of an optimised
for-range loop.
For example, in the following code the "j" variable from the inner for loop
was not being popped off the Python stack:
for i in range(4):
for j in range(4):
pass
else:
continue
This is now fixed with this patch.
In CPython 3.4 this raises a SyntaxError. In CPython 3.5+ having a
positional after * is allowed but uPy has the wrong semantics and passes
the arguments in the incorrect order. To prevent incorrect use of a
function going unnoticed it is important to raise the SyntaxError in uPy,
until the behaviour is fixed to follow CPython 3.5+.
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 refactors the handling of the special super() call within the
compiler. It removes the need for a global (to the compiler) state variable
which keeps track of whether the subject of an expression is super. The
handling of super() is now done entirely within one function, which makes
the compiler a bit cleaner and allows to easily add more optimisations to
super calls.
Changes to the code size are:
bare-arm: +12
minimal: +0
unix x64: +48
unix nanbox: -16
stmhal: +4
cc3200: +0
esp8266: -56
With this optimisation enabled the compiler optimises the if-else
expression within a return statement. The optimisation reduces bytecode
size by 2 bytes for each use of such a return-if-else statement. Since
such a statement is not often used, and costs bytes for the code, the
feature is disabled by default.
For example the following code:
def f(x):
return 1 if x else 2
compiles to this bytecode with the optimisation disabled (left column is
bytecode offset in bytes):
00 LOAD_FAST 0
01 POP_JUMP_IF_FALSE 8
04 LOAD_CONST_SMALL_INT 1
05 JUMP 9
08 LOAD_CONST_SMALL_INT 2
09 RETURN_VALUE
and to this bytecode with the optimisation enabled:
00 LOAD_FAST 0
01 POP_JUMP_IF_FALSE 6
04 LOAD_CONST_SMALL_INT 1
05 RETURN_VALUE
06 LOAD_CONST_SMALL_INT 2
07 RETURN_VALUE
So the JUMP to RETURN_VALUE is optimised and replaced by RETURN_VALUE,
saving 2 bytes and making the code a bit faster.
Otherwise the type of parse-node and its kind has to be re-extracted
multiple times. This optimisation reduces code size by a bit (16 bytes on
bare-arm).
With this patch all illegal assignments are reported as "can't assign to
expression". Before the patch there were special cases for a literal on
the LHS, and for augmented assignments (eg +=), but it seems a waste of
bytes (and there are lots of bytes used in error messages) to spend on
distinguishing such errors which a user will rarely encounter.
The self variable may be closed-over in the function, and in that case the
call to super() should load the contents of the closure cell using
LOAD_DEREF (before this patch it would just load the cell directly).
It improves readability of code and reduces the chance to make a mistake.
This patch also fixes a bug with nan-boxing builds by rounding up the
calculation of the new NSLOTS variable, giving the correct number of slots
(being 4) even if mp_obj_t is larger than the native machine size.
Previous to this patch any non-interned str/bytes objects would create a
special parse node that held a copy of the str/bytes data. Then in the
compiler this data would be turned into a str/bytes object. This actually
lead to 2 copies of the data, one in the parse node and one in the object.
The parse node's copy of the data would be freed at the end of the compile
stage but nevertheless it meant that the peak memory usage of the
parse/compile stage was higher than it needed to be (by an amount equal to
the number of bytes in all the non-interned str/bytes objects).
This patch changes the behaviour so that str/bytes objects are created
directly in the parser and the object stored in a const-object parse node
(which already exists for bignum, float and complex const objects). This
reduces peak RAM usage of the parse/compile stage, simplifies the parser
and compiler, and reduces code size by about 170 bytes on Thumb2 archs,
and by about 300 bytes on Xtensa archs.
It's much more efficient in RAM and code size to do implicit literal string
concatenation in the lexer, as opposed to the compiler.
RAM usage is reduced because the concatenation can be done right away in the
tokeniser by just accumulating the string/bytes literals into the lexer's
vstr. Prior to this patch adjacent strings/bytes would create a parse tree
(one node per string/bytes) and then in the compiler a whole new chunk of
memory was allocated to store the concatenated string, which used more than
double the memory compared to just accumulating in the lexer.
This patch also significantly reduces code size:
bare-arm: -204
minimal: -204
unix x64: -328
stmhal: -208
esp8266: -284
cc3200: -224
Grammar rules have 2 variants: ones that are attached to a specific
compile function which is called to compile that grammar node, and ones
that don't have a compile function and are instead just inspected to see
what form they take.
In the compiler there is a table of all grammar rules, with each entry
having a pointer to the associated compile function. Those rules with no
compile function have a null pointer. There are 120 such rules, so that's
120 words of essentially wasted code space.
By grouping together the compile vs no-compile rules we can put all the
no-compile rules at the end of the list of rules, and then we don't need
to store the null pointers. We just have a truncated table and it's
guaranteed that when indexing this table we only index the first half,
the half with populated pointers.
This patch implements such a grouping by having a specific macro for the
compile vs no-compile grammar rules (DEF_RULE vs DEF_RULE_NC). It saves
around 460 bytes of code on 32-bit archs.
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.
It is split into 2 functions, one to make small ints and the other to make
a non-small-int leaf node. This reduces code size by 32 bytes on
bare-arm, 64 bytes on unix (x64-64) and 144 bytes on stmhal.
When an exception is raised and is to be handled by the VM, it is stored
on the Python value stack so the bytecode can access it. CPython stores
3 objects on the stack for each exception: exc type, exc instance and
traceback. uPy followed this approach, but it turns out not to be
necessary. Instead, it is enough to store just the exception instance on
the Python value stack. The only place where the 3 values are needed
explicitly is for the __exit__ handler of a with-statement context, but
for these cases the 3 values can be extracted from the single exception
instance.
This patch removes the need to store 3 values on the stack, and instead
just stores the exception instance.
Code size is reduced by about 50-100 bytes, the compiler and VM are
slightly simpler, generate bytecode is smaller (by 2 bytes for each try
block), and the Python value stack is reduced in size for functions that
handle exceptions.
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.
They are sugar for marking function as generator, "yield from"
and pep492 python "semantically equivalents" respectively.
@dpgeorge was the original author of this patch, but @pohmelie made
changes to implement `async for` and `async with`.
Previous to this patch, the "**b" in "a**b" had its own parse node with
just one item (the "b"). Now, the "b" is just the last element of the
power parse-node. This saves (a tiny bit of) RAM when compiling.
This new compile-time option allows to make the bytecode compiler
configurable at runtime by setting the fields in the mp_dynamic_compiler
structure. By using this feature, the compiler can generate bytecode
that targets any MicroPython runtime/VM, regardless of the host and
target compile-time settings.
Options so far that fall under this dynamic setting are:
- maximum number of bits that a small int can hold;
- whether caching of lookups is used in the bytecode;
- whether to use unicode strings or not (lexer behaviour differs, and
therefore generated string constants differ).
Before this patch, (x+y)*z would be parsed to a tree that contained a
redundant identity parse node corresponding to the parenthesis. With
this patch such nodes are optimised away, which reduces memory
requirements for expressions with parenthesis, and simplifies the
compiler because it doesn't need to handle this identity case.
A parenthesis parse node is still needed for tuples.
MICROPY_ENABLE_COMPILER can be used to enable/disable the entire compiler,
which is useful when only loading of pre-compiled bytecode is supported.
It is enabled by default.
MICROPY_PY_BUILTINS_EVAL_EXEC controls support of eval and exec builtin
functions. By default they are only included if MICROPY_ENABLE_COMPILER
is enabled.
Disabling both options saves about 40k of code size on 32-bit x86.
To use, put the following in mpconfigport.h:
#define MICROPY_OBJ_REPR (MICROPY_OBJ_REPR_D)
#define MICROPY_FLOAT_IMPL (MICROPY_FLOAT_IMPL_DOUBLE)
typedef int64_t mp_int_t;
typedef uint64_t mp_uint_t;
#define UINT_FMT "%llu"
#define INT_FMT "%lld"
Currently does not work with native emitter enabled.
It makes much more sense to do constant folding in the parser while the
parse tree is being built. This eliminates the need to create parse
nodes that will just be folded away. The code is slightly simpler and a
bit smaller as well.
Constant folding now has a configuration option,
MICROPY_COMP_CONST_FOLDING, which is enabled by default.
With this patch parse nodes are allocated sequentially in chunks. This
reduces fragmentation of the heap and prevents waste at the end of
individually allocated parse nodes.
Saves roughly 20% of RAM during parse stage.
Function annotations are only needed when the native emitter is enabled
and when the current scope is emitted in viper mode. All other times
the annotations can be skipped completely.
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 there were some cases where line numbers for
errors were 0 (unknown). Now the compiler attempts to give a better
line number where possible, in some cases giving the line number of the
closest statement, and other cases the line number of the inner-most
scope of the error (eg the line number of the start of the function).
This helps to give good (and sometimes exact) line numbers for
ViperTypeError exceptions.
This patch also makes sure that the first compile error (eg SyntaxError)
that is encountered is reported (previously it was the last one that was
reported).
ViperTypeError now includes filename and function name where the error
occurred. The line number is the line number of the start of the
function definition, which is the best that can be done without a lot
more work.
Partially addresses issue #1381.
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.
These allow to fine-tune the compiler to select whether it optimises
tuple assignments of the form a, b = c, d and a, b, c = d, e, f.
Sensible defaults are provided.
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.
To enable parsing constants more efficiently, mp_parse should be allowed
to raise an exception, and mp_compile can already raise a MemoryError.
So these functions need to be protected by an nlr push/pop block.
This patch adds that feature in all places. This allows to simplify how
mp_parse and mp_compile are called: they now raise an exception if they
have an error and so explicit checking is not needed anymore.
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.
With this patch str/bytes construction is streamlined. Always use a
vstr to build a str/bytes object. If the size is known beforehand then
use vstr_init_len to allocate only required memory. Otherwise use
vstr_init and the vstr will grow as needed. Then use
mp_obj_new_str_from_vstr to create a str/bytes object using the vstr
memory.
Saves code ROM: 68 bytes on stmhal, 108 bytes on bare-arm, and 336 bytes
on unix x64.
Bytecode also needs a pass to compute the stack size. This is because
the state size of the bytecode function is encoded as a variable uint,
so we must know the value of this uint before we encode it (otherwise
the size of the generated code changes from one pass to the next).
Having an entire pass for this seems wasteful (in time). Alternative is
to allocate fixed space for the state size (would need 3-4 bytes to be
general, when 1 byte is usually sufficient) which uses a bit of extra
RAM per bytecode function, and makes the code less elegant in places
where this uint is encoded/decoded.
So, for now, opt for an extra pass.
Previously to this patch all constant string/bytes objects were
interned by the compiler, and this lead to crashes when the qstr was too
long (noticeable now that qstr length storage defaults to 1 byte).
With this patch, long string/bytes objects are never interned, and are
referenced directly as constant objects within generated code using
load_const_obj.
Compiler optimises lookup of module.CONST when enabled (an existing
feature). Disabled by default; enabled for unix, windows, stmhal.
Costs about 100 bytes ROM on stmhal.
This patch makes the MICROPY_PY_BUILTINS_SLICE compile-time option
fully disable the builtin slice operation (when set to 0). This
includes removing the slice sytanx from the grammar. Now, enabling
slice costs 4228 bytes on unix x64, and 1816 bytes on stmhal.
This patch makes MICROPY_PY_BUILTINS_SET compile-time option fully
disable the builtin set object (when set to 0). This includes removing
set constructor/comprehension from the grammar, the compiler and the
emitters. Now, enabling set costs 8168 bytes on unix x64, and 3576
bytes on stmhal.
This patch gives proper SyntaxError exceptions for bad global/nonlocal
declarations. It also reduces code size: 304 bytes on unix x64, 132
bytes on stmhal.
You can now assign to the range end variable and the for-loop still
works correctly. This fully addresses issue #565.
Also fixed a bug with the stack not being fully popped when breaking out
of an optimised for-loop (and it's actually impossible to write a test
for this case!).
mp_parse_node_free now frees the memory associated with non-interned
strings. And the parser calls mp_parse_node_free when discarding a
non-used node (such as a doc string).
Also, the compiler now frees the parse tree explicitly just before it
exits (as opposed to relying on the caller to do this).
Addresses issue #708 as best we can.
Damien George