Add a new MICROPY_COMP_CONST_FLOAT feature, enabled by in mpy-cross and
when compiling with MICROPY_CONFIG_ROM_LEVEL_CORE_FEATURES. The new
feature leverages the code of MICROPY_COMP_CONST_FOLDING to support folding
of floating point constants.
If MICROPY_COMP_MODULE_CONST is defined as well, math module constants are
made available at compile time. For example:
_DEG_TO_GRADIANT = const(math.pi / 180)
_INVALID_VALUE = const(math.nan)
A few corner cases had to be handled:
- The float const folding code should not fold expressions resulting into
complex results, as the mpy parser for complex immediates has
limitations.
- The constant generation code must distinguish between -0.0 and 0.0, which
are different even if C consider them as ==.
This change removes previous limitations on the use of `const()`
expressions that would result in floating point number, so the test cases
of micropython/const_error have to be updated.
Additional test cases have been added to cover the new repr() code (from a
previous commit). A few other simple test cases have been added to handle
the use of floats in `const()` expressions, but the float folding code
itself is also tested when running general float test cases, as float
expressions often get resolved at compile-time (with this change).
Signed-off-by: Yoctopuce dev <dev@yoctopuce.com>
Following discussions in PR #16666, this commit updates the float
formatting code to improve the `repr` reversibility, i.e. the percentage of
valid floating point numbers that do parse back to the same number when
formatted by `repr` (in CPython it's 100%).
This new code offers a choice of 3 float conversion methods, depending on
the desired tradeoff between code size and conversion precision:
- BASIC method is the smallest code footprint
- APPROX method uses an iterative method to approximate the exact
representation, which is a bit slower but but does not have a big impact
on code size. It provides `repr` reversibility on >99.8% of the cases in
double precision, and on >98.5% in single precision (except with REPR_C,
where reversibility is 100% as the last two bits are not taken into
account).
- EXACT method uses higher-precision floats during conversion, which
provides perfect results but has a higher impact on code size. It is
faster than APPROX method, and faster than the CPython equivalent
implementation. It is however not available on all compilers when using
FLOAT_IMPL_DOUBLE.
Here is the table comparing the impact of the three conversion methods on
code footprint on PYBV10 (using single-precision floats) and reversibility
rate for both single-precision and double-precision floats. The table
includes current situation as a baseline for the comparison:
PYBV10 REPR_C FLOAT DOUBLE
current = 364688 12.9% 27.6% 37.9%
basic = 364812 85.6% 60.5% 85.7%
approx = 365080 100.0% 98.5% 99.8%
exact = 366408 100.0% 100.0% 100.0%
Signed-off-by: Yoctopuce dev <dev@yoctopuce.com>
Current longlong implementation does not allow a float as RHS of mathematic
operators, as it lacks the delegation code present in mpz.
Signed-off-by: Yoctopuce dev <dev@yoctopuce.com>
Commit dc2fcfcc55 seems to have accidentally
changed the ruff quote style to "preserve", instead of keeping it at the
default which is "double".
Put it back to the default and update relevant .py files with this rule.
Signed-off-by: Damien George <damien@micropython.org>
Current implementation of REPR_C works by clearing the two lower bits of
the mantissa to zero. As this happens after each floating point operation,
this tends to bias floating point numbers towards zero, causing decimals
like .9997 instead of rounded numbers. This is visible in test cases
involving repeated computations, such as `tests/misc/rge_sm.py` for
instance.
The suggested fix fills in the missing bits by copying the previous two
bits. Although this cannot recreate missing information, it fixes the bias
by inserting plausible values for the lost bits, at a relatively low cost.
Some float tests involving irrational numbers have to be softened in case
of REPR_C, as the 30 bits are not always enough to fulfill the expectations
of the original test, and the change may randomly affect the last digits.
Such cases have been made explicit by testing for REPR_C or by adding a
clear comment.
The perf_test fft code was also missing a call to round() before casting a
log_2 operation to int, which was causing a failure due to a last-decimal
change.
Signed-off-by: Yoctopuce dev <dev@yoctopuce.com>
CPython math.nan is positive with regards to copysign. The signaling bit
(aka sign flag) was incorrectly set.
In addition, REPR_C and REPR_D should only use the _true_ nan to prevent
system crash in case of hand-crafted floats. For instance, with REPR_C,
any nan-like float following the pattern
`01111111 1xxxxxxx xxxxxxxx xxxxx1xx` would be switched to an immediate
object or a qstr string. When the qstr index is too large, this would
cause a crash.
This commit fixes the issue, and adds the relevant test cases.
Signed-off-by: Yoctopuce dev <dev@yoctopuce.com>
If a complex literal had a negative real part and a positive imaginary
part, it was not parsed properly because the imaginary part also came out
negative.
Includes a test of complex parsing, which fails without this fix.
Co-authored-by: ComplexSymbol <141301057+ComplexSymbol@users.noreply.github.com>
Signed-off-by: Jeff Epler <jepler@gmail.com>
The existing test for `math.e` and `math.pi` constants can fail on certain
targets if the functions `math.exp()` and/or `math.cos()` are not accurate
enough (eg out by an LSB of float precision). For example this test
currently fails on PYBD_SF6 which uses double precision floats (and that's
due to the `lib/libm_dbl/exp.c` implementation not being exact).
This commit changes this constant test so that it tests the actual constant
value, not the evaluation of `exp()` and `cos()` functions.
Signed-off-by: Damien George <damien@micropython.org>
The mantissa parsing code uses a floating point variable to accumulate
digits. Using an `mp_float_uint_t` variable instead and casting to
`mp_float_t` at the very end reduces code size. In some cases, it also
improves the rounding behaviour as extra digits are taken into account
by the int-to-float conversion code.
An extra test case handles the special case where mantissa overflow occurs
while processing deferred trailing zeros.
Signed-off-by: Yoctopuce dev <dev@yoctopuce.com>
Otherwise CPython gives a deprecation warning.
This test is not actually testing inversion of bools, rather that bit of
the test is used to compute the pass/fail result.
Signed-off-by: Damien George <damien@micropython.org>
This commit implements the 'e' half-float format: 10-bit mantissa, 5-bit
exponent. It uses native _Float16 if supported by the compiler, otherwise
uses custom bitshifting encoding/decoding routines.
Signed-off-by: Matthias Urlichs <matthias@urlichs.de>
Signed-off-by: Damien George <damien@micropython.org>
Instead of having a special set of arguments to test for each math-module
function, just test all functions with all sets of arguments. This gives
improved test cases to prevent regressions.
Signed-off-by: Damien George <damien@micropython.org>
Prior to this fix, pow(1.5, inf) and pow(0.5, -inf) (among other things)
would incorrectly raise a ValueError, because the result is inf with the
first argument being finite. This commit fixes this by allowing the result
to be infinite if the first or second (or both) argument is infinite.
This fix doesn't affect the other three math functions that have two
arguments:
- atan2 never returns inf, so always fails isinf(ans)
- copysign returns inf only if the first argument x is inf, so will never
reach the isinf(y) check
- fmod never returns inf, so always fails isinf(ans)
Signed-off-by: Damien George <damien@micropython.org>
Rework the conversion of floats to decimal strings so it aligns precisely
with the conversion of strings to floats in parsenum.c. This is to avoid
rendering 1eX as 9.99999eX-1 etc. This is achieved by removing the power-
of-10 tables and using pow() to compute the exponent directly, and that's
done efficiently by first estimating the power-of-10 exponent from the
power-of-2 exponent in the floating-point representation.
Code size is reduced by roughly 100 to 200 bytes by this commit.
Signed-off-by: Dan Ellis <dan.ellis@gmail.com>
Formerly, py/formatfloat would print whole numbers inaccurately with
nonzero digits beyond the decimal place. This resulted from its strategy
of successive scaling of the argument by 0.1 which cannot be exactly
represented in floating point. The change in this commit avoids scaling
until the value is smaller than 1, so all whole numbers print with zero
fractional part.
Fixes issue #4212.
Signed-off-by: Dan Ellis dan.ellis@gmail.com
This commit simplifies mp_obj_get_complex_maybe() by first calling
mp_obj_get_float_maybe() to handle the cases corresponding to floats.
Only if that fails does it attempt to extra a full complex number.
This reduces code size and also means that mp_obj_get_complex_maybe() now
supports user-defined classes defining __float__; in particular this allows
user-defined classes to be used as arguments to cmath-module function.
Furthermore, complex_make_new() can now be simplified to directly call
mp_obj_get_complex(), instead of mp_obj_get_complex_maybe() followed by
mp_obj_get_float(). This also improves error messages from complex with
an invalid argument, it now raises "can't convert <type> to complex" rather
than "can't convert <type> to float".
Signed-off-by: Damien George <damien@micropython.org>
Prior to this commit, complex("j") would return 0j, and complex("nanj")
would return nan+0j. This commit makes sure "j" is tested for after
parsing the number (nan, inf or a decimal), and also supports the case of
"j" on its own.
Signed-off-by: Damien George <damien@micropython.org>
2.5 can be represented correctly in object representation C, but 2.3 cannot
(it is slightly truncated).
Signed-off-by: Damien George <damien@micropython.org>
Prior to this commit, pow(-2, float('nan')) would return (nan+nanj), or
raise an exception on targets that don't support complex numbers. This is
fixed to return simply nan, as CPython does.
Signed-off-by: Damien George <damien@micropython.org>
This is consistent with the other 'micro' modules and allows implementing
additional features in Python via e.g. micropython-lib's sys.
Note this is a breaking change (not backwards compatible) for ports which
do not enable weak links, as "import sys" must now be replaced with
"import usys".
Updating to Black v20.8b1 there are two changes that affect the code in
this repository:
- If there is a trailing comma in a list (eg [], () or function call) then
that list is now written out with one line per element. So remove such
trailing commas where the list should stay on one line.
- Spaces at the start of """ doc strings are removed.
Signed-off-by: Damien George <damien@micropython.org>
So they can be skipped if __rOP__'s are not supported on the target. Also
fix the typo in the complex_special_methods.py filename.
Signed-off-by: Damien George <damien@micropython.org>
This allows complex binary operations to fail gracefully with unsupported
operation rather than raising an exception, so that special methods work
correctly.
Signed-off-by: Damien George <damien@micropython.org>
Older implementations deal with infinity/negative zero incorrectly. This
commit adds generic fixes that can be enabled by any port that needs them,
along with new tests cases.
For combinations of certain versions of glibc and gcc the definition of
fpclassify always takes float as argument instead of adapting itself to
float/double/long double as required by the C99 standard. At the time of
writing this happens for instance for glibc 2.27 with gcc 7.5.0 when
compiled with -Os and glibc 3.0.7 with gcc 9.3.0. When calling fpclassify
with double as argument, as in objint.c, this results in an implicit
narrowing conversion which is not really correct plus results in a warning
when compiled with -Wfloat-conversion. So fix this by spelling out the
logic manually.
When the unix and windows ports use MICROPY_FLOAT_IMPL_FLOAT instead of
MICROPY_FLOAT_IMPL_DOUBLE, the test output has for example
complex(-0.15052, 0.34109) instead of the expected
complex(-0.15051, 0.34109).
Use one decimal place less for the output printing to fix this.
Since automatically formatting tests with black, we have lost one line of
code coverage. This adds an explicit test to ensure we are testing the
case that is no longer covered implicitly.
This adds the Python files in the tests/ directory to be formatted with
./tools/codeformat.py. The basics/ subdirectory is excluded for now so we
aren't changing too much at once.
In a few places `# fmt: off`/`# fmt: on` was used where the code had
special formatting for readability or where the test was actually testing
the specific formatting.
As per PEP 485, this function appeared in for Python 3.5. Configured via
MICROPY_PY_MATH_ISCLOSE which is disabled by default, but enabled for the
ports which already have MICROPY_PY_MATH_SPECIAL_FUNCTIONS enabled.
Nan and inf (signed and unsigned) are also handled correctly by using
signbit (they were also handled correctly with "val<0", but that didn't
handle -0.0 correctly). A test case is added for this behaviour.
This commit adds the math.factorial function in two variants:
- squared difference, which is faster than the naive version, relatively
compact, and non-recursive;
- a mildly optimised recursive version, faster than the above one.
There are some more optimisations that could be done, but they tend to take
more code, and more storage space. The recursive version seems like a
sensible compromise.
The new function is disabled by default, and uses the non-optimised version
by default if it is enabled. The options are MICROPY_PY_MATH_FACTORIAL
and MICROPY_OPT_MATH_FACTORIAL.
