The previous simple one could be wrong in several low-order digits due to the inaccuracy in the representation of the exponent (1/3). This version effectively breaks the number up into the form a*8^b, computes the cube root of 8^b exactly (i.e. 2^b), and uses the slightly inaccurate exponentiation only for a.
This uses the obvious calculation, except with scaling to avoid unnecessary overflow/underflow.
There is a discussion of hypot implementations in C. Borges, An Improved Algorithm for hypot(a,b) (https://arxiv.org/pdf/1904.09481.pdf). This implementation is similar to the "Naive (Unfused)" version discussed in that paper. As the paper notes, it is possible to get better accuracy by adding a correction term, but the "naive" version is already reasonably good, so we skip the correction in the interest of code size and speed.
The lrint functions could give the wrong result for negative numbers in upward/downward rounding modes. Casts to comp could also have different rounding behavior.
This is basically the implementation recommended in Apple Numerics Manual Ch. 9, except that there is an added case for large values that would otherwise cause an overflow.
These SANE operations can sometimes return incorrect values for certain negative integers such as -2147483648 and -53021371269120 (numbers with at least 16 low-order zero bits in their two's-complement representation). To work around this, we now avoid calling FX2C or FX2L on negative numbers, generally by saving and restoring the sign separately.
These workarounds are used in several of the new <math.h> rounding functions, and also for code that converts floating-point values to comp or long long. There are some places in SysFloat that should be patched similarly, so we may still hit this problem in certain situations until that is done.
This should work, and mostly does. However, it is affected by a bug in FX2L (and FX2C) which can sometimes give the wrong results for certain negative integers (such as -2147483648). I believe this can occur when at least the lower 16 bits if the integer (in two's-complement representation) are zeros.
This is a bit more complex than other rounding functions, because it rounds to nearest but always away from zero in halfway cases, which is not a rounding direction directly supported by SANE.
This differs from scalbn in that the exponent has type long. When scaling an extended value, exponents slightly outside the range of int can actually be used meaningfully. We address this by doing multiple SCALBX calls (at most 2) in a loop.
Most of these actually just jump to the existing functions, since they really use extended precision anyway. The exception is the modf functions, which need a separate implementation for each type because they store a value through a pointer to that type.
The functions implemented so far are largely the ones that map (nearly) directly to SANE calls.
Note that C99 specifies separate float/double/long double versions of each of these functions, but under ORCA/C they generally use the same code.
These macros differ from the normal comparison operators in that they will not signal invalid due to the operands being unordered. However, this implementation will still signal invalid for SNaNs. That is clearly OK according to the wording in draft C23. C17 and earlier do not mention that possibility, but they do not really specify the behavior of SNaNs in general.
This will be used in conjunction with the new implementation of variable arguments, where they are not removed from the stack until the end of the function.
This is intended to be a complete implementation of strftime as specified in C17, although it lacks meaningful support for time zones or non-C locales.
This affects code where multiple structs or unions are assigned by successive = operators in one expression, e.g. "s1=s2=s3". The middle struct assignment(s) would use the ~Move2 or ~LongMove2 helper functions (for <64k or >=64k moves, respectively). These functions are supposed to leave the destination pointer on the stack so it can be used as the source of a subsequent move, but they both had bugs where they could modify dest and leave that modified value on the stack, which would cause subsequent moves to use the wrong source location. In the case of ~Move2, this only happened if the size was odd.
Here is a program that demonstrated the problems with both functions:
#pragma memorymodel 1
#include <stdio.h>
struct S1 {
char s[80000];
} a,b,c;
int main(void) {
struct S2 {
int x,y;
char z;
} d,e,f;
c.s[66000] = 123;
f.y = 5678;
a = b = c;
d = e = f;
printf("%i %i %i\n", a.s[66000], b.s[66000], c.s[66000]);
printf("%i %i %i\n", d.y, e.y, f.y);
}
The previous code relied on the data bank value at entry being the bank that contained the library code. This might not be the case when called from code using the large memory model, which could lead to the wrong value being returned.
This may be either 50 or 60, depending on the system's video frequency setting (50Hz PAL or 60Hz NTSC). The video setting can be determined by inspecting bit 4 of the LANGSEL soft switch, documented in Appendix E of the Apple IIGS Firmware Reference.
If the upper byte of the int argument was nonzero, it could write the wrong value (the OR of the upper and lower bytes). It should convert the value to unsigned char, i.e. just use the lower byte.
Previously, it could behave incorrectly in some cases where count was 0. This would happen if the address calculation &base[-1] wrapped around and produced a large number, either because base was NULL or because size was larger than the address of base.
This fixes#33.
The core loop uses the 65816's decimal mode to create a BCD representation of the number, then we convert that to ASCII to print it. The idea is based on some code from Kent Dickey, with adjustments to fit this codebase. This technique should be faster, and also should typically result in smaller code, since the long long division routine will not need to be linked into every program that uses printf.
For conversion to signed long long, we can almost just use SANE's FX2C call to convert it to comp. But LLONG_MIN corresponds to the bit pattern of the comp NaN, so it has to be treated specially. (You might think SANE would return the comp NaN when converting out-of-range values, which would give the value we want, but somewhat strangely it does not.)
For conversion to unsigned long long, we check if the value is greater than LLONG_MAX (which is also the max comp value). If so, we subtract LLONG_MAX+1 from the floating-point value before conversion and add it back in to the integer afterward.
This is almost the same as ~CMP8 in SysLib, except that it does not use a static location to save the DP. This makes it slightly smaller/faster, and also makes it thread/interrupt safe.
The new m16.int64 file contains a new "negate8" macro (for 64-bit negation), as well as an improved version of "ph8". These have been added to the individual *.macros files, but if those are regenerated, m16.int64 should be used as an input to macgen ahead of m16.ORCA (which contains the original version of ph8).
SANE does not have built in operations for this, but all the values are exactly representable in the SANE extended format, so we just do the conversion ourselves based on the definition of the extended format.
EOF should not be returned in this case.
I think it shouldn't actually write anything to the destination in this case, but it currently does. That problem remains unfixed for the moment; addressing it would require us to have our own internal buffer.
Conversions using it will not actually work right yet (unless assignment is suppressed), because the necessary changes have not been made in SysFloat, but the infrastructure is now in place to allow for those changes.