The branch range calculation treated dcl directives as taking 2 bytes rather than 4, which could result in out-of-range branches. These could result in linker errors (for forward branches) or silently generating wrong code (for backward branches).
This patch now treats dcb, dcw, and dcl as separate directives in the native-code layer, so the appropriate length can be calculated for each.
Here is an example of code affected by this:
int main(int argc, char **argv) {
top:
if (!argc) { /* this caused a linker error */
asm {
dcl 0
dcl 0
dcl 0
dcl 0
dcl 0
dcl 0
dcl 0
dcl 0
dcl 0
dcl 0
dcl 0
dcl 0
dcl 0
dcl 0
dcl 0
dcl 0
dcl 0
dcl 0
dcl 0
dcl 0
dcl 0
dcl 0
dcl 0
dcl 0
dcl 0
dcl 0
dcl 0
dcl 0
dcl 0
dcl 0
dcl 0
dcl 0
dcl 0
}
goto top; /* this generated bad code with no error */
}
}
Previously, the assembly-level optimizations applied to code in asm statements. In many cases, this was fine (and could even do useful optimizations), but occasionally the optimizations could be invalid. This was especially the case if the assembly involved tricky things like self-modifying code.
To avoid these problems, this patch makes the assembly optimizers ignore code from asm statements, so it is always emitted as-is, without any changes.
This fixes#34.
If one step of peephole optimization produced code that can be further optimized with more peephole optimizations, that additional optimization was not always done. This makes sure the additional optimization is done in several such cases.
This was particularly likely to affect functions containing asm blocks (because CheckLabels would never trigger rescanning in them), but could also occur in other cases.
Here is an example affected by this (generating inefficient code to load a[1]):
#pragma optimize 1
int a[10];
void f(int x) {}
int main(int argc, char **argv) {
if (argc) return 0;
f(a[1]);
}
This accords with its definition in the C standards. For the time being, the old form of three separate tokens is still accepted too, because the ... token may not be scanned correctly in the obscure case where there is a line continuation between the second and third dots.
One observable effect of this is that there are no longer spaces between the dots in #pragma expand output.
This enforces the constraint from C17 section 6.7 p2 that declarations "shall declare at least a declarator (other than the parameters of a function or the members of a structure or union), a tag, or the members of an enumeration."
Somewhat relaxed rules are used for enums in the default loose type checking mode, similar to what GCC and Clang do.
A declaration of this exact form always declares the tag T within the current scope, and as such makes this "struct T" a distinct type from any other "struct T" type in an outer scope. (Similarly for unions.)
See C17 section 6.7.2.3 p7 (and corresponding places in all other C standards).
Here is an example of a program affected by this:
struct S {char a;};
int main(void) {
struct S;
struct S *sp;
struct S {long b;} s;
sp = &s;
sp->b = sizeof(*sp);
return s.b;
}
They are now represented in local structures instead. This keeps the representation of declaration specifiers together and eliminates the need for awkward and error-prone code to save and restore the global variables.
This differs from the usual ORCA/C behavior of treating all floating-point parameters as extended. With the option enabled, they will still be passed in the extended format, but will be converted to their declared type at the start of the function. This is needed for strict standards conformance, because you should be able to take the address of a parameter and get a usable pointer to its declared type. The difference in types can also affect the behavior of _Generic expressions.
The implementation of this is based on ORCA/Pascal, which already did the same thing (unconditionally) with real/double/comp parameters.
The added characters are accented roman letters that were added to the Mac OS Roman character set at some time after it was first defined. Some IIGS fonts include them, although others do not.
The "known issue" about not issuing required diagnostics is removed because ORCA/C has gotten significantly better about that, particularly if strict type checking is enabled. There are still probably some diagnostics that are missed, but it is no longer a big enough issue to be called out more prominently than other bugs.
If strict type checking is enabled, this will prohibit redefinition of enums, like:
enum E {a,b,c};
enum E {x,y,z};
It also prohibits use of an "enum E" type specifier if the enum has not been previously declared (with its constants).
These things were historically supported by ORCA/C, but they are prohibited by constraints in section 6.7.2.3 of C99 and later. (The C90 wording was different and less clear, but I think they were not intended to be valid there either.)
This affects code like the following:
enum E {a,b,c};
int main(void) {
enum E e;
struct E {int x;}; /* or: enum E {x,y,z}; */
}
The line "enum E e;" should refer to the enum type declared in the outer scope, but not redeclare it in the inner scope. Therefore, a subsequent struct, union, or enum declaration using the same tag in the same scope is acceptable.
They now use a jmp (addr,X) instruction, rather than a more complicated code sequence using rts. This is an improvement that was suggested in an old Genie message from Todd Whitesel.
This avoids strangeness where an enum tag declared within a typedef declaration would act like a typedef. For example, the following would compile without error:
typedef enum E {a,b,c} T;
E e;
This covers code like the following, which is very dubious but does not seem to be clearly prohibited by the standards:
int main(void) {
void *vp;
*vp;
}
Previously, this would do an indirect load of a four-byte value at the location, but then treat it as void. This could lead to the four-byte value being left on the stack, eventually causing a crash. Now we just evaluate the pointer expression (in case it has side effects), but effectively cast it to void without dereferencing it.
This avoids any possible issue with code possibly expecting __FE_DFL_ENV to be in the data bank when using the large memory model, although I don't think that happened in practice.
This converts comparisons like x > N (with constant N) to instead be evaluated as x >= N+1, since >= comparisons generate better code. This is possible as long as N is not the maximum value in the type, but in that case the comparison is always false. There are also a few other tweaks to the generated code in some cases.
The optimization could turn an unsigned comparison "x <= 0xFFFF" into "x < 0".
Here is an example affected by this:
int main(void) {
unsigned i = 1;
return (i <= 0xffff);
}
The TAY instruction would set the flags, but that is unnecessary because pc_cnv is a "NeedsCondition" operation (and some other conversions also do not reliably set the flags).
The code is also changed to preserve the Y register, possibly facilitating register optimizations.
This was a problem introduced in commit 393b7304a0. It could cause a compiler error for unoptimized array indexing code, e.g.:
int a[100];
int main(void) {
return a[5];
}
The new value of maxLocalLabel is aligned with the C99+ requirement to support "511 identifiers with block scope declared in one block".
The value of maxLabel is now the maximum it can be while keeping the size of the labelTab array under 32 KiB. (I'm not entirely sure the address calculations in the code generated by ORCA/Pascal would work correctly beyond that.)
This affects 16-bit subtractions where where only the left operand is "complex" (i.e. most things other than constants and simple loads). They were using an unnecessarily complicated code path suitable for the case where both operands are complex.
These could occur because the code for certain operations was assumed to set the z flag based on the result value, but did not actually do so. The affected operations were shifts, loads or stores of bit-fields, and ? : expressions.
Here is an example showing the problem with a shift:
#pragma optimize 1
int main(void) {
int i = 1, j = 0;
return (i >> j) ? 1 : 0;
}
Here is an example showing the problem with a bit-field load:
struct {
signed int i : 16;
} s = {1};
int main(void) {
return (s.i) ? 1 : 0;
}
Here is an example showing the problem with a bit-field store:
#pragma optimize 1
struct {
signed int i : 16;
} s;
int main(void) {
return (s.i = 1) ? 1 : 0;
}
Here is an example showing the problem with a ? : expression:
#pragma optimize 1
int main(void) {
int a = 5;
return (a ? (a<<a) : 0) ? 0 : 1;
}
This optimizes most multiplications by a power of 2 or the sum of two powers of 2, converting them to equivalent operations using shifts which should be faster than the general-purpose multiplication routine.
This changes unsigned 16-bit multiplies to use the new ~CUMul2 routine in ORCALib, rather than ~UMul2 in SysLib. They differ in that ~CUMul2 gives the low-order 16 bits of the true result in case of overflow. The C standards require this behavior for arithmetic on unsigned types.
