https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119455
James K. Lowden <jklowden at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Severity|normal |enhancement
Assignee|unassigned at gcc dot gnu.org |rdubner at gcc dot
gnu.org
CC| |jklowden at gcc dot gnu.org
--- Comment #2 from James K. Lowden <jklowden at gcc dot gnu.org> ---
> the GCC COBOL frontend seems to use something like a type/object descriptor
> for variables and constants and dispatching to the GCC COBOL runtime for all
> operations. That means the static types are not exposed to GCCs
> optimizations
Bob can speak to this better than I can, but your observation apt. We always
knew that when our work was scrutinized by GCC, there would be (ahem)
suggestions. All to the good! I fully expected that GCC feedback would
improve the FE, and COBOL's unique requirements may likewise induce changes to
the middle-end.
I can say for sure it won't be difficult to change, because I won't be doing
any of the work.
Yes, COBOL is statically typed. But it has a much richer type system than any
other language I know of. One way to look at it is that each variable (in
COBOL, "data item") has a unique type. Much COBOL data is organized in
hierarchical records, sometimes with fixed offsets that are not word-aligned.
Numeric types, besides binary integers and floating point, it has a fixed-point
decimal type of arbitrary size with an implied decimal point, and the sign can
be represented in several ways. For all arithmetic of all types, rounding can
be done in 7 ways for any single operation. There is also Packed Decimal,
similar to BCD at a high level, but different in detail.
Known as "Numeric Display", decimal types use one character for each digit.
Because it's a commonly used feature, the COBOL FE has its own optimized
computation routines for it.
There are "edited" types (alphanumeric and numeric) that have embedded static
data interleaved. Character strings have known size and encoding, not
something fobbed off to the runtime and subject to the environment. MOVE
CORRESPONDING exists in no other language I know of.
COBOL variables have metadata that is used at runtime. Certain alphanumeric
positions can be constrained to be digits. The initial value, if supplied, can
be restored via INITIALIZE. A variable can have an enumerated domain. Unlike
C, it's not restricted to integers, and one value in the domain can be defined
as "false".
How all this is defined in GENERIC is a years-long improvisation. Memory is
allocated per-record, and individual data items are pointers into that record.
IIUC, the concern is how those data-items are defined. It may that some are
ripe for optimization if defined differently and others cannot be, and still
other could be, if GENERIC/Gimple were enhanced.
The instant case of moving 'a' to a single-byte variable does look like a
textbook example of "ripe". We will give it a look.
