Removed use of dyn_cast, llvm no compiles
without exceptions and rtti by
default. We do need exceptions for the libconfig stuff, but rtti isn't
necessary (anymore).
Debug info needs to be rewritten, as in LLVM 2.7 the format has
completely changed. To have something to look at while rewriting, the
old code has been wrapped inside #ifndef DISABLE_DEBUG_INFO , this means
that you have to define this to compile at the moment.
Updated tango 0.99.9 patch to include updated EH runtime code, which is
needed for LLVM 2.7 as well.
crucial difference being special handling of `llvm::Type`s so they get printed
by name rather than printing their full representation (which can be positively
*huge*).
This allows re-enabling some logger calls that were disabled due to extreme
verbosity.
or because it's a static array) by giving it the pointer type instead of the
type itself.
This fixes Derelict compilation on x86-64, where the ABI wasn't expecting a
static array to be passed in at all.
Added -v-cg switch, which right now just prints "codegen: module.name (module/name.d)" to stdout, this can really help figuring out where, in some complex build command, things go wrong.
parameter types.
This was intended to fix the following test case:
{{{
void delegate(Hit) dg;
struct Hit { void delegate(Hit) a; }
}}}
which ChristianK reduced from a problem downs had on IRC.
It also seems to fix mini/compile_delegate.d (which turns out to be quite
similar, but produced a different error message).
to D functions, we can apply noalias and nocapture. They are sret parameters,
'nest' pointers passed to nested functions, and _argptr:
Nocapture:
- Sret and nest are nocapture because they don't represent D-level variables,
and thus the callee can't (validly) obtain a pointer to them, let alone keep
it around after it returns.
- _argptr is nocapture because although the callee has access to it as a
pointer, that pointer is invalidated when it returns.
All three are noalias because they're function-local variables
- Sret and _argptr are noalias because they're freshly alloca'd memory only
used for a single function call that's not allowed to keep an aliasing
pointer to it around (since the parameter is nocapture).
- 'Nest' is noalias because the callee only ever has access to one such pointer
per parent function, and every parent function has a different one.
This commit also ensures attributes set on sret, _arguments and _argptr are
propagated to calls to such functions.
It also adds one exception to the general rule that attributes on function types
should propagate to calls: the type of a delegate's function pointer has a
'nest' parameter, but this can either be a true 'nest' (for delegates to nested
functions) or a 'this' (for delegates to member functions). Since 'this' is
neither noalias nor nocapture, and there's generally no way to tell which one it
is, we remove these attributes at the call site if the callee is a delegate.
- See through typedefs and enums in passByVal
- Don't depend on TypeFunction::parameters->dim being the actual number of
parameters; it contains unexpanded tuples as single elements.
- Implement x86-64 extern(C), hopefully correctly.
- Tried to be a bit smarter about extern(D) while I was there.
Interestingly, this code seems to be generating more efficient code than
gcc and llvm-gcc in some edge cases, like returning a `{ [7 x i8] }` loaded from
a stack slot from an extern(C) function. (gcc generates 7 1-byte loads, while
this code generates a 4-byte, a 2-byte and a 1-byte load)
I also added some changes to make sure structs being returned from functions or
passed in as parameters are stored in memory where the rest of the backend seems
to expect them to be. These should be removed when support for first-class
aggregates improves.
