Now with *almost* working EH codegen. Does not compile Phobos yet
because we run into the "instruction does not dominate all uses"
issue when an r-value result of toElemDtor is used and we need to
run cleanups in between. Should easily be fixed by promoting those
values to allocas.
Most of the changes outside of ir/irfunction.{h, cpp} are just
because CreateCallOrInvoke moved locations. I took the
opportunity to also make use of the different arg count
overloads where possible.
In case of a singleobj build and a provided output file (.exe or .obj).
The dmd-testsuite tests used to produce lots of .o files while expecting
.obj ones as specified in the command line.
0.15.1 used stdcall convention everywhere. But the ABI change in 0.15.2
requires use of the C calling convention for variadic calls on x86
because the stdcall convention does not support vararg functions.
Solution is to check the llvm::FunctionType if the function has varargs.
This commit also makes the C calling convention the global default
because according to the LLVM documentation the fastcc convention used
for D linkage does not support varargs.
This fixes issue #1000.
This is likely an LLVM bug: The code in InlineFunctions.cpp
only updates the source location metadata for the instructions
in the inlined callee when the call site has a source location
set. When the caller has no location metadata, the scope for
the inlined instructions thus still points to the DISubprogram
for the original callee. This then leads to an assertion during
codegen.
GitHub: Fixes#998.
If a target file name is given (`-of`) and a library is created then
`-singleObj` is implied. This results in behaviour not compatible
with dmd. This commit checks for this situation and does not set
`-singleObj`. This fixes issue #978.
- Do not create debug info after we called finalize.
- Clean up debug info generation for module ctor/dtor.
- Rename EmitSubProgramInternal() to EmitModuleCTor().
- Remove unused parameter from CreateTypeDescription().
Also being only a flag, it is safer to use the i8 type and compare against 0.
This adds only an icmp instruction to the IR and should not change the generated machine code.
The performance impact should be completely immeasurable,
but the main benefit is that this makes IR dumps slightly
nicer to look at (as the redundant "empty:" labels are
now omitted entirely).
The new module discovery scheme requires the following section order:
.minfo_beg
.minfo
.minfo_end
This works for non-PIC code because the segments have the same attributes.
However, if -relocation-model=pic is passed to ldc2 then the .minfo section
becomes writeable and the sequence of sections is changed.
The quick fix is to mark the data always as writeable. Then all sections are
always writeable.
I think a better solution would be to base this in the used relocation model.
But this information is currently only available in the driver.
This fixes the failure in test case runnable/eh2.d
