On 02/09/2020 18:49, Tom Tromey wrote:
"Andrew" == Andrew Stubbs <a...@codesourcery.com> writes:
Andrew> To be fair, the DWARF standard makes a similar assumption; the
Andrew> engineers working on LLVM and GDB, at AMD, have therefore invented
Andrew> some new DWARF operators that they plan to propose for a future
Andrew> standard. Only one is relevant here, however:
Andrew> DW_OP_LLVM_piece_end. (Unfortunately this clashes with an AArch64
Andrew> extension, but I think we can cope using an alias -- only GCC dumps
Andrew> will be confusing.)
Andrew> +/* AMD GCN extensions (originally for LLVM). */
Andrew> +// This clashes with DW_OP_AARCH64_operation, so use an alias instead
Andrew> +// DW_OP (DW_OP_LLVM_piece_end, 0xea)
Andrew> +#define DW_OP_LLVM_piece_end DW_OP_AARCH64_operation
Andrew> DW_END_OP
Is it too late to pick a non-clashing value?
Also, we have tried pretty hard in recent years to document all of gcc's
DWARF extensions. Please add a link to the documentation for this one.
If there aren't docs -- I guess it would be ideal if you could write
them. Putting them on the GCC wiki is fine.
I didn't select this number; I'm just following out-of-tree LLVM and GDB
usage from AMD. It's allocated from the user extension range, so I
imagine it'll enter the standard with a different number (and probably
name).
The documentation is here:
http://llvm.org/docs/AMDGPUDwarfExtensionsForHeterogeneousDebugging.html#composite-location-description-operations
Andrew
