For JIT applications, the code and .eh_frame maybe placed very far apart.
The large code model for AArch64 should handle FDE references more than +/-2GB range.
Details
Diff Detail
Event Timeline
llvm/test/MC/ELF/cfi-large-model.s | ||
---|---|---|
8–11 | -triple aarch64 is sufficient. This tests generic ELF behavior. Add -triple aarch64_be. You can move aarch64 before x86_64 to be closer (oh,yes,x86_64->powerpc64le is an inversion pair) to an alphabetical order. |
llvm/test/MC/ELF/cfi-large-model.s | ||
---|---|---|
8–11 | using generic aarch64 and triple aarch64_be added. Here I keep the original reverse order so that it don't need to change the order of the large chunk of test code |
I realized I just ran into this issue:) I could not add a test for large code model because it does not use sdata8 yet: 3073a3aa1ef1ce8c9cac9b97a8e5905dd8779e16
Sorry, I think it is better dumping the output with llvm-dwarfdump --eh-frame, instead of inspecting the raw bytes.
-triple aarch64 is sufficient. This tests generic ELF behavior. Add -triple aarch64_be.
You can move aarch64 before x86_64 to be closer (oh,yes,x86_64->powerpc64le is an inversion pair) to an alphabetical order.