================
@@ -1169,7 +1169,11 @@
DEFINE_LIBUNWIND_FUNCTION(_ZN9libunwind15Registers_riscv6jumptoEv)
ILOAD x\i, (RISCV_ISIZE * \i)(a0)
.endr
// skip a0 for now
+#if defined(__riscv_abi_rve)
----------------
lenary wrote:
> The non-e case seems to be saving/restoring all registers (just like every
> other architecture)? I'm not sure where you're seeing that it doesn't save
> all of them?
Sorry, my fault, I didn't realise there were two loops in this function and
thus two different lists of registers.
>
> The reason the lists aren't symmetric is because register a0 is x10, so
> restoring it is deferred until the end. I'm not sure why it doesn't just use
> one list, but this pattern is also used for MIPS and LoongArch.
Yeah, I'm happy with this.
Broadly, going back to your/craig's original question: I feel this should be
keyed on architecture, not ABI. This function is restoring all registers,
including temporaries, not just caller-saves, so on the e-abi it should be
restoring all temporaries as well. (This is different to my change to the
save/restore routines, which only touch callee-saves).
https://github.com/llvm/llvm-project/pull/98855
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
