On 2/20/24 07:21, Alexandre Oliva wrote:
On Feb 20, 2024, Jeff Law <jeffreya...@gmail.com> wrote:
On 2/19/24 21:26, Alexandre Oliva wrote:
This backport for gcc-13 is required for pr90838.c to get the expected
count of andi instructions on riscv64-elf
.
In general, shouldn't backports be focused on correctness issues?
*nod*.
It's unclear what the motivation is for backporting this change into
gcc-13.
There's this unexpected fail in gcc-13 (pr90838.c), one out of a handful
that we've hit while transitioning our riscv toolchains to gcc-13.
I set out to understand them, I identified the patches that got them to
pass in the trunk, and so I've proposed their backports to fix the fails
in gcc-13.
Surely there are other ways to address each one of the fails.
But even if we choose to just xfail them, or leave them failing noisily,
I've already gone through the process of identifying the fix, so I
figured I might as well share it.
Thanks for explaining things. I had a feeling the motivation might be
something along those lines.
I'd tend to think we don't want this backported. It doesn't fix any
correctness issue and the performance impact is small. I also don't
expect gcc-13 is going to be of significant long term interest in the
RISC-V space as it predates any RVV support.
So this feels like it ought to be left as-is on the gcc-13 branch.
jeff
