On 10/21/26 8:26 AM, Nathan Sidwell wrote:
On 20/10/2026 10:10, Jason Merrill wrote:
On 10/20/26 9:17 AM, Nathan Sidwell wrote:
In tandem with my riscv ECF flag patch I fell over this, where the C+
+ FE doesn't propagate the NO_VOPS flag. I happen to be using that
on our builtins, and we also explicitly declare our builtins (so that
code analysis tools are not confused).
Boom, optimizations stop working :(
I wasn't sure whether this was the best place or merge_attribute_bits
would be better. I went with here, as that more closely matches the
C FE, which is already propagating these bits.
Do the other two places that call merge_attribute_bits also want these
bits copied?
I should have looked more closely at the paths adjacent to those calls
-- explicit template specializations do not inherit from the general
template. And hence merge_attribute_bits is the right place.
Sounds good. And good to see you on gcc-patches!
Incidentally, please include the rationale (e.g. the three quoted
paragraphs from the OP) in the commit message before the ChangeLog entries.
Jason
