On 8/26/25 05:26, Jakub Jelinek wrote:
On Mon, Aug 25, 2025 at 08:24:18PM -0600, Sandra Loosemore wrote:
On 8/25/25 00:13, Jakub Jelinek wrote:
Yes, GCC doesn't have it implemented fully, but that doesn't mean it should
be ripped off, the implementation should be simply finished.
Well, I have no idea how to make this work, nor any enthusiasm for devoting
a big block of time to this. So I've taken Tobias's suggestion of emitting
a "sorry" for the incomplete functionality instead of silently emitting bad
code. New part 2 patch attached. Is this OK?
I wouldn't emit sorry for targets where we already implement it "correctly",
i.e. at least all targets which have NULL
targetm.simd_clone.compute_vecsize_and_simdlen
There the argument transformation is identity.
That would be the targetm.xxx for the offload target, right? Is that
information actually available at parse time? Passing an invalid
call_expr all the way through to the offload compiler just so we can
scan for it later somehow and emit a "sorry" then seems like not such a
good idea to me.
-Sandra
