On 30/09/2026 13:34, Richard Biener wrote:
On Mon, Sep 28, 2026 at 3:20 PM Tobias Burnus <[email protected]> wrote:
Andrew Stubbs wrote:
Although, I don't know why we want to chase all these when we have
gfx11-generic and the features aren't different (and we can't test it).
Besides the "for completeness",
I think that's reasonable.
I think it is mainly for users of
those GPUs that just pass the argument to the compiler. And instead
of
gcn mkoffload: error: unrecognized argument in option ‘-march=gfx1154’
it is much more helpful to get the following error message:
gcn mkoffload: fatal error: GCC was built without library support for
‘-march=gfx1154’; consider compiling for the associated generic
architecture ‘-march=gfx11-generic’ instead
Can we make that
gcn mkoffload: note: GCC was built without library support for
'-march=gfx1154'; using architecture '-march=gfx11-generic' instead
aka, fall back to gfx11-generic automagically for all gfx11*? Or do you
consider that too dangerous (well, not more than this untested patch?)
Is there currently _any_ benefit to using a special gfx11* target over
gfx11-generic?
There's no benefit that I'm aware of, for GFX11 anyway. Certainly the
compiler doesn't do anything different.
As for auto-fallback, that would complicate generating the specs,
perhaps, but we could do it. The only people that might care would be
those who have external libraries to link in (in which case the ELF
flags wouldn't match), but probably they would have already configure
the --with-multilib-list appropriately.
Andrew
