| Message ID | 20230523223844.719056-1-richard.henderson@linaro.org |
|---|---|
| State | New |
| Headers | show |
| Series | meson: Adjust check for __int128_t | expand |
On Tue, 23 May 2023 at 23:39, Richard Henderson <richard.henderson@linaro.org> wrote: > > Remove the signed * signed check, leaving the signed * unsigned check. > This link test runs foul of -fsanitize=undefined, where clang-11 has > an undefined reference to __muloti4 to check for signed overflow. If you can't do a signed * signed multiply then that sounds to me like "int128_t doesn't work on this compiler". We specifically added this check to catch "some clang with -fsanitize=undefined don't actually correctly compile int128_t * int128_t" in commit 464e3671f9. -- PMM
Queued, thanks. Paolo
On Wed, 24 May 2023 at 13:32, Paolo Bonzini <pbonzini@redhat.com> wrote: > > Queued, thanks. Can you unqueue it, please? This is reverting a fix we deliberately put in in commit 464e3671f9d5c. thanks -- PMM
On 5/24/23 12:30, Peter Maydell wrote: > On Tue, 23 May 2023 at 23:39, Richard Henderson > <richard.henderson@linaro.org> wrote: >> >> Remove the signed * signed check, leaving the signed * unsigned check. >> This link test runs foul of -fsanitize=undefined, where clang-11 has >> an undefined reference to __muloti4 to check for signed overflow. > > If you can't do a signed * signed multiply then that sounds > to me like "int128_t doesn't work on this compiler". We > specifically added this check to catch "some clang with > -fsanitize=undefined don't actually correctly compile > int128_t * int128_t" in commit 464e3671f9. Oops, missed this reply before queuing (and will leave the patch aside until the discussion is over). As long as -fsanitize=undefined is properly covered by CI and we don't need int128_t * int128_t multiplication, removing the test is ok IMO, though perhaps it should have a comment about why it's avoided. Paolo
On 5/24/23 14:33, Peter Maydell wrote: > On Wed, 24 May 2023 at 13:32, Paolo Bonzini <pbonzini@redhat.com> wrote: >> >> Queued, thanks. > > Can you unqueue it, please? This is reverting a fix we > deliberately put in in commit 464e3671f9d5c. Yes, I have done so. Paolo
diff --git a/meson.build b/meson.build index ef181ff2df..865bdd8d11 100644 --- a/meson.build +++ b/meson.build @@ -2542,10 +2542,8 @@ has_int128 = cc.links(''' int main (void) { a = a + b; b = a * b; - a = a * a; return 0; }''') - config_host_data.set('CONFIG_INT128', has_int128) if has_int128
Remove the signed * signed check, leaving the signed * unsigned check. This link test runs foul of -fsanitize=undefined, where clang-11 has an undefined reference to __muloti4 to check for signed overflow. This failure prevents us properly detecting atomic128 support. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- meson.build | 2 -- 1 file changed, 2 deletions(-)