[PATCH] Use 64-bit atomic on sem_t only with 8-byte alignment [BZ #33632]

[Wilco Dijkstra]

Hi Andreas/HJ,

> It's not universally true that you need 8-byte alignment for 64-bit
> atomic operation.

I believe it is true for the targets that GLIBC supports.

I like the direction of this since it allows us to reenable the configure check
sooner rather than later.

-#if USE_64B_ATOMICS
+#if USE_64B_ATOMICS && SEM_T_ALIGN >= 8

Wouldn't it be better to use a single define with all the preconditions?

Cheers,
Wilco