Rollup merge of #140072 - folkertdev:miri-fn-align, r=RalfJung

handle function alignment in miri

tracking issue: rust-lang/rust#82232
Fixes #4282

The `#[repr(align(N))]` attribute on functions was ignored when using miri. For such a function, its address should be a multiple of `N`.

There is some further discussion in the thread [#t-compiler/const-eval > function address alignment](https://rust-lang.zulipchat.com/#narrow/channel/146212-t-compiler.2Fconst-eval/topic/function.20address.20alignment) on how `dyn Fn` should be handled. The behavior there appears to be consistent between miri and nightly, though both may be incorrect. In any case, that can be resolved separately.

Author: ChrisDenton

tests/pass/fn_align.rs

@@ -0,0 +1,21 @@
+//@compile-flags: -Zmin-function-alignment=8
+#![feature(fn_align)]
+
+// When a function uses `repr(align(N))`, the function address should be a multiple of `N`.
+
+#[repr(align(256))]
+fn foo() {}
+
+#[repr(align(16))]
+fn bar() {}
+
+#[repr(align(4))]
+fn baz() {}
+
+fn main() {
+    assert!((foo as usize).is_multiple_of(256));
+    assert!((bar as usize).is_multiple_of(16));
+
+    // The maximum of `repr(align(N))` and `-Zmin-function-alignment=N` is used.
+    assert!((baz as usize).is_multiple_of(8));
+}