Alloc of > half of address space is complicated

[nunoplopes]

LLVM does this (from @regehr):

define i1 @fn(i64 %x) {
  %m = malloc i64 %x
  %cmp = icmp eq * %m, null, use_provenance
  ret i1 %cmp
}
=>
define i1 @fn(i64 %x) {
  ret i1 0
}
Transformation doesn't verify!

ERROR: Value mismatch

Example:
i64 %x = #x8000000000000000 (9223372036854775808, -9223372036854775808)

Source:
* %m = null
i1 %cmp = #x1 (1)

Target:
Source value: #x1 (1)
Target value: #x0 (0)

Considering that LLVM sign-extends offsets in GEPs and etc, can we support such large allocations? How does the inbounds check looks like then?