Handling of upper bits for array indices wider than 32 bits

[larsclausen]

The current behavior for array indices wider than 32 bits is that the upper bits are silently truncated and the lower bits are used as the index E.g.

reg [32:0] i = {1'b1,32'h1};
integer a[2];
a[1] = 10;
$display(a[i]); // -> 10

In my opinion this should be treated as out-of-bounds instead. And it is if the index is constant. E.g.

integer a[2];
a[1] = 10;
$display(a[{1'b1,32'h1}]); // -> x

But we have a test that explicitly checks that upper bits for dynamic indices are discarded.

@caryr its been a while, but you created that test. Any thoughts on this?

[martinwhitaker]

I would be inclined to agree with you, but a quick test on some other simulators indicates they also truncate, either silently or with a warning. I found this discussion which relates, but doesn't reach any conclusion.

[caryr]

I'll need to take a look. I don't remember the details without more context, but I think we are implicitly assuming that all this eventually gets flattened, normalized and limited to something that is addressed by 32-bits. Maybe the limiting is happening too early. There's also the issue that we do not use types with explicit widths (e.g. uint32_t, int64_t, etc.) so there is some variation depending on the environment.