[aarch64-port-dev ] Weird thing in c1 compiler code

[Andrew Haley]

On 19/08/14 09:31, Edward Nevill wrote:
> On Mon, 2014-08-18 at 14:49 +0100, Andrew Dinn wrote:
>> On 08/08/14 10:08, Andrew Haley wrote:
>>> On 07/08/14 21:59, Andrew Haley wrote:
>>>>> Specifically, why is the else if branch testing for T_DOUBLE?
>>>> It shouldn't be: AFAICS that case is unreachable.
>>>
>>> Ah, but it might not be: it is possible to stash a double in an
>>> integer register.  That might be why it's there.
>>
>> Ok, that does indeed make sense -- although the same case is not present
>> in the x86 code yet the same possibility is there with x86, no? (also,
>> mutatis mutandis, with a float value and an integer register).
>>
>> I was just wondering if whoever added this case had a good reason why.
>> Ed, maybe?
> 
> Beats me.
> 
> The change was added by Andrew on May 22 last year.
> 
> # HG changeset patch
> # User aph
> # Date 1369218144 -3600
> #      Wed May 22 11:22:24 2013 +0100
> # Branch aarch64_c1
> # Node ID 31a822a80cae49157ef2c02baabfd5018822b171
> # Parent  9baee4e65ac5665fa3d8c6ba75d6b58018c317f4
> More minor C1 bug fixes.
> 
> ...
> @@ -688,7 +703,7 @@
>      if (type == T_ARRAY || type == T_OBJECT) {
>        __ str(src->as_register(), frame_map()->address_for_slot(dest->single_stack_ix()));
>        __ verify_oop(src->as_register());
> -    } else if (type == T_METADATA) {
> +    } else if (type == T_METADATA || type == T_DOUBLE) {
>        __ str(src->as_register(), frame_map()->address_for_slot(dest->single_stack_ix()));
>      } else {
>        __ strw(src->as_register(), frame_map()->address_for_slot(dest->single_stack_ix()));
> ...
> 
> The change seems to have been added as a specific change. IE It does not seem to be the result of an accidental merge.
> 
> I guess the reasons are just lost in history.

I don't think so: I wouldn't have done it unless it actually triggered.  I was
careful with C1 to make sure that all those cases were executed, at least once.

Andrew.