[COMMITTED] PowerPC: Add support for RFC02653 - Dense Math Facility
Tom de Vries
tdevries@suse.de
Thu Nov 3 06:33:50 GMT 2022
On 11/2/22 20:25, Peter Bergner wrote:
> On 11/2/22 1:41 PM, Tom de Vries wrote:
>> On 11/2/22 19:32, Peter Bergner wrote:
>>> On 11/2/22 1:27 PM, Tom de Vries wrote:
>>>> The disassembly shows:
>>>> ...
>>>> 0x000000000000063c <+1596>: pmdmxvbf16ger2 a4,vs0,vs1,0,0,0^M
>>>> ...
>>>> while the test expects pmxvbf16ger2.
>>>
>>> Yes, Carl notified me of the fallout. I gave him a patch that updates
>>> the expected output to look for the new name. Carl has agreed to
>>> shepherd the patch to fix the fallout.
>>
>> Could someone explain why that is the right solution?
>>
>> I don't see any dm<n> registers in the instruction "pmxvbf16ger2 a4,vs0,vs1,0,0,0".
>>
>> Why then should the pmdmxvbf16ger2 name be used?
>
> a4 (ie, accumulator register 4) is a DMR register...more specifically,
> half of a DMR register. The two RFCs which I pushed support for changed
> the "official" ISA mnemonic to the *dm* name because that is more
> architecturally correct. We kept the old names as extended mnemonics,
> so the only visible change to users is that the disassembly behavior
> has changed, hence the need for the change to the expected output.
Understood, thanks for the explanation.
- Tom
More information about the Binutils
mailing list