|
From: | Richard Henderson |
Subject: | Re: [PATCH v2 09/15] target/ppc: Move ADDI, ADDIS to decodetree, implement PADDI |
Date: | Wed, 28 Apr 2021 08:23:42 -0700 |
User-agent: | Mozilla/5.0 (X11; Linux x86_64; rv:78.0) Gecko/20100101 Thunderbird/78.7.1 |
On 4/28/21 7:10 AM, Matheus K. Ferst wrote:
In our first attempt, we did some efforts to keep prefixed instructions type 0b10 and 0b11 under the same implementation as their word-size counterpart, i.e. trans_ADDI and trans_PADDI had the same signature and just forwarded their arguments to a third method that does the real work. Is this kind of approach desirable? We initially achieved this by using const_elt to set r=0 for addi, which is not particularly nice, but we can look for other solutions.
Yes, I could have tried harder to share the implementation here. And in retrospect, using a &PLS_D argument set for the non-prefixed integer load/store insns would have been fairly easy, and reduce 30 lines of unnecessary duplication.
For the MMIRR prefixed instructions, that pain of duplication would be a lot higher.
r~
[Prev in Thread] | Current Thread | [Next in Thread] |