|
From: | Dmitry Antipov |
Subject: | Re: Benchmarking temporary Lisp objects [Was: Re: [RFC] temporary Lisp_Strings] |
Date: | Fri, 05 Sep 2014 13:28:43 +0400 |
User-agent: | Mozilla/5.0 (X11; Linux x86_64; rv:31.0) Gecko/20100101 Thunderbird/31.0 |
On 09/05/2014 08:24 AM, Stefan Monnier wrote:
Your benchmark for `cons' showed that there's some potential benefit, but then we have to figure out which Fcons calls can be replaced by alloca ones, and then assess whether the result is worth the effort (and the risk, since such alloca-allocated thingies need to be handled with care, making sure they can't escape to Elisp, and also using the stack more intensively increases the risk of crashing into the stack depth limit).
Surely this is a bit risky and requires careful looking how this object is used. On the other side, it's not too hard to debug; use of local object outside of its scope causes the very likely eassert/crash at next GC, and the fault address is very different from what we see with heap objects. Dmitry
[Prev in Thread] | Current Thread | [Next in Thread] |