|
From: | Paul Eggert |
Subject: | Re: Inlining policy |
Date: | Mon, 6 Apr 2020 11:26:16 -0700 |
User-agent: | Mozilla/5.0 (X11; Linux x86_64; rv:68.0) Gecko/20100101 Thunderbird/68.4.1 |
On 4/6/20 11:15 AM, Stefan Monnier wrote:
I haven't checked whether it has the same performance impact, and neither do I know how it compares w.r.t. "portability/reliability", but judging just from the patch, I really like it. Paul?
Although I like the idea, it significantly hurts performance with -O0 (GCC 9.3.1, x86-64) because it causes GCC to emit excess instructions to save and restore inline function arguments to the stack.
I'm testing a patch to use the idea for -Og but not -O0. Here it seems to help performance. Unfortunately this means we still need to keep those macros, for -O0.
[Prev in Thread] | Current Thread | [Next in Thread] |