This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH][AArch64] Change FP reassociation width
- From: Wilco Dijkstra <Wilco dot Dijkstra at arm dot com>
- To: Richard Earnshaw <Richard dot Earnshaw at arm dot com>, GCC Patches <gcc-patches at gcc dot gnu dot org>
- Cc: nd <nd at arm dot com>, James Greenhalgh <James dot Greenhalgh at arm dot com>
- Date: Tue, 13 Jun 2017 09:43:05 +0000
- Subject: Re: [PATCH][AArch64] Change FP reassociation width
- Authentication-results: sourceware.org; auth=none
- Authentication-results: spf=none (sender IP is ) smtp.mailfrom=Wilco dot Dijkstra at arm dot com;
- Nodisclaimer: True
- References: <AM5PR0802MB2610946B99FCF37A2BB819BD83CD0@AM5PR0802MB2610.eurprd08.prod.outlook.com>,<acd5c386-03e7-9fa6-5c2e-9fc89e5f1a74@arm.com>
- Spamdiagnosticmetadata: NSPM
- Spamdiagnosticoutput: 1:99
Richard Earnshaw (lists) wrote:
>
> Why 1 and not 2? Many processors have 2 fp pipes and forcing this down
> to a sequential stream is not obviously the right thing.
1 was faster than 2. Like I said, the reassociation is too aggressive and even
splits multiply-add rather than keeping them. Until reassociation is fixed and
able to split a complex expression into 2 independent chains of similar depth
while keeping multiply-accumulate operations, it is best to set it to 1 for now.
> If reassociation is is causing excess spilling, then the right fix for
> that is to look at the pressure model, not hammer the problem away.
The problem of the GCC scheduler hugely increasing register pressure has
existed for many years with no progress being made. Unless we're willing to
start a project improving this I do not believe there will be a solution any time
soon. So changing the association width is the best solution for the time being.
Wilco