This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH 3/6] Fix prefetch heuristic calculation.
- From: Richard Biener <richard dot guenther at gmail dot com>
- To: Maxim Kuvyrkov <maxim dot kuvyrkov at linaro dot org>
- Cc: GCC Patches <gcc-patches at gcc dot gnu dot org>, Kyrylo Tkachov <kyrylo dot tkachov at arm dot com>, Andrew Pinski <apinski at cavium dot com>
- Date: Mon, 30 Jan 2017 13:37:09 +0100
- Subject: Re: [PATCH 3/6] Fix prefetch heuristic calculation.
- Authentication-results: sourceware.org; auth=none
- References: <F7C2520D-866C-4293-831D-815BF466DFA2@linaro.org> <A3BE143E-DE33-4737-B99A-CA27ED4BD87E@linaro.org>
On Mon, Jan 30, 2017 at 12:43 PM, Maxim Kuvyrkov
<maxim.kuvyrkov@linaro.org> wrote:
> This patch fixes heuristic in loop array prefetching to use "round-to-nearest" instead of "floor", which is what _all_ other similar heuristics in the pass are doing.
_all_ is a bit over-exaggregating... In the context we are testing
this round value against a min_insn_to_mem_ratio so rounding
down makes sense to me.
What am I missing?
Richard.
> This subtle change allows a critical loop in 429.mcf to get prefetches without making the pass too aggressive, which causes regressions on other benchmarks.
>
> Bootstrapped and regtested on x86_64-linux-gnu and aarch64-linux-gnu.
>
> --
> Maxim Kuvyrkov
> www.linaro.org
>