On Sun, Sep 28, 2025, 10:26 AM Hongtao Liu <[email protected]> wrote:
> On Sun, Sep 28, 2025 at 9:49 AM H.J. Lu <[email protected]> wrote: > > > > On Sun, Sep 28, 2025 at 9:29 AM Liu, Hongtao <[email protected]> > wrote: > > > > > > > > > > > > > -----Original Message----- > > > > From: H.J. Lu <[email protected]> > > > > Sent: Sunday, September 28, 2025 9:11 AM > > > > To: Hongtao Liu <[email protected]> > > > > Cc: Liu, Hongtao <[email protected]>; [email protected] > > > > Subject: Re: [PATCH] Deprecate -mmove-max= and related tuning. > > > > > > > > On Sun, Sep 28, 2025 at 9:08 AM Hongtao Liu <[email protected]> > wrote: > > > > > > > > > > On Sun, Sep 28, 2025 at 8:32 AM H.J. Lu <[email protected]> > wrote: > > > > > > > > > > > > On Sun, Sep 28, 2025 at 8:18 AM Liu, Hongtao < > [email protected]> > > > > wrote: > > > > > > > > > > > > > > > > > > > > > > > > > > > > > -----Original Message----- > > > > > > > > From: H.J. Lu <[email protected]> > > > > > > > > Sent: Sunday, September 28, 2025 6:33 AM > > > > > > > > To: Liu, Hongtao <[email protected]> > > > > > > > > Cc: [email protected] > > > > > > > > Subject: Re: [PATCH] Deprecate -mmove-max= and related > tuning. > > > > > > > > > > > > > > > > On Fri, Sep 26, 2025 at 9:40 AM liuhongt < > [email protected]> > > > > wrote: > > > > > > > > > > > > > > > > > > For memset, the size of used vector is decided by > > > > > > > > > MIN(MOVE_MAX_PIECES, STORE_MAX_PIECES). > > > > > > > > > Unless there's u-arch prefer big size vector for memcpy and > > > > > > > > > small size vector for memset, there's no need to have a > > > > > > > > > separate option or tune for it. > > > > > > > > > In general, x86 backend always prefer big size vector for > > > > > > > > > memset due to STLF issue. > > > > > > > > > So for maintaince convenience, the patch remove store_max > > > > > > > > > related tune and marked the mstore-max as deprecated. > > > > > > > > > > > > > > > > > > Bootstrapped and regtested on x86_64-pc-linux{-m32,}. > > > > > > > > > H.J Any comments? > > > > > > > > > > > > > > > > Does it change -mstore-max or -mmove-max? The subject says > > > > > > > > -mmove-max, but the commit message says -mstore-max. > > > > > > > > > > > > > > Typo, The patch only changes -mstore-max 1. It removes tune of > > > > > > > {AVX256,AVX512}_STORE_BY_PIECES. > > > > > > > 2. Deprecate mstore-max by aliasing it to -mmove-max and > issues a > > > > warning. > > > > > > > > > > > > Does the patch include tests for -mmove-max aliasing and the > warning > > > > message? > > > > > No, do we need it? It seems simple enough. > > > > > > > > Any change should have tests to verify that it is done correctly. > > > The change is also ready verified by adjusting testcase, those memset > related testcases will fail due to new warning message. > > > > I didn't see the -mstore-max tests. The new warning message should be > > part of the tests. > Ok, Will add a testcase like > > /* { dg-do compile } */ > /* { dg-options "-O2 -march=x86-64-v4 -mstore-max=256" } */ > // { dg-warning ".-mstore-max=. is deprecated; use .-mmove-max=. > instead" "" { target *-*-* } 0 } > Can you add a warning message to the existing -mstore-max test? > > > > > > > > > > > -- > > > > H.J. > > > > > > > > -- > > H.J. > > > > -- > BR, > Hongtao > >
