public inbox for gcc-bugs@sourceware.org help / color / mirror / Atom feed
From: "cvs-commit at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org> To: gcc-bugs@gcc.gnu.org Subject: [Bug rtl-optimization/100328] IRA doesn't model matching constraint well Date: Tue, 06 Jul 2021 02:35:19 +0000 [thread overview] Message-ID: <bug-100328-4-WXmuS0xN84@http.gcc.gnu.org/bugzilla/> (raw) In-Reply-To: <bug-100328-4@http.gcc.gnu.org/bugzilla/> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100328 --- Comment #9 from CVS Commits <cvs-commit at gcc dot gnu.org> --- The master branch has been updated by Kewen Lin <linkw@gcc.gnu.org>: https://gcc.gnu.org/g:8ffe25eefae57fb3a228a2d31a57af5bdab8911f commit r12-2045-g8ffe25eefae57fb3a228a2d31a57af5bdab8911f Author: Kewen Lin <linkw@linux.ibm.com> Date: Mon Jul 5 20:53:19 2021 -0500 ira: Support more matching constraint forms with param [PR100328] This patch is to make IRA consider matching constraint heavily, even if there is at least one other alternative with non-NO_REG register class constraint, it will continue and check matching constraint in all available alternatives and respect the matching constraint with preferred register class. One typical case is destructive FMA style instruction on rs6000. Without this patch, for the mentioned FMA instruction, IRA won't respect the matching constraint on VSX_REG since there are some alternative with FLOAT_REG which doesn't have matching constraint. It can cause extra register copies since later reload has to make code to respect the constraint. This patch make IRA respect this matching constraint on VSX_REG which is the preferred regclass, but it excludes some cases where for one preferred register class there can be two or more alternatives, one of them has the matching constraint, while another doesn't have. It also considers the possibility of free register copy. With option Ofast unroll, this patch can help to improve SPEC2017 bmk 508.namd_r +2.42% and 519.lbm_r +2.43% on Power8 while 508.namd_r +3.02% and 519.lbm_r +3.85% on Power9 without any remarkable degradations. It also improved something on SVE as testcase changes showed and Richard's confirmation. Bootstrapped & regtested on powerpc64le-linux-gnu P9, x86_64-redhat-linux and aarch64-linux-gnu. gcc/ChangeLog: PR rtl-optimization/100328 * doc/invoke.texi (ira-consider-dup-in-all-alts): Document new parameter. * ira.c (ira_get_dup_out_num): Adjust as parameter param_ira_consider_dup_in_all_alts. * params.opt (ira-consider-dup-in-all-alts): New. * ira-conflicts.c (process_regs_for_copy): Add one parameter single_input_op_has_cstr_p. (get_freq_for_shuffle_copy): New function. (add_insn_allocno_copies): Adjust as single_input_op_has_cstr_p. * ira-int.h (ira_get_dup_out_num): Add one bool parameter. gcc/testsuite/ChangeLog: PR rtl-optimization/100328 * gcc.target/aarch64/sve/acle/asm/div_f16.c: Remove one xfail. * gcc.target/aarch64/sve/acle/asm/div_f32.c: Likewise. * gcc.target/aarch64/sve/acle/asm/div_f64.c: Likewise. * gcc.target/aarch64/sve/acle/asm/divr_f16.c: Likewise. * gcc.target/aarch64/sve/acle/asm/divr_f32.c: Likewise. * gcc.target/aarch64/sve/acle/asm/divr_f64.c: Likewise. * gcc.target/aarch64/sve/acle/asm/mad_f16.c: Likewise. * gcc.target/aarch64/sve/acle/asm/mad_f32.c: Likewise. * gcc.target/aarch64/sve/acle/asm/mad_f64.c: Likewise. * gcc.target/aarch64/sve/acle/asm/mla_f16.c: Likewise. * gcc.target/aarch64/sve/acle/asm/mla_f32.c: Likewise. * gcc.target/aarch64/sve/acle/asm/mla_f64.c: Likewise. * gcc.target/aarch64/sve/acle/asm/mls_f16.c: Likewise. * gcc.target/aarch64/sve/acle/asm/mls_f32.c: Likewise. * gcc.target/aarch64/sve/acle/asm/mls_f64.c: Likewise. * gcc.target/aarch64/sve/acle/asm/msb_f16.c: Likewise. * gcc.target/aarch64/sve/acle/asm/msb_f32.c: Likewise. * gcc.target/aarch64/sve/acle/asm/msb_f64.c: Likewise. * gcc.target/aarch64/sve/acle/asm/mulx_f16.c: Likewise. * gcc.target/aarch64/sve/acle/asm/mulx_f32.c: Likewise. * gcc.target/aarch64/sve/acle/asm/mulx_f64.c: Likewise. * gcc.target/aarch64/sve/acle/asm/nmad_f16.c: Likewise. * gcc.target/aarch64/sve/acle/asm/nmad_f32.c: Likewise. * gcc.target/aarch64/sve/acle/asm/nmad_f64.c: Likewise. * gcc.target/aarch64/sve/acle/asm/nmla_f16.c: Likewise. * gcc.target/aarch64/sve/acle/asm/nmla_f32.c: Likewise. * gcc.target/aarch64/sve/acle/asm/nmla_f64.c: Likewise. * gcc.target/aarch64/sve/acle/asm/nmls_f16.c: Likewise. * gcc.target/aarch64/sve/acle/asm/nmls_f32.c: Likewise. * gcc.target/aarch64/sve/acle/asm/nmls_f64.c: Likewise. * gcc.target/aarch64/sve/acle/asm/nmsb_f16.c: Likewise. * gcc.target/aarch64/sve/acle/asm/nmsb_f32.c: Likewise. * gcc.target/aarch64/sve/acle/asm/nmsb_f64.c: Likewise. * gcc.target/aarch64/sve/acle/asm/sub_f16.c: Likewise. * gcc.target/aarch64/sve/acle/asm/sub_f32.c: Likewise. * gcc.target/aarch64/sve/acle/asm/sub_f64.c: Likewise. * gcc.target/aarch64/sve/acle/asm/subr_f16.c: Likewise. * gcc.target/aarch64/sve/acle/asm/subr_f32.c: Likewise. * gcc.target/aarch64/sve/acle/asm/subr_f64.c: Likewise.
next prev parent reply other threads:[~2021-07-06 2:35 UTC|newest] Thread overview: 12+ messages / expand[flat|nested] mbox.gz Atom feed top 2021-04-29 6:41 [Bug rtl-optimization/100328] New: IRA doesn't model dup num " linkw at gcc dot gnu.org 2021-04-30 8:30 ` [Bug rtl-optimization/100328] " linkw at gcc dot gnu.org 2021-06-23 18:28 ` [Bug rtl-optimization/100328] IRA doesn't model matching " vmakarov at gcc dot gnu.org 2021-06-24 12:11 ` linkw at gcc dot gnu.org 2021-06-24 12:36 ` linkw at gcc dot gnu.org 2021-06-28 3:27 ` linkw at gcc dot gnu.org 2021-06-28 5:25 ` linkw at gcc dot gnu.org 2021-06-29 16:01 ` rsandifo at gcc dot gnu.org 2021-07-01 6:18 ` linkw at gcc dot gnu.org 2021-07-06 2:35 ` cvs-commit at gcc dot gnu.org [this message] 2021-07-06 2:35 ` cvs-commit at gcc dot gnu.org 2021-07-07 3:07 ` linkw at gcc dot gnu.org
Reply instructions: You may reply publicly to this message via plain-text email using any one of the following methods: * Save the following mbox file, import it into your mail client, and reply-to-all from there: mbox Avoid top-posting and favor interleaved quoting: https://en.wikipedia.org/wiki/Posting_style#Interleaved_style * Reply using the --to, --cc, and --in-reply-to switches of git-send-email(1): git send-email \ --in-reply-to=bug-100328-4-WXmuS0xN84@http.gcc.gnu.org/bugzilla/ \ --to=gcc-bugzilla@gcc.gnu.org \ --cc=gcc-bugs@gcc.gnu.org \ /path/to/YOUR_REPLY https://kernel.org/pub/software/scm/git/docs/git-send-email.html * If your mail client supports setting the In-Reply-To header via mailto: links, try the mailto: linkBe sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox; as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).