public inbox for gcc-bugs@sourceware.org help / color / mirror / Atom feed
From: "rguenth at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org> To: gcc-bugs@gcc.gnu.org Subject: [Bug tree-optimization/112331] Fail vectorization after loop interchange Date: Thu, 02 Nov 2023 10:48:09 +0000 [thread overview] Message-ID: <bug-112331-4-JcCuAyNiej@http.gcc.gnu.org/bugzilla/> (raw) In-Reply-To: <bug-112331-4@http.gcc.gnu.org/bugzilla/> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112331 --- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> --- I'm not sure what the problem is with a zero DR step for an inner loop reference (possibly dependence analysis runs into some unhandled cases - who knows). The following vectorizes the inner loop (the load is hoisted as invariant, but the store is not sunk - there's no sinking phase after interchange). diff --git a/gcc/tree-vect-data-refs.cc b/gcc/tree-vect-data-refs.cc index d5c9c4a11c2..7d1f0697fe7 100644 --- a/gcc/tree-vect-data-refs.cc +++ b/gcc/tree-vect-data-refs.cc @@ -2944,6 +2944,7 @@ vect_analyze_data_ref_access (vec_info *vinfo, dr_vec_info *dr_info) DR_GROUP_FIRST_ELEMENT (stmt_info) = NULL; if (!nested_in_vect_loop_p (loop, stmt_info)) return DR_IS_READ (dr); +#if 0 /* Allow references with zero step for outer loops marked with pragma omp simd only - it guarantees absence of loop-carried dependencies between inner loop iterations. */ @@ -2954,6 +2955,7 @@ vect_analyze_data_ref_access (vec_info *vinfo, dr_vec_info *dr_info) "zero step in inner loop of nest\n"); return false; } +#endif } if (loop && nested_in_vect_loop_p (loop, stmt_info)) Note when we don't vectorize we are eliding the inner loop later, when we vectorize we don't. unvectorized: s111: .LFB0: .cfi_startproc xorl %eax, %eax .L2: movl Y(%rax), %ecx addq $4, %rax leal 1(%rcx), %edx movl %edx, X-4(%rax) cmpq $128000, %rax jne .L2 xorl %eax, %eax ret vectorized: s111: .LFB0: .cfi_startproc movdqa .LC0(%rip), %xmm1 xorl %ecx, %ecx .L2: movdqa Y(%rcx), %xmm0 leaq X(%rcx), %rdx movl $400000, %eax paddd %xmm1, %xmm0 .p2align 4,,10 .p2align 3 .L3: movaps %xmm0, (%rdx) subl $2, %eax jne .L3 addq $16, %rcx cmpq $128000, %rcx jne .L2 xorl %eax, %eax ret
prev parent reply other threads:[~2023-11-02 10:48 UTC|newest] Thread overview: 6+ messages / expand[flat|nested] mbox.gz Atom feed top 2023-11-01 9:20 [Bug c/112331] New: middle-end: Fail vectorization juzhe.zhong at rivai dot ai 2023-11-01 9:24 ` [Bug c/112331] " juzhe.zhong at rivai dot ai 2023-11-02 7:42 ` [Bug tree-optimization/112331] " juzhe.zhong at rivai dot ai 2023-11-02 10:00 ` [Bug tree-optimization/112331] Fail vectorization after loop interchange rguenth at gcc dot gnu.org 2023-11-02 10:14 ` juzhe.zhong at rivai dot ai 2023-11-02 10:48 ` rguenth at gcc dot gnu.org [this message]
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-112331-4-JcCuAyNiej@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).