public inbox for gcc-bugs@sourceware.org help / color / mirror / Atom feed
From: "soap at gentoo dot org" <gcc-bugzilla@gcc.gnu.org> To: gcc-bugs@gcc.gnu.org Subject: [Bug tree-optimization/94617] New: Simple if condition not optimized Date: Thu, 16 Apr 2020 11:28:24 +0000 [thread overview] Message-ID: <bug-94617-4@http.gcc.gnu.org/bugzilla/> (raw) https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94617 Bug ID: 94617 Summary: Simple if condition not optimized Product: gcc Version: 10.0 Status: UNCONFIRMED Severity: normal Priority: P3 Component: tree-optimization Assignee: unassigned at gcc dot gnu.org Reporter: soap at gentoo dot org Target Milestone: --- Given the following C++ snippet const char* vanilla_bandpass(int a, int b, int x, const char* low, const char* high) { const bool within_interval { (a <= x) && (x < b) }; return (within_interval ? high : low); } GCC trunk yields with -O3 -march=znver2 the following assembly vanilla_bandpass(int, int, int, char const*, char const*): mov rax, r8 cmp edi, edx jg .L4 cmp edx, esi jge .L4 ret .L4: mov rax, rcx ret which is terrible. On the other hand, Clang emits vanilla_bandpass(int, int, int, char const*, char const*): cmp edx, esi cmovge r8, rcx cmp edi, edx cmovg r8, rcx mov rax, r8 ret which is a lot better. There exists an unbranched version for which I'm not 100% certain whether it's free of UB: #include <cstdint> const char* funky_bandpass(int a, int b, int x, const char* low, const char* high) { const bool within_interval { (a <= x) && (x < b) }; const auto low_ptr = reinterpret_cast<uintptr_t>(low) * (!within_interval); const auto high_ptr = reinterpret_cast<uintptr_t>(high) * within_interval; const auto ptr_sum = low_ptr + high_ptr; const auto* result = reinterpret_cast<const char*>(ptr_sum); return result; } which yields funky_bandpass(int, int, int, char const*, char const*): cmp edi, edx setle al cmp edx, esi setl dl and eax, edx mov edx, eax xor edx, 1 movzx edx, dl movzx eax, al imul rcx, rdx imul rax, r8 add rax, rcx ret which is jump-free and in practice executes at the same observable rate as Clang's assembly, but still looks needlessly complex. Clang manages to compile this code to the same assembly as vanilla_bandpass. Any chance of getting the optimizer ironed out for this?
next reply other threads:[~2020-04-16 11:28 UTC|newest] Thread overview: 12+ messages / expand[flat|nested] mbox.gz Atom feed top 2020-04-16 11:28 soap at gentoo dot org [this message] 2020-04-16 11:57 ` [Bug tree-optimization/94617] " rguenth at gcc dot gnu.org 2020-04-16 12:02 ` soap at gentoo dot org 2020-04-16 12:10 ` rguenth at gcc dot gnu.org 2020-04-16 12:17 ` rguenth at gcc dot gnu.org 2020-04-16 12:18 ` soap at gentoo dot org 2020-04-16 12:47 ` jakub at gcc dot gnu.org 2020-04-16 13:11 ` rguenth at gcc dot gnu.org 2021-07-19 3:40 ` [Bug target/94617] " pinskia at gcc dot gnu.org 2022-11-26 20:43 ` pinskia at gcc dot gnu.org 2023-06-25 22:29 ` pinskia at gcc dot gnu.org 2023-06-25 22:45 ` pinskia 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-94617-4@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).