public inbox for gcc-prs@sourceware.org help / color / mirror / Atom feed
From: iain.templeton@cisra.canon.com.au To: gcc-gnats@gcc.gnu.org Subject: optimization/5328: Optimisation causes code ordering to be incorrect on x86 Date: Tue, 08 Jan 2002 18:56:00 -0000 [thread overview] Message-ID: <20020109024845.0790C2015E@yves.research.canon.com.au> (raw) >Number: 5328 >Category: optimization >Synopsis: Optimisation causes code ordering to be incorrect on x86 >Confidential: no >Severity: serious >Priority: medium >Responsible: unassigned >State: open >Class: wrong-code >Submitter-Id: net >Arrival-Date: Tue Jan 08 18:56:00 PST 2002 >Closed-Date: >Last-Modified: >Originator: Iain Templeton >Release: 3.0.2 >Organization: Canon Information Systems Research Australia >Environment: System: Linux yves.research.canon.com.au 2.4.2-2 #1 Sun Apr 8 20:41:30 EDT 2001 i686 unknown Architecture: i686 PIII-850/Linux 2.4.2/Redhat 7.1 host: i686-pc-linux-gnu build: i686-pc-linux-gnu target: i686-pc-linux-gnu configured with: ./configure --prefix=/opt/gcc-3.0.2 >Description: A really simple bit of code shown below comes out with the comparison instruction and the mov instruction the wrong way around when using -O2. Using -O2 -fno-schedule-insns2 doesn't show this problem, and neither does -O1 -fschedule-insns -fschedule-insns2 This problem also effects 2.96 (but we don't talk about that). 2.95.3 (on FreeBSD though) doesn't do this, rather it does it using some other kind of optimisation and never outputs the assert test. Ie: gcc test.c -S -o test.S -O2 WRONG gcc test.c -S -o test.S -O2 -fno-schedule-insns2 OK gcc test.c -S -o test.S -O1 -fschedule-insns -fschedule-insns2 OK gcc -v gives: yves% gcc -v -save-temps -O2 -S simpletest.c -o simpletest.S Reading specs from /usr/lib/gcc-lib/i386-redhat-linux/2.96/specs gcc version 2.96 20000731 (Red Hat Linux 7.1 2.96-85) /usr/lib/gcc-lib/i386-redhat-linux/2.96/cpp0 -lang-c -v -iprefix /opt/bin/../lib/gcc-lib/i386-redhat-linux/2.96/ -D__GNUC__=2 -D__GNUC_MINOR__=96 -D__GNUC_PATCHLEVEL__=0 -D__ELF__ -Dunix -Dlinux -D__ELF__ -D__unix__ -D__linux__ -D__unix -D__linux -Asystem(posix) -D__OPTIMIZE__ -Acpu(i386) -Amachine(i386) -Di386 -D__i386 -D__i386__ -D__tune_i386__ simpletest.c simpletest.i GNU CPP version 2.96 20000731 (Red Hat Linux 7.1 2.96-85) (cpplib) (i386 Linux/ELF) ignoring nonexistent directory "/opt/lib/gcc-lib/i386-redhat-linux/2.96/include" ignoring nonexistent directory "/opt/i386-redhat-linux/include" ignoring nonexistent directory "/usr/local/include" ignoring nonexistent directory "/usr/i386-redhat-linux/include" #include "..." search starts here: #include <...> search starts here: /usr/lib/gcc-lib/i386-redhat-linux/2.96/include /usr/include End of search list. /usr/lib/gcc-lib/i386-redhat-linux/2.96/cc1 simpletest.i -quiet -dumpbase simpletest.c -O2 -version -o simpletest.S GNU C version 2.96 20000731 (Red Hat Linux 7.1 2.96-85) (i386-redhat-linux) compiled by GNU C version 2.96 20000731 (Red Hat Linux 7.1 2.96-85). The assembly below gives the output, the cmpl is before the movl. .file "simpletest.c" .version "01.01" gcc2_compiled.: .section .rodata .LC0: .string "main" .LC1: .string "simpletest.c" .align 32 .LC2: .string "(f1 = -12345.6789F, *(unsigned int *)&f1 == 0xC640E6B7U)" .text .align 4 .globl main .type main,@function main: pushl %ebp movl %esp, %ebp subl $8, %esp cmpl $-968825161, -4(%ebp) movl $0xc640e6b7, -4(%ebp) je .L4 pushl $.LC0 pushl $7 pushl $.LC1 pushl $.LC2 call __assert_fail .p2align 2 .L4: xorl %eax, %eax leave ret .Lfe1: .size main,.Lfe1-main .ident "GCC: (GNU) 2.96 20000731 (Red Hat Linux 7.1 2.96-85)" >How-To-Repeat: Compile the C code and look at the assembler output. -- Code Start extern void __assert_fail (__const char *__assertion, __const char *__file, unsigned int __line, __const char *__function) __attribute__ ((__noreturn__)); extern void __assert_perror_fail (int __errnum, __const char *__file, unsigned int __line, __const char *__function) __attribute__ ((__noreturn__)); extern void __assert (const char *__assertion, const char *__file, int __line) __attribute__ ((__noreturn__)); int main(int argc, char **argv) { float f1; /* assert((f1 = -12345.6789F, *(unsigned int *)&f1 == 0xC640E6B7U)); */ ((void) (((f1 = -12345.6789F, *(unsigned int *)&f1 == 0xC640E6B7U)) ? 0 : (__assert_fail ("(f1 = -12345.6789F, *(unsigned int *)&f1 == 0xC640E6B7U)", "simpletest.c", 7, __PRETTY_FUNCTION__), 0))); return 0; } -- Code End >Fix: >Release-Note: >Audit-Trail: >Unformatted:
next reply other threads:[~2002-01-09 2:56 UTC|newest] Thread overview: 2+ messages / expand[flat|nested] mbox.gz Atom feed top 2002-01-08 18:56 iain.templeton [this message] 2002-02-06 8:05 jakub
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=20020109024845.0790C2015E@yves.research.canon.com.au \ --to=iain.templeton@cisra.canon.com.au \ --cc=gcc-gnats@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).