public inbox for gcc-bugs@sourceware.org
help / color / mirror / Atom feed
From: "ubizjak at gmail dot com" <gcc-bugzilla@gcc.gnu.org>
To: gcc-bugs@gcc.gnu.org
Subject: [Bug regression/39914] 96% performance regression in floating point code; part of the problem started 2009/03/12-13
Date: Mon, 27 Apr 2009 08:16:00 -0000 [thread overview]
Message-ID: <20090427081621.19946.qmail@sourceware.org> (raw)
In-Reply-To: <bug-39914-271@http.gcc.gnu.org/bugzilla/>
------- Comment #2 from ubizjak at gmail dot com 2009-04-27 08:16 -------
(In reply to comment #0)
> (same .i file, same instructions for reproducing, same compiler options, same
> everything)
I guess that this is direct.i compiled with -O1?
Trunk, revision: 146825 -O1 on x86_64 linux gives:
.L27:
leaq 4(%rbx), %rbp
movq %rbx, %rdx
addq (%r11), %rdx
movq %rdx, (%rsi)
addq $4, %rdx
movq %rdx, (%r10)
movq (%r11), %rdx
addq (%rsi), %rdx
movq %rdx, (%rcx)
addq $4, %rdx
movq %rdx, (%r9)
movq (%r11), %r12
addq (%rcx), %r12
movq %r12, (%r8)
addq $4, %r12
movq %r12, (%r15)
movq (%rax), %rdx
addq $7, %rdx
movsd (%rdx,%r12,2), %xmm2
movsd %xmm2, -96(%rsp)
movq (%r8), %r12
movsd (%rdx,%r12,2), %xmm2
movsd %xmm2, -64(%rsp)
movq (%r9), %r12
movsd (%rdx,%r12,2), %xmm2
movsd %xmm2, -56(%rsp)
movq (%rcx), %r12
movsd (%rdx,%r12,2), %xmm2
movsd %xmm2, -48(%rsp)
movq (%r10), %r12
movsd (%rdx,%r12,2), %xmm2
movsd %xmm2, -104(%rsp)
movq (%rsi), %r12
movsd (%rdx,%r12,2), %xmm2
movsd %xmm2, -88(%rsp)
leaq (%rbp,%rbp), %r12
movsd (%r12,%rdx), %xmm2
movsd %xmm2, -80(%rsp)
leaq (%rdx,%rbx,2), %r14
movq 24(%rdi), %rdx
movsd 31(%rdx), %xmm2
movsd %xmm2, -32(%rsp)
movsd 23(%rdx), %xmm2
movsd %xmm2, -40(%rsp)
movsd 15(%rdx), %xmm2
movsd %xmm2, -120(%rsp)
movsd 7(%rdx), %xmm2
movsd %xmm2, -112(%rsp)
movapd %xmm2, %xmm3
mulsd -96(%rsp), %xmm3
movsd -120(%rsp), %xmm2
mulsd -64(%rsp), %xmm2
addsd %xmm2, %xmm3
movsd %xmm3, -24(%rsp)
movsd -112(%rsp), %xmm3
mulsd -64(%rsp), %xmm3
movsd -120(%rsp), %xmm2
mulsd -96(%rsp), %xmm2
subsd %xmm2, %xmm3
movsd %xmm3, -96(%rsp)
movsd -112(%rsp), %xmm3
mulsd -56(%rsp), %xmm3
movsd -120(%rsp), %xmm2
mulsd -48(%rsp), %xmm2
addsd %xmm2, %xmm3
movsd %xmm3, -64(%rsp)
movsd -112(%rsp), %xmm3
mulsd -48(%rsp), %xmm3
movsd -120(%rsp), %xmm2
mulsd -56(%rsp), %xmm2
subsd %xmm2, %xmm3
movsd %xmm3, -120(%rsp)
movsd -104(%rsp), %xmm2
subsd -24(%rsp), %xmm2
movsd %xmm2, -112(%rsp)
movsd -88(%rsp), %xmm2
subsd -96(%rsp), %xmm2
movsd %xmm2, -56(%rsp)
movsd -80(%rsp), %xmm2
subsd -64(%rsp), %xmm2
movsd %xmm2, -48(%rsp)
movsd (%r14), %xmm2
subsd %xmm3, %xmm2
movsd %xmm2, -16(%rsp)
movsd -104(%rsp), %xmm2
addsd -24(%rsp), %xmm2
movsd %xmm2, -104(%rsp)
movsd -88(%rsp), %xmm2
addsd -96(%rsp), %xmm2
movsd %xmm2, -88(%rsp)
movsd -80(%rsp), %xmm2
addsd -64(%rsp), %xmm2
movsd %xmm2, -80(%rsp)
movsd (%r14), %xmm2
addsd %xmm3, %xmm2
movsd %xmm2, -72(%rsp)
movsd -32(%rsp), %xmm2
xorpd %xmm0, %xmm2
movsd %xmm2, -120(%rsp)
movapd %xmm2, %xmm3
mulsd -112(%rsp), %xmm3
movsd -40(%rsp), %xmm2
mulsd -56(%rsp), %xmm2
addsd %xmm2, %xmm3
movsd %xmm3, -96(%rsp)
movsd -120(%rsp), %xmm3
mulsd -56(%rsp), %xmm3
movsd -40(%rsp), %xmm2
mulsd -112(%rsp), %xmm2
subsd %xmm2, %xmm3
movsd %xmm3, -120(%rsp)
movsd -40(%rsp), %xmm3
mulsd -104(%rsp), %xmm3
movsd -32(%rsp), %xmm2
mulsd -88(%rsp), %xmm2
addsd %xmm2, %xmm3
movsd %xmm3, -112(%rsp)
movsd -40(%rsp), %xmm3
mulsd -88(%rsp), %xmm3
movsd -32(%rsp), %xmm2
mulsd -104(%rsp), %xmm2
subsd %xmm2, %xmm3
movsd %xmm3, -104(%rsp)
movsd -72(%rsp), %xmm2
addsd %xmm3, %xmm2
movsd %xmm2, (%r14)
movq (%rax), %rdx
movsd -80(%rsp), %xmm2
addsd -112(%rsp), %xmm2
movsd %xmm2, 7(%r12,%rdx)
movq (%rsi), %r12
movq (%rax), %rdx
movsd -72(%rsp), %xmm2
subsd -104(%rsp), %xmm2
movsd %xmm2, 7(%rdx,%r12,2)
movq (%r10), %r12
movq (%rax), %rdx
movsd -80(%rsp), %xmm2
subsd -112(%rsp), %xmm2
movsd %xmm2, 7(%rdx,%r12,2)
movq (%rcx), %r12
movq (%rax), %rdx
movsd -16(%rsp), %xmm2
addsd -120(%rsp), %xmm2
movsd %xmm2, 7(%rdx,%r12,2)
movq (%r9), %r12
movq (%rax), %rdx
movsd -48(%rsp), %xmm2
addsd -96(%rsp), %xmm2
movsd %xmm2, 7(%rdx,%r12,2)
movq (%r8), %r12
movq (%rax), %rdx
movsd -16(%rsp), %xmm2
subsd -120(%rsp), %xmm2
movsd %xmm2, 7(%rdx,%r12,2)
movq (%r15), %r12
movq (%rax), %rdx
movsd -48(%rsp), %xmm2
subsd -96(%rsp), %xmm2
movsd %xmm2, 7(%rdx,%r12,2)
addq $8, %rbx
cmpq %rbx, -8(%rsp)
jg .L27
The code above looks similar to your gcc version 4.4.0 20090313 code.
Using -O2, I get:
.L27:
movq -96(%rsp), %r14
leaq (%rax,%rcx,2), %rdi
leaq -8(%rax,%rcx,2), %rbp
leaq (%rax,%rsi,2), %r8
leaq -8(%rax,%rsi,2), %r9
leaq 8(%rax,%rdx,2), %r12
movsd (%rdi), %xmm2
leaq 8(%rax,%rbx,2), %r10
movsd (%r14), %xmm4
movq -88(%rsp), %r14
movsd (%rbp), %xmm6
leaq (%rax,%rbx,2), %r11
movsd (%r8), %xmm9
leaq (%rax,%rdx,2), %r13
movsd (%r14), %xmm1
movq -120(%rsp), %r14
movsd (%r9), %xmm10
movq %rcx, -80(%rsp)
movapd %xmm1, %xmm14
addq $8, %rdx
movsd (%r14), %xmm5
addq $8, %rcx
mulsd %xmm6, %xmm14
addq $8, %rsi
addq $8, %rbx
movapd %xmm5, %xmm7
mulsd %xmm5, %xmm6
movsd (%r12), %xmm11
cmpq %rdx, -112(%rsp)
mulsd %xmm2, %xmm7
mulsd %xmm1, %xmm2
movsd (%r15), %xmm8
movsd (%r11), %xmm3
addsd %xmm14, %xmm7
movapd %xmm1, %xmm14
subsd %xmm2, %xmm6
movapd %xmm5, %xmm2
mulsd %xmm10, %xmm14
mulsd %xmm9, %xmm2
mulsd %xmm9, %xmm1
movapd %xmm11, %xmm9
mulsd %xmm10, %xmm5
movsd (%r10), %xmm15
addsd %xmm14, %xmm2
movsd (%r13), %xmm0
movapd %xmm15, %xmm14
subsd %xmm1, %xmm5
movapd %xmm3, %xmm1
subsd %xmm7, %xmm14
movapd %xmm0, %xmm10
subsd %xmm2, %xmm9
addsd %xmm2, %xmm11
movapd %xmm8, %xmm2
subsd %xmm6, %xmm1
xorpd %xmm12, %xmm2
subsd %xmm5, %xmm10
addsd %xmm3, %xmm6
movapd %xmm4, %xmm3
addsd %xmm0, %xmm5
movapd %xmm2, %xmm0
mulsd %xmm1, %xmm3
addsd %xmm15, %xmm7
mulsd %xmm2, %xmm1
mulsd %xmm14, %xmm0
movapd %xmm4, %xmm2
mulsd %xmm4, %xmm14
mulsd %xmm7, %xmm2
addsd %xmm3, %xmm0
movapd %xmm8, %xmm3
mulsd %xmm8, %xmm7
subsd %xmm14, %xmm1
mulsd %xmm6, %xmm3
addsd %xmm3, %xmm2
movapd %xmm4, %xmm3
movapd %xmm5, %xmm4
mulsd %xmm6, %xmm3
subsd %xmm7, %xmm3
addsd %xmm3, %xmm4
subsd %xmm3, %xmm5
movsd %xmm4, (%r13)
movapd %xmm11, %xmm4
subsd %xmm2, %xmm11
addsd %xmm2, %xmm4
movapd %xmm10, %xmm2
subsd %xmm1, %xmm10
addsd %xmm1, %xmm2
movsd %xmm4, (%r12)
movsd %xmm5, (%r11)
movsd %xmm11, (%r10)
movsd %xmm2, (%r9)
movapd %xmm9, %xmm2
subsd %xmm0, %xmm9
addsd %xmm0, %xmm2
movsd %xmm2, (%r8)
movsd %xmm10, (%rbp)
movsd %xmm9, (%rdi)
jg .L27
It is not clear from your report, if -O1 flag is problematic, -O2 code looks
good to me.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39914
next prev parent reply other threads:[~2009-04-27 8:16 UTC|newest]
Thread overview: 17+ messages / expand[flat|nested] mbox.gz Atom feed top
2009-04-26 18:24 [Bug regression/39914] New: " lucier at math dot purdue dot edu
2009-04-26 18:43 ` [Bug regression/39914] " ubizjak at gmail dot com
2009-04-27 8:16 ` ubizjak at gmail dot com [this message]
2009-04-27 15:07 ` lucier at math dot purdue dot edu
2009-04-27 15:11 ` lucier at math dot purdue dot edu
2009-04-27 15:26 ` pinskia at gcc dot gnu dot org
2009-04-27 15:32 ` lucier at math dot purdue dot edu
2009-04-27 15:35 ` lucier at math dot purdue dot edu
2009-04-27 16:29 ` lucier at math dot purdue dot edu
2009-04-27 18:21 ` ubizjak at gmail dot com
2009-04-27 19:04 ` [Bug regression/39914] [4.4/4.5 Regression] " bonzini at gnu dot org
2009-04-27 20:38 ` lucier at math dot purdue dot edu
2009-04-28 1:40 ` lucier at math dot purdue dot edu
2009-04-28 16:18 ` uros at gcc dot gnu dot org
2009-04-28 16:19 ` [Bug rtl-optimization/39914] [4.4 " ubizjak at gmail dot com
2009-05-03 19:40 ` uros at gcc dot gnu dot org
2009-05-03 19:41 ` ubizjak at gmail dot com
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=20090427081621.19946.qmail@sourceware.org \
--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: link
Be 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).