[Bug c/35709] New: severe perfromance degradation with "float complex" type

[Bug c/35709] New: severe perfromance degradation with "float complex" type

[sergstesh at yahoo dot com]

Compiling and running the same test case with gcc-4.2.3 and gcc-4.3.0 shows
that in the latter case performance takes an almost 3x hit:

"
sergei@amdam2:/mnt/sda1/sergei/learning_octave>
/maxtor5/sergei/AppsFromScratchWD/install/gcc-4.2.3/binsh/gcc -Wall
-mtune=native -march=native -O2 -msse -lm complex_multiplication_testcase.c -o
complex_multiplication_testcase
sergei@amdam2:/mnt/sda1/sergei/learning_octave> time
./complex_multiplication_testcase
executions of straightforward_multiply_with_ptrs() took 6.77 seconds at line
number 136 of 'complex_multiplication_testcase.c' file
executions of straightforward_multiply() took 6.74 seconds at line number 154
of 'complex_multiplication_testcase.c' file
executions of straightforward_multiply_2_signals() took 8.62 seconds at line
number 172 of 'complex_multiplication_testcase.c' file

real    0m22.326s
user    0m22.121s
sys     0m0.012s
sergei@amdam2:/mnt/sda1/sergei/learning_octave>
/maxtor5/sergei/AppsFromScratchWD/install/gcc-4.3.0/binsh/gcc -Wall
-mtune=native -march=native -O2 -msse -lm complex_multiplication_testcase.c -o
complex_multiplication_testcase
sergei@amdam2:/mnt/sda1/sergei/learning_octave> time
./complex_multiplication_testcase
executions of straightforward_multiply_with_ptrs() took 18.17 seconds at line
number 136 of 'complex_multiplication_testcase.c' file
executions of straightforward_multiply() took 16.91 seconds at line number 154
of 'complex_multiplication_testcase.c' file
executions of straightforward_multiply_2_signals() took 30.64 seconds at line
number 172 of 'complex_multiplication_testcase.c' file

real    1m6.306s
user    1m5.676s
sys     0m0.056s
sergei@amdam2:/mnt/sda1/sergei/learning_octave>
"

- see, for example,

executions of straightforward_multiply() took 6.74 seconds
vs
executions of straightforward_multiply() took 16.91 seconds

and

user    0m22.121s
vs
user    1m5.676s
.

FWIW, gcc-3.4.6 shows comparable to gcc-4.2.3 results, albeit slightly worse,
so it looks like the issue is very much gcc-4.3.0 specific.

I'll upload the test case.


-- 
           Summary: severe perfromance degradation with "float complex" type
           Product: gcc
           Version: 4.3.0
            Status: UNCONFIRMED
          Severity: major
          Priority: P3
         Component: c
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: sergstesh at yahoo dot com
 GCC build triplet: i686-pc-linux-gnu
  GCC host triplet: i686-pc-linux-gnu
GCC target triplet: i686-pc-linux-gnu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35709

[Bug c/35709] severe perfromance degradation with "float complex" type

[sergstesh at yahoo dot com]

------- Comment #1 from sergstesh at yahoo dot com  2008-03-26 19:23 -------
Created an attachment (id=15383)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=15383&action=view)
the test case


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35709

[Bug c/35709] severe perfromance degradation with "float complex" type

[sergstesh at yahoo dot com]

------- Comment #2 from sergstesh at yahoo dot com  2008-03-26 19:40 -------
My system:

Linux amdam2 2.6.18.8-0.9-default #1 SMP Sun Feb 10 22:48:05 UTC 2008 i686
athlon i386 GNU/Linux

- SUSE 10.2, the CPU is: "AMD Athlon(tm) 64 X2 Dual Core Processor 4600+
stepping 02".


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35709

[Bug middle-end/35709] severe perfromance degradation with "float complex" type

[rguenth at gcc dot gnu dot org]

------- Comment #3 from rguenth at gcc dot gnu dot org  2008-03-26 20:33 -------
This was an intentional change for correctness.  4.3 uses the library
implementation to handle the case of NaNs correctly.  Use -fcx-limited-range
if you want back the performance.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|                            |INVALID


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35709

[Bug c/35709] severe perfromance degradation with "float complex" type

[sergstesh at yahoo dot com]

------- Comment #4 from sergstesh at yahoo dot com  2008-03-26 20:44 -------
(In reply to comment #3)
> This was an intentional change for correctness.  4.3 uses the library
> implementation to handle the case of NaNs correctly.  Use -fcx-limited-range
> if you want back the performance.
> 

Well, I'm using "float complex" for real time audio, so correctness of NaNs is
not an issue.

I think the default should have been the opposite. Why to break existing
applications ?

When "float" rather than "double" is used, it's most likely for non-sicentific
needs, so correctness in canse of NaNs most likely isn't an issue.


-- 

sergstesh at yahoo dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|middle-end                  |c


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35709

[Bug c/35709] severe perfromance degradation with "float complex" type

[pinskia at gcc dot gnu dot org]

------- Comment #5 from pinskia at gcc dot gnu dot org  2008-03-26 20:45 -------
As mentioned, use -fcx-limited-range.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35709

[Bug c/35709] severe perfromance degradation with "float complex" type

[sergstesh at yahoo dot com]

------- Comment #6 from sergstesh at yahoo dot com  2008-03-26 20:58 -------
(In reply to comment #5)
> As mentioned, use -fcx-limited-range.
> 

Can't you add just _one_ command line switch: -old_behavior ?
Or -gcc_4_2_3_behavior ?

For example, I need FFTW3 with "float" rather than "double", and I need its
speed. I am not sure FFTW3 'configure' will add -fcx-limited-range.

Thanks in advance.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35709

[Bug c/35709] severe perfromance degradation with "float complex" type

[rguenth at gcc dot gnu dot org]

------- Comment #7 from rguenth at gcc dot gnu dot org  2008-03-26 22:22 -------
I would suggest for real time audio you should use -ffast-math which will
enable
-fcx-limited-range as well.  It also disables generally handling of NaNs, Infs
and other optimization limiting things.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35709

[Bug c/35709] severe perfromance degradation with "float complex" type

[pinskia at gcc dot gnu dot org]

-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|major                       |normal


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35709