llvm dev - Sep 2011 - [LLVMdev] sincos functions

Hello,

  I was trying to compare the performance of icc, gcc and llvm on the
program almabench.c in Coyote Benchmark suite. Here is a line of code
from the program.


 da   = da + (ca[np][k] * cos(arga) + sa[np][k] * sin(arga)) * 0.0000001;

gcc and icc are performing way better than llvm as they are using
'sincos' library function to compute the sin and the cos of the
argument in a single shot. However llvm is computing sin and cos
seperately.

Is there any compiler option which can enable the code generator to
use 'sincos' function against 'sin' and 'cos' functions
seperately.

-Suresh

Hi Suresh,
>    I was trying to compare the performance of icc, gcc and llvm on the
> program almabench.c in Coyote Benchmark suite. Here is a line of code
> from the program.
>
>
>   da   = da + (ca[np][k] * cos(arga) + sa[np][k] * sin(arga)) * 0.0000001;
>
> gcc and icc are performing way better than llvm as they are using
> 'sincos' library function to compute the sin and the cos of the
> argument in a single shot. However llvm is computing sin and cos
> seperately.
>
> Is there any compiler option which can enable the code generator to
> use 'sincos' function against 'sin' and 'cos'
functions seperately.
no, it hasn't been implemented in LLVM yet.  Please open a bugreport if this
optimization is important for you.

Ciao, Duncan.

Duncan,

I am not sure how important the optimization is. I found this
optimization, when I am trying to improve the performance of the
almabench.c program in Coyote. On my machine:

1) With icc it takes 4 seconds.
2) With gcc it takes 8 seconds
3) With llvm it takes 12 seconds.

gcc and icc are using 'sincos' functions with icc using a vectorized
implementation of 'sincos' function.

However, I will open a bugreport on this.

-Suresh

On Thu, Sep 15, 2011 at 5:59 PM, Duncan Sands <baldrick at free.fr>
wrote:
> Hi Suresh,
>
>>    I was trying to compare the performance of icc, gcc and llvm on the
>> program almabench.c in Coyote Benchmark suite. Here is a line of code
>> from the program.
>>
>>
>>   da   = da + (ca[np][k] * cos(arga) + sa[np][k] * sin(arga)) *
0.0000001;
>>
>> gcc and icc are performing way better than llvm as they are using
>> 'sincos' library function to compute the sin and the cos of the
>> argument in a single shot. However llvm is computing sin and cos
>> seperately.
>>
>> Is there any compiler option which can enable the code generator to
>> use 'sincos' function against 'sin' and 'cos'
functions seperately.
>
> no, it hasn't been implemented in LLVM yet.  Please open a bugreport if
this
> optimization is important for you.
>
> Ciao, Duncan.
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>