[LLVMdev] Polly issue

Tobias Grosser tobias at grosser.es
Mon May 20 07:06:11 PDT 2013


On 05/20/2013 03:02 AM, maxs wrote:
> Hi,
>      When I test "matmul" in the polly directory, I get the following
> performance data:
> //===============================
> --> 12. Compare the runtime of the executables
> time ./matmul.normalopt.exe
> 0:23.53 real, 23.48 user, 0.00 sys
> time ./matmul.polly.interchanged.exe
> 0:22.86 real, 22.82 user, 0.01 sys
> time ./matmul.polly.interchanged+tiled.exe
> 0:22.87 real, 22.83 user, 0.00 sys
> time ./matmul.polly.interchanged+tiled+vector.exe
> 0:22.53 real, 22.48 user, 0.02 sys
> time ./matmul.polly.interchanged+tiled+vector+openmp.exe
> 0:05.88 real, 46.71 user, 0.02 sys
> //===============================
>      The data is different from the official site
> "http://polly.llvm.org/example_manual_matmul.html" results. Any advices?

Hi maxs,

thanks for trying Polly.

Did you run the above tests using the ./runall.sh script? Did you verify 
its output corresponds to the output shown in the previous steps? When I 
run it I realized the naming of the scops tested, such that Polly failed 
to read in the transformations. I fixed this on trunk (commit 182265) 
and now get (on my new machine):

time ./matmul.normalopt.exe
0:26.21 real, 26.10 user, 0.03 sys
time ./matmul.polly.interchanged.exe
0:02.39 real, 2.38 user, 0.00 sys
time ./matmul.polly.interchanged+tiled.exe
0:03.16 real, 3.14 user, 0.00 sys
time ./matmul.polly.interchanged+tiled+vector.exe
0:00.93 real, 0.92 user, 0.00 sys
time ./matmul.polly.interchanged+tiled+vector+openmp.exe
0:00.42 real, 1.64 user, 0.00 sys

Tobias




More information about the llvm-dev mailing list