[aarch64-port-dev ] Some benchmarking help required

[Dmitry Chuyko]

Some numbers for fixed patch #2 
http://cr.openjdk.java.net/~aph/aarch64_8203006-3.patch :

==T88
Benchmark            #2 %    p
compress             -1.32   0.005
crypto.aes           -3.38   0.000
crypto.signverify    0.80    0.043
derby                -3.51   0.000
mpegaudio            0.95    0.007
scimark.sor.large    5.33    0.000
scimark.lu.small     -6.42   0.000
xml.transform        -3.52   0.000
xml.validation       -1.53   0.045

==T99
Benchmark              #2 %    p
compress               2.86    0.041
crypto.rsa             -0.86   0.032
scimark.fft.small      -0.64   0.015
scimark.lu.small       10.82   0.011
scimark.sparse.small   -5.32   0.000
scimark.monte_carlo    1.39    0.030
sunflow                1.04    0.007

No failed runs but also no univocal improvement (in contrast with #1) 
while the difference is present.

I also made runs with merged patch, the numbers:

==T88
Benchmark              #1#2 %  p
compress               -1.35   0.000
crypto.aes              29.42   0.000
crypto.signverify       1.13    0.047
derby                   -1.48   0.042
mpegaudio               3.14    0.000
scimark.sor.large       5.35    0.000
scimark.sparse.large    -4.01   0.000
scimark.lu.small        -23.97  0.000
scimark.sparse.small    -4.64   0.000
serial                  2.56    0.034
sunflow                 1.81    0.000
xml.transform           11.49   0.000

==T99
Benchmark               #1#2 %  p
crypto.aes              26.08   0.000
crypto.rsa              1.87    0.000
crypto.signverify       0.71    0.000
mpegaudio               16.61   0.000
scimark.sor.large       0.57    0.031
scimark.sparse.large    19.02   0.023
scimark.fft.small       -0.86   0.002
scimark.sor.small       8.51    0.000
scimark.sparse.small    -1.54   0.019
scimark.monte_carlo     7.68    0.000
sunflow                 4.59    0.000
xml.transform           4.38    0.000
xml.validation          1.33    0.003

On T99 lu.small and derby shown insignificant -3-4%. And sparse.large 
shown insignificant +10-30% with #1 and #2.

-Dmitry


On 05/16/2018 05:44 PM, Andrew Haley wrote:
> ..................................