Fix ChauvenetOutlierDetection on docker / ubuntu

[gfoidl]

See https://github.com/gfoidl/Stochastics/pull/36#issuecomment-360477559

This test failes on CircleCI, on all other tests platforms it is OK:

https://github.com/gfoidl/Stochastics/blob/9c22392e49e50647339482c4add12804cfb30058/tests/gfoidl.Stochastics.Tests/Statistics/OutlierDetection/ChauvenetOutlierDetectionTests/GetOutliers.cs#L10-L22

[gfoidl]

ubuntu-x64

uname -a

Linux ubuntu 4.4.0-62-generic #83-Ubuntu SMP Wed Jan 18 14:10:15 UTC 2017 x86_64 x86_64 x86_64 GNU/Linux

CPU-Info / lscpu

Architecture:          x86_64
CPU op-mode(s):        32-bit, 64-bit
Byte Order:            Little Endian
CPU(s):                4
On-line CPU(s) list:   0-3
Thread(s) per core:    1
Core(s) per socket:    4
Socket(s):             1
NUMA node(s):          1
Vendor ID:             GenuineIntel
CPU family:            6
Model:                 158
Model name:            Intel(R) Core(TM) i7-7700HQ CPU @ 2.80GHz
Stepping:              9
CPU MHz:               2808.002
BogoMIPS:              5616.00
Hypervisor vendor:     KVM
Virtualization type:   full
L1d cache:             32K
L1i cache:             32K
L2 cache:              256K
L3 cache:              6144K
NUMA node0 CPU(s):     0-3
Flags:                 fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx rdtscp lm constant_tsc rep_good nopl xtopology nonstop_tsc pni pclmulqdq ssse3 cx16 sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx rdrand hypervisor lahf_lm abm 3dnowprefetch avx2 rdseed clflushopt

probOutside

 0      0.815724735289837
 3      0.826191062791331
 4      0.829686811399171
 1      0.819209958020078
 2      0.822698756058058
 3      0.826191062791331
 0      0.815724735289837
 2      0.822698756058058
 1      0.819209958020078
 3      0.826191062791331
 2      0.822698756058058
 0      0.815724735289837
 2      0.822698756058058
 2      0.822698756058058
 3      0.826191062791331
 2      0.822698756058058
 5      0.833185934856398
 2      0.822698756058058
 3      0.826191062791331
 999    2.15247343841125E-05

CircleCI

uname -a

Linux 143fb24f23d3 4.4.0-111-generic #134~14.04.1-Ubuntu SMP Mon Jan 15 15:39:56 UTC 2018 x86_64 GNU/Linux

CPU-Info / lscpu

Architecture:          x86_64
CPU op-mode(s):        32-bit, 64-bit
Byte Order:            Little Endian
CPU(s):                32
On-line CPU(s) list:   0-31
Thread(s) per core:    2
Core(s) per socket:    8
Socket(s):             2
NUMA node(s):          2
Vendor ID:             GenuineIntel
CPU family:            6
Model:                 62
Model name:            Intel(R) Xeon(R) CPU E5-2680 v2 @ 2.80GHz
Stepping:              4
CPU MHz:               2800.076
BogoMIPS:              5600.15
Hypervisor vendor:     Xen
Virtualization type:   full
L1d cache:             32K
L1i cache:             32K
L2 cache:              256K
L3 cache:              25600K
NUMA node0 CPU(s):     0-7,16-23
NUMA node1 CPU(s):     8-15,24-31
Flags:                 fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx rdtscp lm constant_tsc rep_good nopl xtopology eagerfpu pni pclmulqdq ssse3 cx16 pcid sse4_1 sse4_2 x2apic popcnt tsc_deadline_timer aes xsave avx f16c rdrand hypervisor lahf_lm kaiser fsgsbase smep erms xsaveopt

probOutside

 0      0.815724735289837
 3      0.826191062791331
 4      0.829686811399171
 1      0.819209958020078
 2      0.822698756058058
 3      0.826191062791331
 0      0.815724735289837
 2      0.822698756058058
 1      0.819209958020078
 3      0.826191062791331
 2      0.822698756058058
 0      0.815724735289837
 2      0.822698756058058
 2      0.822698756058058
 3      0.826191062791331
 2      0.822698756058058
 5      0.833185934856398
 2      0.822698756058058
 3      1
 999    1

[gfoidl]

The problem goes down to

Standard Output Messages:
    at System.Environment.get_StackTrace()
    at gfoidl.Stochastics.SpecialFunctions.Erfc(Double[] values) in /root/repo/source/gfoidl.Stochastics/SpecialFunctions.cs

https://github.com/gfoidl/Stochastics/blob/9c22392e49e50647339482c4add12804cfb30058/source/gfoidl.Stochastics/Statistics/OutlierDetection/ChauvenetOutlierDetection.cs#L85-L92

Where the values for tsusSqrt2Inv are at CircleCI:

 0.164789272799277
 0.155273049153506
 0.152100974604915
 0.161617198250687
 0.158445123702096
 0.155273049153506
 0.164789272799277
 0.158445123702096
 0.161617198250687
 0.155273049153506
 0.158445123702096
 0.164789272799277
 0.158445123702096
 0.158445123702096
 0.155273049153506
 0.158445123702096
 0.148928900056325
 0.158445123702096
 0
 0

So the last two 0 result in (correct) Erfc -> 1. This means the problem is in https://github.com/gfoidl/Stochastics/blob/9c22392e49e50647339482c4add12804cfb30058/source/gfoidl.Stochastics/Statistics/OutlierDetection/ChauvenetOutlierDetection.cs#L101

[gfoidl]

When https://github.com/gfoidl/Stochastics/blob/9c22392e49e50647339482c4add12804cfb30058/source/gfoidl.Stochastics/Statistics/OutlierDetection/ChauvenetOutlierDetection.cs#L114 is set to false then the test passes.

[gfoidl]

OK, found the problem.

https://github.com/gfoidl/Stochastics/blob/9c22392e49e50647339482c4add12804cfb30058/source/gfoidl.Stochastics/Statistics/OutlierDetection/ChauvenetOutlierDetection.cs#L163-L164

On the other test-platforms Vector<double>.Count was 4, and the size of the test-array is 20. So with 8x + 1x the whole array is processed. The normal for-loop isn't touched -> problem does not show up.

On CircleCI Vector<double>.Count is 2, so the normal for-loop gets executed and it indexes on the pointers that get incremented in the simd-loops. So the access is far outside the array-bounds. unsafe caught me 😜 (i.e. with safe code there would have been an IndexOutOfRangeException)

So two possibilities:

for (; i < n; ++i)
    *tsus++ = Abs(*zTrans++) * Sqrt2Inv;

or

for (; i < n; ++i)
    pTarget[i] = Abs(pSource[i]) * Sqrt2Inv;

Additionally there should be a testcase that forces to use the normal-loop, i.e. an array size that is not a multiple of simd-size.

[gfoidl]

Closing as https://github.com/gfoidl/Stochastics/pull/38 is merged.