Fix a data race for NNUE build - App crash for the bmi version

[Coolchessguykevin]

Latest build from abrok.eu (BMI2 version), terminal engine window works well, but when trying to start analisys in any GUI (checked at Fritz and Aquarium), engine is crashing. NNUE=true, default net used.

Yesterday's build worked well at the same conditions.

Some info from windows event log:

P1: stockfish_20080923_x64_bmi2.exe
P2: 0.0.0.0
P3: 00000000
P4: stockfish_20080923_x64_bmi2.exe
P5: 0.0.0.0
P6: 00000000
P7: c0000005
P8: 00000000000eaaca
P9:
P10:

[Coolchessguykevin]

It happens ONLY when NNUE=true. With classic evaluation engine works without crashes.

[Coolchessguykevin]

From the engine log file before its crashes:

setoption name Contempt value 0
setoption name Threads value 7
<< info string Hash table allocation: Windows large pages not used.
setoption name Use NNUE value true
setoption name EvalFile value C:\NNUE\nn-9931db908a9b.nnue
quit

[crossbr]

I have the same problem, both with bmi and with avx2, in Arena GUI. It works if I uncheck the "Use NNUE" box, but not if I check that box. I have the same problem with the "Avoid special casing for MinGW" commit (also named 20080923). But Viz's 20080921 works fine for me.

[rooklift]

I have what seems to be this crash - I agree the relevant commit is the previous one, a6e8929 ("Avoid special casing for MinGW")

[vondele]

OK, compiler issue identified with gcc 7.3, I'll revert the offending commit.