Thanks for that!
I got it to compile with regular repo code - however, about 15% SLOWER than wallet miner and your precompiled binary.
Changing algo options (-mno-sha etc.) didn't change anything, always ~46kH/s (precompiled & wallet do 53-55) on a Ryzen 1700.
What compile options did you use?
Check winbuild-cross.sh file.
Which version of mingw/gcc did you use?
Thank you!
gcc-5.4
Its a Ryzen thing! Quite a story: the issue seems to be that GCC (esp. 5.4) is not optimized fro Ryzen at all. Compiling with your -march=corei7-avx gives me exactly the same speed as your compiled binary. Doing a -march=native results in the 15% decrease in speed. The often recommended -march=bdver4 doesn't help at all, same speed decrease. I installed GCC 6.3 to compile with -march=znver1 on Ubuntu, however that leads to segfaults.
Went to my Gentoo box to compile with GCC 6.4, finally setting CFLAGS="-O3 -march=native -Wall" CXXFLAGS="$CFLAGS -std=gnu++11 -fpermissive -march=znver1 -DFOUR_WAY" gives the same speed as with -march=corei7-avx....but no increase! So -march=corei7-avx seems to be the best thing for Ryzen CPUs.