It's worth playing with optimization level more than some of those. It depends on engine if Os, O2, or O3 produces the fastest build on these devices. For zerowin, a Cortex-a7 running Stockfish, O2 is fastest.lucasart wrote: ↑Tue Nov 27, 2018 2:22 amAll of this seems completely gratuitous:MikeB wrote: ↑Tue Nov 27, 2018 1:19 amCode: Select all
$(CC) $(CF) -DVERSION=\"$(VERSION)\" ./src/*.c -o $(OUT) -march=armv8-a+crc -mfpu=neon-vfpv4 -mtune=cortex-a53 -ftree-vectorize -mfloat-abi=hard -O3 $(LF)
-mfpu=neon-vfpv4 -mtune=cortex-a53 -ftree-vectorize -mfloat-abi=hard
For weaker devices, there are several things worth testing in additon to optimization level:
linktime optimizations on/off
pgo on/off
mthumb on/off
I can't tell you how many devices experienced a reduction with -flto or with PGO. Fruit definitely was faster with -mthumb on some devices. You have to experiment. In the few places it mattered, -mthumb was the bomb. over 50% difference.