Vector Class Discussion

float uf[8][8];
Vec8f uL;
uL = gather8f<0,8,16,24,32,40,48,56>(&(uf[0][0]));

The goal of this code is to take the elements uf[*][0] and put them in a single Vec8f

It segfaulted in normal use (g++ 5.4.0 under cygwin on Windows 10), flags g++ -O3 -mavx2 -mfma
When I ran under gdb it didn't segfault and did what I expected.

I have equivalent code that does it with 8 inserts for less than AVX which seems inefficient (Which deals with the lack of gather8f for < AVX).

I'd appreciate any suggestions.

It is part of a larger program. I could try to make a small program that shows the issue but I am not sure how easy that would be.

Regards,

James

The description for the underlying gather instruction says:
"If any pair of the index, mask, or destination registers are the same, this instruction results a UD fault"
Perhaps the compiler has used the same vector register for input and output here. This would be a compiler error. You can check this by debugging or disassembling the version of the exe file that fails. (A good disassembler is my objconv program).