internal::round_float gives incorrect results about hlslpp HOT 4 CLOSED

The issue was that std::round was not giving me the same results as hlsl for 0.5, but rounding 0.9 to -1 is definitely wrong, I'll take a look at that again, thanks for reporting

from hlslpp.

Can you try latest code? There was an issue with one of the conditions, it should be good now. I also verified that std::round is incorrect wrt to HLSL and the SIMD _mm_round_ps

from hlslpp.

That fixes my immediate issue, but I did not double check that the fixed code is correct for rounding negative numbers, does the unit test cover these cases? (In my particular case where I ran into this bug, negative numbers are clamped to zero after the rounding operation so any bugs in rounding negative numbers would not show up there.)

What are the differences between HLSL and std::round?

from hlslpp.

It's at x.5 that it goes wrong, not just with hlsl but with the simd intrinsics as well. 0.5 rounds to 1 with std::round but to 0 in the other cases.

The unit tests have negative numbers as well. However I may have to improve them because they clearly didn't catch the issue last time. I did do a few more tests locally.

from hlslpp.