Compute values for rational magnitude powers #601
Conversation
Since this will only ever be done at compile time (as guaranteed by using `consteval`), we can afford to prioritize precision over speed. To compute an Nth root, we simply do a binary search over representable floating point numbers, looking for the number whose Nth power most closely matches the original number. Fixes mpusz#494. We have included a test case reproducing the original problem exactly. All tests use "within 4 ULPs" as the criterion, which is (I believe) equivalent to the googletest `EXPECT_DOUBLE_EQ` criterion.
|
"Formatting CI" fails. Please run pre-commit according to the instructions in https://github.com/mpusz/mp-units/blob/master/CONTRIBUTING.md#unified-code-formatting. |
|
mp-units/test/static/math_test.cpp Line 39 in f5eda4a |
|
It seems that the libc++-18 issues are fixed now, and a few problems have surfaced in our code. I fixed all of them today. Please rebase to master and fix the remaining issues in your code. There are at least two more issues to be fixed: |
|
Apple clang does not like this linkage:
It's weird, because the undefined symbol is Maybe the solution is more obvious to you. |
|
This was also the case for clang-16: Hopefully, now it will be OK. |
|
Really glad I asked for help instead of trying to solve it on my own --- I would never have come up with that! |
|
Yeah, I had a few issues with clang-16 already 😉 |
|
I had to manually restart each of 3 failed jobs. The jobs gave no reason for their failure, and all of them passed on retry. In any case, we're at a 100% pass rate right now! |
Since this will only ever be done at compile time (as guaranteed by
using
consteval), we can afford to prioritize precision over speed. Tocompute an Nth root, we simply do a binary search over representable
floating point numbers, looking for the number whose Nth power most
closely matches the original number.
Fixes #474. We have included a test case reproducing the original
problem exactly. All tests use "within 4 ULPs" as the criterion, which
is (I believe) equivalent to the googletest
EXPECT_DOUBLE_EQcriterion.