[libc][math] Fix signaling nans and floating point exceptions for some math functions.

[lntue]

The following issues with signaling nans and floating point exceptions are reported by @zimmermann6 with the CORE-MATH test suites:

CORE_MATH_CHECK_STD=true LIBM=/localdisk/zimmerma/llvm-project/build/projects/libc/lib/libllvmlibc.a ./check.sh $f

atan2
Running worst cases check in --rndn mode...
Missing inexact exception for x=0x0.0000000000001p-1022 y=0x1p+0 (z=0x0.0000000000001p-1022)

atan2f
Running worst cases check in --rndn mode...
Missing invalid exception for x=nan y=0x1p+0

exp
Running worst cases check in --rndn mode...
Error, foo(sNaN) should be qNaN, got sNaN=fff0000000000001

exp10
Running worst cases check in --rndn mode...
Error, foo(sNaN) should be qNaN, got sNaN=fff0000000000001

exp2
Running worst cases check in --rndn mode...
Error, foo(sNaN) should be qNaN, got sNaN=fff0000000000001

log
Running worst cases check in --rndn mode...
Error, foo(sNaN) should be qNaN, got sNaN=7ff0000000000001

log10
Running worst cases check in --rndn mode...
Error, foo(sNaN) should be qNaN, got sNaN=7ff0000000000001

log1p
Running worst cases check in --rndn mode...
Error, foo(sNaN) should be qNaN, got sNaN=7ff0000000000001

log2
Running worst cases check in --rndn mode...
Error, foo(sNaN) should be qNaN, got sNaN=7ff0000000000001

pow
Running worst cases check in --rndn mode...
Error, foo(sNaN,1.0) should be qNaN, got sNaN=7ff0000000000001

powf
Running worst cases check in --rndn mode...
Error, foo(sNaN,x) should be qNaN, got sNaN=7f800001

[llvmbot]

@llvm/issue-subscribers-libc

Author: None (lntue)

The following issues with signaling nans and floating point exceptions are reported by @zimmerman6 with the CORE-MATH test suites: ``` CORE_MATH_CHECK_STD=true LIBM=/localdisk/zimmerma/llvm-project/build/projects/libc/lib/libllvmlibc.a ./check.sh $f

atan2
Running worst cases check in --rndn mode...
Missing inexact exception for x=0x0.0000000000001p-1022 y=0x1p+0 (z=0x0.0000000000001p-1022)

atan2f
Running worst cases check in --rndn mode...
Missing invalid exception for x=nan y=0x1p+0

exp
Running worst cases check in --rndn mode...
Error, foo(sNaN) should be qNaN, got sNaN=fff0000000000001

exp10
Running worst cases check in --rndn mode...
Error, foo(sNaN) should be qNaN, got sNaN=fff0000000000001

exp2
Running worst cases check in --rndn mode...
Error, foo(sNaN) should be qNaN, got sNaN=fff0000000000001

log
Running worst cases check in --rndn mode...
Error, foo(sNaN) should be qNaN, got sNaN=7ff0000000000001

log10
Running worst cases check in --rndn mode...
Error, foo(sNaN) should be qNaN, got sNaN=7ff0000000000001

log1p
Running worst cases check in --rndn mode...
Error, foo(sNaN) should be qNaN, got sNaN=7ff0000000000001

log2
Running worst cases check in --rndn mode...
Error, foo(sNaN) should be qNaN, got sNaN=7ff0000000000001

pow
Running worst cases check in --rndn mode...
Error, foo(sNaN,1.0) should be qNaN, got sNaN=7ff0000000000001

powf
Running worst cases check in --rndn mode...
Error, foo(sNaN,x) should be qNaN, got sNaN=7f800001

</details>

hello can I get this assigned to me please. thanks

[wldfngrs]

@lara254 any updates on this issue?

[lntue]

@wldfngrs @lara254 for now I'm assigning this to both of you. There are many functions need to be investigate and update signaling nan treatment. Can you leave your comments here about functions that you investigated / going to investigate / fix so that we won't make redundant work? Thanks!

[wldfngrs]

Sure thing @lntue . One thing tho, how to quickly check for the functions with incorrect signaling nan treatment? You know, without having to read each function definition file directly. Is that possible?

[lntue]

Sure thing @lntue . One thing tho, how to quickly check for the functions with incorrect signaling nan treatment? You know, without having to read each function definition file directly. Is that possible?

So you can start with the list that the CORE-MATH test suite gave us above. After that, I think we can start going through the smoke tests, and see which ones are missing tests for qNaN, sNaN, and start adding them according to the standard to see if they fail/pass before looking into the implementations.

hello guys. hey @lntue can you unassigned me this issue. ill look for another issue later. thanks

[wldfngrs]

@lara254 any problems? If you're uncertain how to go about resolving this issue, I could guide you through it. If not, that's fine too.

@lara254 any problems? If you're uncertain how to go about resolving this issue, I could guide you through it. If not, that's fine too.

thanks for trying to help. but I just want to pick another issue.