Add copysignf16, copysignf128, fabsf16, and fabsf128
#320
Conversation
tgross35
force-pushed
the
generic-algorithms
branch
2 times, most recently
from
eb24e93
to
965cf5a
Compare
|
Using traits blocks us from using functions in |
|
It would indeed prevent us from making any of the public functions I have some more details here if you are interested #310 |
Who has decided this? |
|
It has been a goal for a while to make the float math functions available in The existing implementations aren't going away either. Generic methods just make it easier to support |
tgross35
force-pushed
the
generic-algorithms
branch
from
9e343ff
to
87f0631
Compare
tgross35
changed the title
copysignf16, copysignf128, fabsf16, and fabsf128
tgross35
force-pushed
the
generic-algorithms
branch
from
87f0631
to
733b768
Compare
Don't try to generate tests for directories, or for files that contain `f16` or `f128` (as these types are not provided by musl's math implementations).
tgross35
force-pushed
the
generic-algorithms
branch
from
733b768
to
c6f95c4
Compare
In preparation of adding generic algorithms to `libm`, add the traits from `compiler-builtins`. Eventually we should be able to unify the two crates so we don't have duplicate implementations.
In preparation of adding routines from these two types, duplicate the `compiler-builtins` configuration here.
tgross35
force-pushed
the
generic-algorithms
branch
from
c6f95c4
to
7b15771
Compare
Currently there is a macro called `llvm_intrinsically_optimized` that uses an intrinsic rather than the function implementation if the configuration is correct. Add a new macro `select_implementation` that is somewhat cleaner to use, and make use of it in two relevant locations. In the future, we can update this macro with more fields to specify other implementations that may be selected, such as something architecture-specific or e.g. using a generic implementation for `f32` routines, rather than those that convert to `f64`. This also introduces a `macros` module within `math/support`. We will be able to move more things here later.
Add generic versions of `abs` and `copysign`. Make use of it for the `f32` and `f64` versions of these functions.
`cfg_if` is helpful for applying `cfg` attributes to groups of items, like we will need to do with `f16` and `f128` to properly gate them. However, `libm` can't have dependencies. The `cfg_if` macro is small, so just vendor it here.
Use the generic algorithm to make these two functions available whenever `f16_enabled` or `f128_enabled` are true. These require the `unstable` feature.
tgross35
force-pushed
the
generic-algorithms
branch
from
7b15771
to
470ab59
Compare
Add generic versions of
absandcopysign, which is used to implementf16andf128versions of these functions. These currently aren't tested because the builtinmusltests don't provide a reference, but the implementations are straightforward and we will be able to test against #311 once it lands. (Technically musl on aarch64 hasfabslandcopysignlasf128, but I don't think it is worth updating the serialization tests).The second commit ("Add float and integer traits...") just copies the exact trait implementations that we use in
compiler-builtins, with the assumption that we will use it more in the future and for testing.The third commit ("Add f16 and f128 configuration...") copies the logic from
compiler-builtinsfor settingenable_f16andenable_f128with the exception that here, theunstablefeature must also be enabled. This will probably need to be tweaked a bit in the future so it isn't tied to other behavior enabled byunstable, but this is enough to get us started.