You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
When I went through openlibm_fenv_i387.h and added OLM_DLLEXPORT, it felt like the __has_sse/__test_sse symbols should have been private, but they were used by the tests so had to be exported.
Looking through other headers, I see openlibm_fenv_amd64.h is better about having OLM_DLLEXPORT, but I notice it is missing from __fe_dfl_env. It seems that this would need to be exported for FE_DFL_ENV macro to work. Probably also need to be done in the arm.h as well.
openlibm_complex.h has noOLM_DLLEXPORT annotations. I assume this is intended to be a public API, yes?
There is some hackery for amd64 with regard to exporting inline functions. This hackery does not seem to be present for i387 or aarch64. I am not sure what precise problem it is intended to solve, but it seems like if it's necessary for amd64 it would be necessary for all.
The text was updated successfully, but these errors were encountered:
We can always replicate the has_sse stuff in the tests. It may be best to not export those from openlibm. Since things have been assembled and edited over time, it is believable that things are not done uniformly.
When I went through openlibm_fenv_i387.h and added
OLM_DLLEXPORT
, it felt like the__has_sse
/__test_sse
symbols should have been private, but they were used by the tests so had to be exported.Looking through other headers, I see openlibm_fenv_amd64.h is better about having
OLM_DLLEXPORT
, but I notice it is missing from__fe_dfl_env
. It seems that this would need to be exported forFE_DFL_ENV
macro to work. Probably also need to be done in the arm.h as well.openlibm_complex.h has no
OLM_DLLEXPORT
annotations. I assume this is intended to be a public API, yes?There is some hackery for amd64 with regard to exporting inline functions. This hackery does not seem to be present for i387 or aarch64. I am not sure what precise problem it is intended to solve, but it seems like if it's necessary for amd64 it would be necessary for all.
The text was updated successfully, but these errors were encountered: