Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Pull 2019-10-09T10-36 Recent NVIDIA Changes #815

Merged
merged 9 commits into from
Oct 9, 2019
Merged

Pull 2019-10-09T10-36 Recent NVIDIA Changes #815

merged 9 commits into from
Oct 9, 2019

Conversation

gklimowicz
Copy link
Contributor

No description provided.

Improve codegen for map clause for explicitly
mapped scalars, pointers and arrays.  Scalars
are passed by value as they are implicitly
firstprivate. However, if they appear in map
clause, they should be pass by reference.
LOGICAL*1 and LOGICAL*2 dummy arguments were
generating incorrect LLVM IR, eventually leading
to a seg fault. This is because MSZ value was
being set as MSZ_WORD.  Adding a case for TY_BLOG
and TY_SLOG and setting MSZ_SBYTE and MSZ_SHWORD
respectively generates correct LLVM IR.
Check OpenMP's target data directive while
calculating array sections in map clause.

Array sections are still not supported in
Fortran, the whole array is sent to the
device, so semantic analysis for the array
section should be skipped for now. If not,
compilation fails as array sections are not
handled in the later phases in the compiler.
The right thing to do for the front end is to
skip it and gives a warning if the directive is
combined with "target" construct.

The compiler was not doing the right thing for
"target data", "target enter/exit data". This
patch adds check for them.

In short this patch fixes compilation fails for
the following codes:
    !$omp target enter data map(to:a(:,:) , b(:))
1) Add compilation flag "-m64"

2) Do not use -O2 and higher

3) Eliminate implicit constructor called at process initialization
   from logf and log10f for 3 constants used in the implementation of
   those routines for AVX2, KNL, and AVX512 ISA extensions.

4) Add header guard to runtime/libpgmath/lib/common/logf/common.h

5) Change system include header file <cmath> to <math.h>.
   Windows' <cmath> has an implicit dependency on runtime MSC_VER >= 1900.
@sscalpone sscalpone merged commit d07daf3 into flang-compiler:master Oct 9, 2019
@sscalpone sscalpone deleted the nv_stage branch October 9, 2019 20:28
kiranchandramohan added a commit to kiranchandramohan/flang that referenced this pull request Mar 31, 2021
A few old PRs (listed below) which carried the old license
were merged recently. This patch switches to the new license.

flang-compiler#839
flang-compiler#737
flang-compiler#765
flang-compiler#815
kiranchandramohan added a commit that referenced this pull request Apr 1, 2021
A few old PRs (listed below) which carried the old license
were merged recently. This patch switches to the new license.

#839
#737
#765
#815
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants