[PATCH] D47849: [OpenMP][Clang][NVPTX] Enable math functions called in an OpenMP NVPTX target device region to be resolved as device-native function calls

[Hal Finkel via Phabricator via cfe-commits]

hfinkel added a comment.

In https://reviews.llvm.org/D47849#1183134, @Hahnfeld wrote:

> In https://reviews.llvm.org/D47849#1124861, @hfinkel wrote:
>
> > In https://reviews.llvm.org/D47849#1124638, @Hahnfeld wrote:
> >
> > > 2. Incidentally I ran into a closely related problem: I can't `#include <math.h>` in translation units compiled for offloading, Clang complains about inline assembly for x86 (see below). Does that work for you?
> > >
> > >
> > >
> > >   In file included from /usr/include/math.h:413:
> > >   /usr/include/bits/mathinline.h:131:43: error: invalid input constraint 'x' in asm
> > >     __asm ("pmovmskb %1, %0" : "=r" (__m) : "x" (__x));
> > >                                             ^
> > >   /usr/include/bits/mathinline.h:143:43: error: invalid input constraint 'x' in asm
> > >     __asm ("pmovmskb %1, %0" : "=r" (__m) : "x" (__x));
> > >                                             ^
> > >   2 errors generated.
> >
> >
> > Hrmm. I thought that we had fixed that already.
> >
> > In case it's helpful, in an out-of-tree experimental target I have I ran into a similar problem, and to fix that I wrote the following code in the target's getTargetDefines function (in lib/Basic/Targets):
> >
> >   // If used as an OpenMP target on x86, x86 target feature macros are defined. math.h
> >   // and other system headers will include inline asm if these are defined.
> >   Builder.undefineMacro("__SSE2_MATH__");
> >   Builder.undefineMacro("__SSE_MATH__");
>
>
> Just found another workaround:
>
>   diff --git a/lib/Sema/SemaStmtAsm.cpp b/lib/Sema/SemaStmtAsm.cpp
>   index 0db15ea..b95f949 100644
>   --- a/lib/Sema/SemaStmtAsm.cpp
>   +++ b/lib/Sema/SemaStmtAsm.cpp
>   @@ -306,7 +306,9 @@ StmtResult Sema::ActOnGCCAsmStmt(SourceLocation AsmLoc, bool IsSimple,
>   
>        TargetInfo::ConstraintInfo Info(Literal->getString(), InputName);
>        if (!Context.getTargetInfo().validateInputConstraint(OutputConstraintInfos,
>   -                                                         Info)) {
>   +                                                         Info) &&
>   +        !(Context.getLangOpts().OpenMPIsDevice &&
>   +          Context.getSourceManager().isInSystemHeader(AsmLoc))) {
>          return StmtError(Diag(Literal->getLocStart(),
>                                diag::err_asm_invalid_input_constraint)
>                           << Info.getConstraintStr());
>
>
> This will ignore all errors during OpenMP device codegen from system headers when the inline assembly is not used. In that case (calling `signbit`) you'll get
>
>   In file included from math.c:2:
>   In file included from /usr/include/math.h:413:
>   /usr/include/bits/mathinline.h:143:10: error: couldn't allocate input reg for constraint 'x'
>     __asm ("pmovmskb %1, %0" : "=r" (__m) : "x" (__x));
>            ^
>   1 error generated.
>
>
> Not sure if that's acceptable...


Hrmm. Doesn't that make it so that whatever functions are implemented using that inline assembly will not be callable from target code (or, perhaps worse, will crash the backend if called)?


Repository:
  rC Clang

https://reviews.llvm.org/D47849