[PATCH] D97949: [AArch64] Add missing intrinsics for vrnd

[JinGu Kang via Phabricator via llvm-commits]

jaykang10 added inline comments.


================
Comment at: clang/lib/Basic/Targets/AArch64.cpp:185
                                                 MacroBuilder &Builder) const {
+  Builder.defineMacro("__ARM_FEATURE_FRINT", "1");
   // Also include the Armv8.4 defines
----------------
SjoerdMeijer wrote:
> jaykang10 wrote:
> > SjoerdMeijer wrote:
> > > SjoerdMeijer wrote:
> > > > I think we need to add a test new macro in:
> > > > 
> > > >   Preprocessor/aarch64-target-features.c
> > > > 
> > > Sorry, wanted to say that we need to add a test for this new macro.
> > The macro is used in "arm_neon.h" which is generated at building time of clang as below so I thought we do not need to add the test for it.
> > ```
> > #if __ARM_ARCH >= 8 && defined(__aarch64__) && defined(__ARM_FEATURE_FRINT)
> > ...
> > #ifdef __LITTLE_ENDIAN__
> > __ai float32x2_t vrnd32x_f32(float32x2_t __p0) {
> >   float32x2_t __ret;
> >   __ret = (float32x2_t) __builtin_neon_vrnd32x_v((int8x8_t)__p0, 9);
> >   return __ret;
> > }
> > #else
> > __ai float32x2_t vrnd32x_f32(float32x2_t __p0) {
> >   float32x2_t __rev0;  __rev0 = __builtin_shufflevector(__p0, __p0, 1, 0);
> >   float32x2_t __ret;
> >   __ret = (float32x2_t) __builtin_neon_vrnd32x_v((int8x8_t)__rev0, 9);
> >   __ret = __builtin_shufflevector(__ret, __ret, 1, 0);
> >   return __ret;
> > }
> > #endif
> > ...
> > ```
> > If it is not enough, I will add a test for it.
> We let the preprocessor emit a new macro, so that's what we need to test, which I believe we need to do in that file I quoted earlier. The fact that it is used elsewhere is an indirect test.
Yep, I will add it.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D97949/new/

https://reviews.llvm.org/D97949