[PATCH] D58666: [OpenCL] Undefine cl_intel_planar_yuv extension

[Anastasia Stulova via Phabricator via cfe-commits]

Anastasia accepted this revision.
Anastasia added a comment.
This revision is now accepted and ready to land.

LGTM! I am ok with this change to remove the macro from the common header. I don't think it belongs to there. However, I believe this functionality should be better represented in the vendors specific headers rather than clang directly. See my comment above!



================
Comment at: test/SemaOpenCL/extension-begin.cl:26
+
 #ifndef IMPLICIT_INCLUDE
 #include "extension-begin.h"
----------------
sidorovd wrote:
> Anastasia wrote:
> > sidorovd wrote:
> > > Anastasia wrote:
> > > > sidorovd wrote:
> > > > > Anastasia wrote:
> > > > > > Can we also test that macro `my_ext` is not defined here but defined above?
> > > > > > 
> > > > > > It seems we are not testing anything like this...
> > > > > #pragma OPENCL EXTENSION my_ext : begin doesn't define an appropriate macro. And so cl-ext=+my_ext.
> > > > But don't you need to expose the definition of it?
> > > Certainly I need, but now the only proper way to do so is by adding an extension via adding it in OpenCLExtensions.def. Previously we decided to avoid adding an extension directly into clang, so with a new approach I'd prefer not to add a definition of the macro in the header but define it somewhere else, otherwise the macro becomes defined  where it's not supposed to be (even for ARM and AMD =) ). 
> > However, my understanding is that you should define the macro when you define the extension itself.
> > 
> > 
> > ```
> > #pragma OPENCL EXTENSION my_ext : begin
> > #define my_ext
> > ...
> > #pragma OPENCL EXTENSION my_ext : end
> > ```
> > 
> > does it not work for you?
> ```
> #pragma OPENCL EXTENSION my_ext : begin
> #define my_ext
> ...
> #pragma OPENCL EXTENSION my_ext : end
> ```
>  ^
>  I
>  It defines the macro regardless begin : end range, so one can consider that it's the same code as:
> 
> ```
> #define my_ext
> #pragma OPENCL EXTENSION my_ext : begin
> ...
> #pragma OPENCL EXTENSION my_ext : end
> ```
> 
> I agree, that it's a better way to define a macro with defining an appropriate extension itself, but it makes it be defined where it's not supposed to be (at least from my perspective).
> 
> To sum up:
> 1. #pragma OPENCL EXTENSION my_ext : begin ... end  - makes an extension known for clang if the header included and if the extension enabled; doesn't affect a macro definition anyhow;
> 2.  OPENCLEXT_INTERNAL(my_ext, *version*, ~0U) -  makes an extension known for clang and defines an appropriate macro if the extension enabled.
> 
> I believe we are okay not to define cl_intel_planar_yuv macro in the header - just make the extension known for clang. 
> I believe we are okay not to define cl_intel_planar_yuv macro in the header - just make the extension known for clang.

We are not adding extensions to Clang unless they are needed for any internal functionality (i.e. to condition usage of half or double types that otherwise available in the standard C mode). If there is no special functionality needed then you should declare the macro in the header instead (probably the same place where you define the extension itself within `begin`/`end` to keep it clear). This is our official guidelines for vendor extensions currently.



CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D58666/new/

https://reviews.llvm.org/D58666