[OpenCL] Add extensions cl_intel_subgroups and cl_intel_subgroups_short

[Anastasia]

lib/Headers/opencl-c.h

16128

Any advantage to use these macros here? They will complicate understanding of errors because preprocessor will remove information about the original BIFs being called in the source code.

[AlexeySotkin]

Replacing #define macros with explicit declarations

[Anastasia]

LGTM! Thanks!

[bader]

LGTM. Thanks!

[AlexeySotkin]

This comment was removed by AlexeySotkin.

include/clang/Basic/OpenCLExtensions.def

 
 // Clang Extensions.
 OPENCLEXT_INTERNAL(cl_clang_storage_class_specifiers, 100, ~0U)
 
 // AMD OpenCL extensions
 OPENCLEXT_INTERNAL(cl_amd_media_ops, 100, ~0U)
 OPENCLEXT_INTERNAL(cl_amd_media_ops2, 100, ~0U)
 
+// Intel OpenCL extensions
+OPENCLEXT_INTERNAL(cl_intel_subgroups, 120, ~0U)
+OPENCLEXT_INTERNAL(cl_intel_subgroups_short, 120, ~0U)
+
 #undef OPENCLEXT_INTERNAL
 
 #ifdef OPENCLEXT
 #undef OPENCLEXT
 #endif

lib/Headers/opencl-c.h

 double  __ovld __conv sub_group_scan_exclusive_max(double x);
 double  __ovld __conv sub_group_scan_inclusive_add(double x);
 double  __ovld __conv sub_group_scan_inclusive_min(double x);
 double  __ovld __conv sub_group_scan_inclusive_max(double x);
 #endif //cl_khr_fp64
 
 #endif //cl_khr_subgroups cl_intel_subgroups
 
+#if defined(cl_intel_subgroups)
+// Intel-Specific Sub Group Functions
+float   __ovld __conv intel_sub_group_shuffle( float  x, uint c );
+float2  __ovld __conv intel_sub_group_shuffle( float2 x, uint c );
+float3  __ovld __conv intel_sub_group_shuffle( float3 x, uint c );
+float4  __ovld __conv intel_sub_group_shuffle( float4 x, uint c );
+float8  __ovld __conv intel_sub_group_shuffle( float8 x, uint c );
+float16 __ovld __conv intel_sub_group_shuffle( float16 x, uint c );
+
+int     __ovld __conv intel_sub_group_shuffle( int  x, uint c );
+int2    __ovld __conv intel_sub_group_shuffle( int2 x, uint c );
+int3    __ovld __conv intel_sub_group_shuffle( int3 x, uint c );
+int4    __ovld __conv intel_sub_group_shuffle( int4 x, uint c );
+int8    __ovld __conv intel_sub_group_shuffle( int8 x, uint c );
+int16   __ovld __conv intel_sub_group_shuffle( int16 x, uint c );
+
+uint    __ovld __conv intel_sub_group_shuffle( uint  x, uint c );
+uint2   __ovld __conv intel_sub_group_shuffle( uint2 x, uint c );
+uint3   __ovld __conv intel_sub_group_shuffle( uint3 x, uint c );
+uint4   __ovld __conv intel_sub_group_shuffle( uint4 x, uint c );
+uint8   __ovld __conv intel_sub_group_shuffle( uint8 x, uint c );
+uint16  __ovld __conv intel_sub_group_shuffle( uint16 x, uint c );
+
+long    __ovld __conv intel_sub_group_shuffle( long x, uint c );
+ulong   __ovld __conv intel_sub_group_shuffle( ulong x, uint c );
+
+float   __ovld __conv intel_sub_group_shuffle_down( float  cur, float  next, uint c );
+float2  __ovld __conv intel_sub_group_shuffle_down( float2 cur, float2 next, uint c );
+float3  __ovld __conv intel_sub_group_shuffle_down( float3 cur, float3 next, uint c );
+float4  __ovld __conv intel_sub_group_shuffle_down( float4 cur, float4 next, uint c );
+float8  __ovld __conv intel_sub_group_shuffle_down( float8 cur, float8 next, uint c );
+float16 __ovld __conv intel_sub_group_shuffle_down( float16 cur, float16 next, uint c );
+
+int     __ovld __conv intel_sub_group_shuffle_down( int  cur, int  next, uint c );
+int2    __ovld __conv intel_sub_group_shuffle_down( int2 cur, int2 next, uint c );
+int3    __ovld __conv intel_sub_group_shuffle_down( int3 cur, int3 next, uint c );
+int4    __ovld __conv intel_sub_group_shuffle_down( int4 cur, int4 next, uint c );
+int8    __ovld __conv intel_sub_group_shuffle_down( int8 cur, int8 next, uint c );
+int16   __ovld __conv intel_sub_group_shuffle_down( int16 cur, int16 next, uint c );
+
+uint    __ovld __conv intel_sub_group_shuffle_down( uint  cur, uint  next, uint c );
+uint2   __ovld __conv intel_sub_group_shuffle_down( uint2 cur, uint2 next, uint c );
+uint3   __ovld __conv intel_sub_group_shuffle_down( uint3 cur, uint3 next, uint c );
+uint4   __ovld __conv intel_sub_group_shuffle_down( uint4 cur, uint4 next, uint c );
+uint8   __ovld __conv intel_sub_group_shuffle_down( uint8 cur, uint8 next, uint c );
+uint16  __ovld __conv intel_sub_group_shuffle_down( uint16 cur, uint16 next, uint c );
+
+long    __ovld __conv intel_sub_group_shuffle_down( long prev, long cur, uint c );
+ulong   __ovld __conv intel_sub_group_shuffle_down( ulong prev, ulong cur, uint c );
+
+float   __ovld __conv intel_sub_group_shuffle_up( float  prev, float  cur, uint c );
+float2  __ovld __conv intel_sub_group_shuffle_up( float2 prev, float2 cur, uint c );
+float3  __ovld __conv intel_sub_group_shuffle_up( float3 prev, float3 cur, uint c );
+float4  __ovld __conv intel_sub_group_shuffle_up( float4 prev, float4 cur, uint c );
+float8  __ovld __conv intel_sub_group_shuffle_up( float8 prev, float8 cur, uint c );
+float16 __ovld __conv intel_sub_group_shuffle_up( float16 prev, float16 cur, uint c );
+
+int     __ovld __conv intel_sub_group_shuffle_up( int  prev, int  cur, uint c );
+int2    __ovld __conv intel_sub_group_shuffle_up( int2 prev, int2 cur, uint c );
+int3    __ovld __conv intel_sub_group_shuffle_up( int3 prev, int3 cur, uint c );
+int4    __ovld __conv intel_sub_group_shuffle_up( int4 prev, int4 cur, uint c );
+int8    __ovld __conv intel_sub_group_shuffle_up( int8 prev, int8 cur, uint c );
+int16   __ovld __conv intel_sub_group_shuffle_up( int16 prev, int16 cur, uint c );
+
+uint    __ovld __conv intel_sub_group_shuffle_up( uint  prev, uint  cur, uint c );
+uint2   __ovld __conv intel_sub_group_shuffle_up( uint2 prev, uint2 cur, uint c );
+uint3   __ovld __conv intel_sub_group_shuffle_up( uint3 prev, uint3 cur, uint c );
+uint4   __ovld __conv intel_sub_group_shuffle_up( uint4 prev, uint4 cur, uint c );
+uint8   __ovld __conv intel_sub_group_shuffle_up( uint8 prev, uint8 cur, uint c );
+uint16  __ovld __conv intel_sub_group_shuffle_up( uint16 prev, uint16 cur, uint c );
+
+long    __ovld __conv intel_sub_group_shuffle_up( long prev, long cur, uint c );
+ulong   __ovld __conv intel_sub_group_shuffle_up( ulong prev, ulong cur, uint c );
+
+float   __ovld __conv intel_sub_group_shuffle_xor( float  x, uint c );
+float2  __ovld __conv intel_sub_group_shuffle_xor( float2 x, uint c );
+float3  __ovld __conv intel_sub_group_shuffle_xor( float3 x, uint c );
+float4  __ovld __conv intel_sub_group_shuffle_xor( float4 x, uint c );
+float8  __ovld __conv intel_sub_group_shuffle_xor( float8 x, uint c );
+float16 __ovld __conv intel_sub_group_shuffle_xor( float16 x, uint c );
+
+int     __ovld __conv intel_sub_group_shuffle_xor( int  x, uint c );
+int2    __ovld __conv intel_sub_group_shuffle_xor( int2 x, uint c );
+int3    __ovld __conv intel_sub_group_shuffle_xor( int3 x, uint c );
+int4    __ovld __conv intel_sub_group_shuffle_xor( int4 x, uint c );
+int8    __ovld __conv intel_sub_group_shuffle_xor( int8 x, uint c );
+int16   __ovld __conv intel_sub_group_shuffle_xor( int16 x, uint c );
+
+uint    __ovld __conv intel_sub_group_shuffle_xor( uint  x, uint c );
+uint2   __ovld __conv intel_sub_group_shuffle_xor( uint2 x, uint c );
+uint3   __ovld __conv intel_sub_group_shuffle_xor( uint3 x, uint c );
+uint4   __ovld __conv intel_sub_group_shuffle_xor( uint4 x, uint c );
+uint8   __ovld __conv intel_sub_group_shuffle_xor( uint8 x, uint c );
+uint16  __ovld __conv intel_sub_group_shuffle_xor( uint16 x, uint c );
+
+long    __ovld __conv intel_sub_group_shuffle_xor( long x, uint c );
+ulong   __ovld __conv intel_sub_group_shuffle_xor( ulong x, uint c );
+
+uint    __ovld __conv intel_sub_group_block_read( read_only image2d_t image, int2 coord );
+uint2   __ovld __conv intel_sub_group_block_read2( read_only image2d_t image, int2 coord );
+uint4   __ovld __conv intel_sub_group_block_read4( read_only image2d_t image, int2 coord );
+uint8   __ovld __conv intel_sub_group_block_read8( read_only image2d_t image, int2 coord );
+
+#if (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+uint    __ovld __conv intel_sub_group_block_read(read_write image2d_t image, int2 coord);
+uint2   __ovld __conv intel_sub_group_block_read2(read_write image2d_t image, int2 coord);
+uint4   __ovld __conv intel_sub_group_block_read4(read_write image2d_t image, int2 coord);
+uint8   __ovld __conv intel_sub_group_block_read8(read_write image2d_t image, int2 coord);
+#endif // (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+
+uint    __ovld __conv intel_sub_group_block_read( const __global uint* p );
+uint2   __ovld __conv intel_sub_group_block_read2( const __global uint* p );
+uint4   __ovld __conv intel_sub_group_block_read4( const __global uint* p );
+uint8   __ovld __conv intel_sub_group_block_read8( const __global uint* p );
+
+void    __ovld __conv intel_sub_group_block_write(write_only image2d_t image, int2 coord, uint data);
+void    __ovld __conv intel_sub_group_block_write2(write_only image2d_t image, int2 coord, uint2 data);
+void    __ovld __conv intel_sub_group_block_write4(write_only image2d_t image, int2 coord, uint4 data);
+void    __ovld __conv intel_sub_group_block_write8(write_only image2d_t image, int2 coord, uint8 data);
+
+#if (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+void    __ovld __conv intel_sub_group_block_write(read_write image2d_t image, int2 coord, uint data);
+void    __ovld __conv intel_sub_group_block_write2(read_write image2d_t image, int2 coord, uint2 data);
+void    __ovld __conv intel_sub_group_block_write4(read_write image2d_t image, int2 coord, uint4 data);
+void    __ovld __conv intel_sub_group_block_write8(read_write image2d_t image, int2 coord, uint8 data);
+#endif // (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+
+void    __ovld __conv intel_sub_group_block_write( __global uint* p, uint data );
+void    __ovld __conv intel_sub_group_block_write2( __global uint* p, uint2 data );
+void    __ovld __conv intel_sub_group_block_write4( __global uint* p, uint4 data );
+void    __ovld __conv intel_sub_group_block_write8( __global uint* p, uint8 data );
+
+#ifdef cl_khr_fp16
+half    __ovld __conv intel_sub_group_shuffle( half x, uint c );
+half    __ovld __conv intel_sub_group_shuffle_down( half prev, half cur, uint c );
+half    __ovld __conv intel_sub_group_shuffle_up( half prev, half cur, uint c );
+half    __ovld __conv intel_sub_group_shuffle_xor( half x, uint c );
+#endif
+
+#if defined(cl_khr_fp64)
+double  __ovld __conv intel_sub_group_shuffle( double x, uint c );
+double  __ovld __conv intel_sub_group_shuffle_down( double prev, double cur, uint c );
+double  __ovld __conv intel_sub_group_shuffle_up( double prev, double cur, uint c );
+double  __ovld __conv intel_sub_group_shuffle_xor( double x, uint c );
+#endif
+
+#endif //cl_intel_subgroups
+
+#if defined(cl_intel_subgroups_short)
+short       __ovld __conv intel_sub_group_broadcast( short  x, uint sub_group_local_id );
+short2      __ovld __conv intel_sub_group_broadcast( short2 x, uint sub_group_local_id );
+short3      __ovld __conv intel_sub_group_broadcast( short3 x, uint sub_group_local_id );
+short4      __ovld __conv intel_sub_group_broadcast( short4 x, uint sub_group_local_id );
+short8      __ovld __conv intel_sub_group_broadcast( short8 x, uint sub_group_local_id );
+
+ushort      __ovld __conv intel_sub_group_broadcast( ushort  x, uint sub_group_local_id );
+ushort2     __ovld __conv intel_sub_group_broadcast( ushort2 x, uint sub_group_local_id );
+ushort3     __ovld __conv intel_sub_group_broadcast( ushort3 x, uint sub_group_local_id );
+ushort4     __ovld __conv intel_sub_group_broadcast( ushort4 x, uint sub_group_local_id );
+ushort8     __ovld __conv intel_sub_group_broadcast( ushort8 x, uint sub_group_local_id );
+
+short       __ovld __conv intel_sub_group_shuffle( short   x, uint c );
+short2      __ovld __conv intel_sub_group_shuffle( short2  x, uint c );
+short3      __ovld __conv intel_sub_group_shuffle( short3  x, uint c );
+short4      __ovld __conv intel_sub_group_shuffle( short4  x, uint c );
+short8      __ovld __conv intel_sub_group_shuffle( short8  x, uint c );
+short16     __ovld __conv intel_sub_group_shuffle( short16 x, uint c);
+
+ushort      __ovld __conv intel_sub_group_shuffle( ushort   x, uint c );
+ushort2     __ovld __conv intel_sub_group_shuffle( ushort2  x, uint c );
+ushort3     __ovld __conv intel_sub_group_shuffle( ushort3  x, uint c );
+ushort4     __ovld __conv intel_sub_group_shuffle( ushort4  x, uint c );
+ushort8     __ovld __conv intel_sub_group_shuffle( ushort8  x, uint c );
+ushort16    __ovld __conv intel_sub_group_shuffle( ushort16 x, uint c );
+
+short       __ovld __conv intel_sub_group_shuffle_down( short   cur, short   next, uint c );
+short2      __ovld __conv intel_sub_group_shuffle_down( short2  cur, short2  next, uint c );
+short3      __ovld __conv intel_sub_group_shuffle_down( short3  cur, short3  next, uint c );
+short4      __ovld __conv intel_sub_group_shuffle_down( short4  cur, short4  next, uint c );
+short8      __ovld __conv intel_sub_group_shuffle_down( short8  cur, short8  next, uint c );
+short16     __ovld __conv intel_sub_group_shuffle_down( short16 cur, short16 next, uint c );
+
+ushort      __ovld __conv intel_sub_group_shuffle_down( ushort   cur, ushort   next, uint c );
+ushort2     __ovld __conv intel_sub_group_shuffle_down( ushort2  cur, ushort2  next, uint c );
+ushort3     __ovld __conv intel_sub_group_shuffle_down( ushort3  cur, ushort3  next, uint c );
+ushort4     __ovld __conv intel_sub_group_shuffle_down( ushort4  cur, ushort4  next, uint c );
+ushort8     __ovld __conv intel_sub_group_shuffle_down( ushort8  cur, ushort8  next, uint c );
+ushort16    __ovld __conv intel_sub_group_shuffle_down( ushort16 cur, ushort16 next, uint c );
+
+short       __ovld __conv intel_sub_group_shuffle_up( short   cur, short   next, uint c );
+short2      __ovld __conv intel_sub_group_shuffle_up( short2  cur, short2  next, uint c );
+short3      __ovld __conv intel_sub_group_shuffle_up( short3  cur, short3  next, uint c );
+short4      __ovld __conv intel_sub_group_shuffle_up( short4  cur, short4  next, uint c );
+short8      __ovld __conv intel_sub_group_shuffle_up( short8  cur, short8  next, uint c );
+short16     __ovld __conv intel_sub_group_shuffle_up( short16 cur, short16 next, uint c );
+
+ushort      __ovld __conv intel_sub_group_shuffle_up( ushort   cur, ushort   next, uint c );
+ushort2     __ovld __conv intel_sub_group_shuffle_up( ushort2  cur, ushort2  next, uint c );
+ushort3     __ovld __conv intel_sub_group_shuffle_up( ushort3  cur, ushort3  next, uint c );
+ushort4     __ovld __conv intel_sub_group_shuffle_up( ushort4  cur, ushort4  next, uint c );
+ushort8     __ovld __conv intel_sub_group_shuffle_up( ushort8  cur, ushort8  next, uint c );
+ushort16    __ovld __conv intel_sub_group_shuffle_up( ushort16 cur, ushort16 next, uint c );
+
+short       __ovld __conv intel_sub_group_shuffle_xor( short   x, uint c );
+short2      __ovld __conv intel_sub_group_shuffle_xor( short2  x, uint c );
+short3      __ovld __conv intel_sub_group_shuffle_xor( short3  x, uint c );
+short4      __ovld __conv intel_sub_group_shuffle_xor( short4  x, uint c );
+short8      __ovld __conv intel_sub_group_shuffle_xor( short8  x, uint c );
+short16     __ovld __conv intel_sub_group_shuffle_xor( short16 x, uint c );
+
+ushort      __ovld __conv intel_sub_group_shuffle_xor( ushort   x, uint c );
+ushort2     __ovld __conv intel_sub_group_shuffle_xor( ushort2  x, uint c );
+ushort3     __ovld __conv intel_sub_group_shuffle_xor( ushort3  x, uint c );
+ushort4     __ovld __conv intel_sub_group_shuffle_xor( ushort4  x, uint c );
+ushort8     __ovld __conv intel_sub_group_shuffle_xor( ushort8  x, uint c );
+ushort16    __ovld __conv intel_sub_group_shuffle_xor( ushort16 x, uint c );
+
+short       __ovld __conv intel_sub_group_reduce_add( short   x );
+ushort      __ovld __conv intel_sub_group_reduce_add( ushort  x );
+short       __ovld __conv intel_sub_group_reduce_min( short   x );
+ushort      __ovld __conv intel_sub_group_reduce_min( ushort  x );
+short       __ovld __conv intel_sub_group_reduce_max( short   x );
+ushort      __ovld __conv intel_sub_group_reduce_max( ushort  x );
+
+short       __ovld __conv intel_sub_group_scan_exclusive_add( short   x );
+ushort      __ovld __conv intel_sub_group_scan_exclusive_add( ushort  x );
+short       __ovld __conv intel_sub_group_scan_exclusive_min( short   x );
+ushort      __ovld __conv intel_sub_group_scan_exclusive_min( ushort  x );
+short       __ovld __conv intel_sub_group_scan_exclusive_max( short   x );
+ushort      __ovld __conv intel_sub_group_scan_exclusive_max( ushort  x );
+
+short       __ovld __conv intel_sub_group_scan_inclusive_add( short   x );
+ushort      __ovld __conv intel_sub_group_scan_inclusive_add( ushort  x );
+short       __ovld __conv intel_sub_group_scan_inclusive_min( short   x );
+ushort      __ovld __conv intel_sub_group_scan_inclusive_min( ushort  x );
+short       __ovld __conv intel_sub_group_scan_inclusive_max( short   x );
+ushort      __ovld __conv intel_sub_group_scan_inclusive_max( ushort  x );
+
+uint       __ovld __conv intel_sub_group_block_read_ui( read_only image2d_t image, int2 byte_coord );
+uint2      __ovld __conv intel_sub_group_block_read_ui2( read_only image2d_t image, int2 byte_coord );
+uint4      __ovld __conv intel_sub_group_block_read_ui4( read_only image2d_t image, int2 byte_coord );
+uint8      __ovld __conv intel_sub_group_block_read_ui8( read_only image2d_t image, int2 byte_coord );
+
+#if (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+uint       __ovld __conv intel_sub_group_block_read_ui( read_write image2d_t image, int2 byte_coord );
+uint2      __ovld __conv intel_sub_group_block_read_ui2( read_write image2d_t image, int2 byte_coord );
+uint4      __ovld __conv intel_sub_group_block_read_ui4( read_write image2d_t image, int2 byte_coord );
+uint8      __ovld __conv intel_sub_group_block_read_ui8( read_write image2d_t image, int2 byte_coord );
+#endif // (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+
+uint       __ovld __conv intel_sub_group_block_read_ui( const __global uint* p );
+uint2      __ovld __conv intel_sub_group_block_read_ui2( const __global uint* p );
+uint4      __ovld __conv intel_sub_group_block_read_ui4( const __global uint* p );
+uint8      __ovld __conv intel_sub_group_block_read_ui8( const __global uint* p );
+
+void       __ovld __conv intel_sub_group_block_write_ui( read_only image2d_t image, int2 byte_coord, uint data );
+void       __ovld __conv intel_sub_group_block_write_ui2( read_only image2d_t image, int2 byte_coord, uint2 data );
+void       __ovld __conv intel_sub_group_block_write_ui4( read_only image2d_t image, int2 byte_coord, uint4 data );
+void       __ovld __conv intel_sub_group_block_write_ui8( read_only image2d_t image, int2 byte_coord, uint8 data );
+
+#if (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+void       __ovld __conv intel_sub_group_block_write_ui( read_write image2d_t image, int2 byte_coord, uint data );
+void       __ovld __conv intel_sub_group_block_write_ui2( read_write image2d_t image, int2 byte_coord, uint2 data );
+void       __ovld __conv intel_sub_group_block_write_ui4( read_write image2d_t image, int2 byte_coord, uint4 data );
+void       __ovld __conv intel_sub_group_block_write_ui8( read_write image2d_t image, int2 byte_coord, uint8 data );
+#endif // (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+
+void       __ovld __conv intel_sub_group_block_write_ui( __global uint* p, uint data );
+void       __ovld __conv intel_sub_group_block_write_ui2( __global uint* p, uint2 data );
+void       __ovld __conv intel_sub_group_block_write_ui4( __global uint* p, uint4 data );
+void       __ovld __conv intel_sub_group_block_write_ui8( __global uint* p, uint8 data );
+
+ushort      __ovld __conv intel_sub_group_block_read_us( read_only image2d_t image, int2 coord );
+ushort2     __ovld __conv intel_sub_group_block_read_us2( read_only image2d_t image, int2 coord );
+ushort4     __ovld __conv intel_sub_group_block_read_us4( read_only image2d_t image, int2 coord );
+ushort8     __ovld __conv intel_sub_group_block_read_us8( read_only image2d_t image, int2 coord );
+
+#if (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+ushort      __ovld __conv intel_sub_group_block_read_us(read_write image2d_t image, int2 coord);
+ushort2     __ovld __conv intel_sub_group_block_read_us2(read_write image2d_t image, int2 coord);
+ushort4     __ovld __conv intel_sub_group_block_read_us4(read_write image2d_t image, int2 coord);
+ushort8     __ovld __conv intel_sub_group_block_read_us8(read_write image2d_t image, int2 coord);
+#endif // (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+
+ushort      __ovld __conv intel_sub_group_block_read_us(  const __global ushort* p );
+ushort2     __ovld __conv intel_sub_group_block_read_us2( const __global ushort* p );
+ushort4     __ovld __conv intel_sub_group_block_read_us4( const __global ushort* p );
+ushort8     __ovld __conv intel_sub_group_block_read_us8( const __global ushort* p );
+
+void        __ovld __conv intel_sub_group_block_write_us(write_only image2d_t image, int2 coord, ushort  data);
+void        __ovld __conv intel_sub_group_block_write_us2(write_only image2d_t image, int2 coord, ushort2 data);
+void        __ovld __conv intel_sub_group_block_write_us4(write_only image2d_t image, int2 coord, ushort4 data);
+void        __ovld __conv intel_sub_group_block_write_us8(write_only image2d_t image, int2 coord, ushort8 data);
+
+#if (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+void        __ovld __conv intel_sub_group_block_write_us(read_write image2d_t image, int2 coord, ushort  data);
+void        __ovld __conv intel_sub_group_block_write_us2(read_write image2d_t image, int2 coord, ushort2 data);
+void        __ovld __conv intel_sub_group_block_write_us4(read_write image2d_t image, int2 coord, ushort4 data);
+void        __ovld __conv intel_sub_group_block_write_us8(read_write image2d_t image, int2 coord, ushort8 data);
+#endif // (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+
+void        __ovld __conv intel_sub_group_block_write_us(  __global ushort* p, ushort  data );
+void        __ovld __conv intel_sub_group_block_write_us2( __global ushort* p, ushort2 data );
+void        __ovld __conv intel_sub_group_block_write_us4( __global ushort* p, ushort4 data );
+void        __ovld __conv intel_sub_group_block_write_us8( __global ushort* p, ushort8 data );
+#endif // cl_intel_subgroups_short
+
 #ifdef cl_amd_media_ops
 uint __ovld amd_bitalign(uint a, uint b, uint c);
 uint2 __ovld amd_bitalign(uint2 a, uint2 b, uint2 c);
 uint3 __ovld amd_bitalign(uint3 a, uint3 b, uint3 c);
 uint4 __ovld amd_bitalign(uint4 a, uint4 b, uint4 c);
 uint8 __ovld amd_bitalign(uint8 a, uint8 b, uint8 c);
 uint16 __ovld amd_bitalign(uint16 a, uint16 b, uint16 c);
 

test/SemaOpenCL/extension-version.cl

 #endif
 #pragma OPENCL EXTENSION cl_amd_media_ops: enable
 
 #ifndef cl_amd_media_ops2
 #error "Missing cl_amd_media_ops2 define"
 #endif
 #pragma OPENCL EXTENSION cl_amd_media_ops2: enable
 
+#if (__OPENCL_C_VERSION__ >= 120)
+#ifndef cl_intel_subgroups
+#error "Missing cl_intel_subgroups define"
+#endif
+#else
+// expected-warning@+2{{unsupported OpenCL extension 'cl_intel_subgroups' - ignoring}}
+#endif
+#pragma OPENCL EXTENSION cl_intel_subgroups : enable
+
+#if (__OPENCL_C_VERSION__ >= 120)
+#ifndef cl_intel_subgroups_short
+#error "Missing cl_intel_subgroups_short define"
+#endif
+#else
+// expected-warning@+2{{unsupported OpenCL extension 'cl_intel_subgroups_short' - ignoring}}
+#endif
+#pragma OPENCL EXTENSION cl_intel_subgroups_short : enable
+