[OpenCL] Remove decls for scalar vloada_half and vstorea_half* fns

[stuart]

Summary

These functions are not part of the OpenCL C specification.

See https://github.com/KhronosGroup/OpenCL-Docs/issues/648 for a clarification regarding the vloada_half declarations.

[Anastasia]

LGTM! Thanks

clang/lib/Headers/opencl-c.h

  * value and the floatn value is returned.
  * The address computed as (p + (offset * n))
  * must be aligned to sizeof (halfn) bytes.
  * For n = 3, vloada_half3 reads a half3 from
  * address (p + (offset * 4)) and returns a float3.
  * The address computed as (p + (offset * 4))
  * must be aligned to sizeof (half) * 4 bytes.
  */
-float __ovld vloada_half(size_t offset, const __constant half *p);
 float2 __ovld vloada_half2(size_t offset, const __constant half *p);
 float3 __ovld vloada_half3(size_t offset, const __constant half *p);
 float4 __ovld vloada_half4(size_t offset, const __constant half *p);
 float8 __ovld vloada_half8(size_t offset, const __constant half *p);
 float16 __ovld vloada_half16(size_t offset, const __constant half *p);
 #if defined(__opencl_c_generic_address_space)
-float __ovld vloada_half(size_t offset, const half *p);
 float2 __ovld vloada_half2(size_t offset, const half *p);
 float3 __ovld vloada_half3(size_t offset, const half *p);
 float4 __ovld vloada_half4(size_t offset, const half *p);
 float8 __ovld vloada_half8(size_t offset, const half *p);
 float16 __ovld vloada_half16(size_t offset, const half *p);
 #else
-float __ovld vloada_half(size_t offset, const __global half *p);
 float2 __ovld vloada_half2(size_t offset, const __global half *p);
 float3 __ovld vloada_half3(size_t offset, const __global half *p);
 float4 __ovld vloada_half4(size_t offset, const __global half *p);
 float8 __ovld vloada_half8(size_t offset, const __global half *p);
 float16 __ovld vloada_half16(size_t offset, const __global half *p);
-float __ovld vloada_half(size_t offset, const __local half *p);
 float2 __ovld vloada_half2(size_t offset, const __local half *p);
 float3 __ovld vloada_half3(size_t offset, const __local half *p);
 float4 __ovld vloada_half4(size_t offset, const __local half *p);
 float8 __ovld vloada_half8(size_t offset, const __local half *p);
 float16 __ovld vloada_half16(size_t offset, const __local half *p);
-float __ovld vloada_half(size_t offset, const __private half *p);
 float2 __ovld vloada_half2(size_t offset, const __private half *p);
 float3 __ovld vloada_half3(size_t offset, const __private half *p);
 float4 __ovld vloada_half4(size_t offset, const __private half *p);
 float8 __ovld vloada_half8(size_t offset, const __private half *p);
 float16 __ovld vloada_half16(size_t offset, const __private half *p);
 #endif //defined(__opencl_c_generic_address_space)
 
 /**
  * The floatn value given by data is converted to
  * a halfn value using the appropriate rounding
  * mode.
  * For n = 1, 2, 4, 8 and 16, the halfn value is
  * written to the address computed as (p + (offset
  * * n)). The address computed as (p + (offset *
  * n)) must be aligned to sizeof (halfn) bytes.
  * For n = 3, the half3 value is written to the
  * address computed as (p + (offset * 4)). The
  * address computed as (p + (offset * 4)) must be
  * aligned to sizeof (half) * 4 bytes.
  * vstorea_halfn uses the current rounding
  * mode. The default current rounding mode is
  * round to nearest even.
  */
 #if defined(__opencl_c_generic_address_space)
-void __ovld vstorea_half(float data, size_t offset, half *p);
 void __ovld vstorea_half2(float2 data, size_t offset, half *p);
 void __ovld vstorea_half3(float3 data, size_t offset, half *p);
 void __ovld vstorea_half4(float4 data, size_t offset, half *p);
 void __ovld vstorea_half8(float8 data, size_t offset, half *p);
 void __ovld vstorea_half16(float16 data, size_t offset, half *p);
 
-void __ovld vstorea_half_rte(float data, size_t offset, half *p);
 void __ovld vstorea_half2_rte(float2 data, size_t offset, half *p);
 void __ovld vstorea_half3_rte(float3 data, size_t offset, half *p);
 void __ovld vstorea_half4_rte(float4 data, size_t offset, half *p);
 void __ovld vstorea_half8_rte(float8 data, size_t offset, half *p);
 void __ovld vstorea_half16_rte(float16 data, size_t offset, half *p);
 
-void __ovld vstorea_half_rtz(float data, size_t offset, half *p);
 void __ovld vstorea_half2_rtz(float2 data, size_t offset, half *p);
 void __ovld vstorea_half3_rtz(float3 data, size_t offset, half *p);
 void __ovld vstorea_half4_rtz(float4 data, size_t offset, half *p);
 void __ovld vstorea_half8_rtz(float8 data, size_t offset, half *p);
 void __ovld vstorea_half16_rtz(float16 data, size_t offset, half *p);
 
-void __ovld vstorea_half_rtp(float data, size_t offset, half *p);
 void __ovld vstorea_half2_rtp(float2 data, size_t offset, half *p);
 void __ovld vstorea_half3_rtp(float3 data, size_t offset, half *p);
 void __ovld vstorea_half4_rtp(float4 data, size_t offset, half *p);
 void __ovld vstorea_half8_rtp(float8 data, size_t offset, half *p);
 void __ovld vstorea_half16_rtp(float16 data, size_t offset, half *p);
 
-void __ovld vstorea_half_rtn(float data, size_t offset, half *p);
 void __ovld vstorea_half2_rtn(float2 data, size_t offset, half *p);
 void __ovld vstorea_half3_rtn(float3 data, size_t offset, half *p);
 void __ovld vstorea_half4_rtn(float4 data, size_t offset, half *p);
 void __ovld vstorea_half8_rtn(float8 data, size_t offset, half *p);
 void __ovld vstorea_half16_rtn(float16 data, size_t offset, half *p);
 
 #ifdef cl_khr_fp64
-void __ovld vstorea_half(double data, size_t offset, half *p);
 void __ovld vstorea_half2(double2 data, size_t offset, half *p);
 void __ovld vstorea_half3(double3 data, size_t offset, half *p);
 void __ovld vstorea_half4(double4 data, size_t offset, half *p);
 void __ovld vstorea_half8(double8 data, size_t offset, half *p);
 void __ovld vstorea_half16(double16 data, size_t offset, half *p);
 
-void __ovld vstorea_half_rte(double data, size_t offset, half *p);
 void __ovld vstorea_half2_rte(double2 data, size_t offset, half *p);
 void __ovld vstorea_half3_rte(double3 data, size_t offset, half *p);
 void __ovld vstorea_half4_rte(double4 data, size_t offset, half *p);
 void __ovld vstorea_half8_rte(double8 data, size_t offset, half *p);
 void __ovld vstorea_half16_rte(double16 data, size_t offset, half *p);
 
-void __ovld vstorea_half_rtz(double data, size_t offset, half *p);
 void __ovld vstorea_half2_rtz(double2 data, size_t offset, half *p);
 void __ovld vstorea_half3_rtz(double3 data, size_t offset, half *p);
 void __ovld vstorea_half4_rtz(double4 data, size_t offset, half *p);
 void __ovld vstorea_half8_rtz(double8 data, size_t offset, half *p);
 void __ovld vstorea_half16_rtz(double16 data, size_t offset, half *p);
 
-void __ovld vstorea_half_rtp(double data, size_t offset, half *p);
 void __ovld vstorea_half2_rtp(double2 data, size_t offset, half *p);
 void __ovld vstorea_half3_rtp(double3 data, size_t offset, half *p);
 void __ovld vstorea_half4_rtp(double4 data, size_t offset, half *p);
 void __ovld vstorea_half8_rtp(double8 data, size_t offset, half *p);
 void __ovld vstorea_half16_rtp(double16 data, size_t offset, half *p);
 
-void __ovld vstorea_half_rtn(double data, size_t offset, half *p);
 void __ovld vstorea_half2_rtn(double2 data, size_t offset, half *p);
 void __ovld vstorea_half3_rtn(double3 data, size_t offset, half *p);
 void __ovld vstorea_half4_rtn(double4 data, size_t offset, half *p);
 void __ovld vstorea_half8_rtn(double8 data, size_t offset, half *p);
 void __ovld vstorea_half16_rtn(double16 data, size_t offset, half *p);
 #endif //cl_khr_fp64
 
 #else
-void __ovld vstorea_half(float data, size_t offset, __global half *p);
 void __ovld vstorea_half2(float2 data, size_t offset, __global half *p);
 void __ovld vstorea_half3(float3 data, size_t offset, __global half *p);
 void __ovld vstorea_half4(float4 data, size_t offset, __global half *p);
 void __ovld vstorea_half8(float8 data, size_t offset, __global half *p);
 void __ovld vstorea_half16(float16 data, size_t offset, __global half *p);
 
-void __ovld vstorea_half_rte(float data, size_t offset, __global half *p);
 void __ovld vstorea_half2_rte(float2 data, size_t offset, __global half *p);
 void __ovld vstorea_half3_rte(float3 data, size_t offset, __global half *p);
 void __ovld vstorea_half4_rte(float4 data, size_t offset, __global half *p);
 void __ovld vstorea_half8_rte(float8 data, size_t offset, __global half *p);
 void __ovld vstorea_half16_rte(float16 data, size_t offset, __global half *p);
 
-void __ovld vstorea_half_rtz(float data, size_t offset, __global half *p);
 void __ovld vstorea_half2_rtz(float2 data, size_t offset, __global half *p);
 void __ovld vstorea_half3_rtz(float3 data, size_t offset, __global half *p);
 void __ovld vstorea_half4_rtz(float4 data, size_t offset, __global half *p);
 void __ovld vstorea_half8_rtz(float8 data, size_t offset, __global half *p);
 void __ovld vstorea_half16_rtz(float16 data, size_t offset, __global half *p);
 
-void __ovld vstorea_half_rtp(float data, size_t offset, __global half *p);
 void __ovld vstorea_half2_rtp(float2 data, size_t offset, __global half *p);
 void __ovld vstorea_half3_rtp(float3 data, size_t offset, __global half *p);
 void __ovld vstorea_half4_rtp(float4 data, size_t offset, __global half *p);
 void __ovld vstorea_half8_rtp(float8 data, size_t offset, __global half *p);
 void __ovld vstorea_half16_rtp(float16 data, size_t offset, __global half *p);
 
-void __ovld vstorea_half_rtn(float data, size_t offset, __global half *p);
 void __ovld vstorea_half2_rtn(float2 data, size_t offset, __global half *p);
 void __ovld vstorea_half3_rtn(float3 data, size_t offset, __global half *p);
 void __ovld vstorea_half4_rtn(float4 data, size_t offset, __global half *p);
 void __ovld vstorea_half8_rtn(float8 data, size_t offset, __global half *p);
 void __ovld vstorea_half16_rtn(float16 data, size_t offset, __global half *p);
 
-void __ovld vstorea_half(float data, size_t offset, __local half *p);
 void __ovld vstorea_half2(float2 data, size_t offset, __local half *p);
 void __ovld vstorea_half3(float3 data, size_t offset, __local half *p);
 void __ovld vstorea_half4(float4 data, size_t offset, __local half *p);
 void __ovld vstorea_half8(float8 data, size_t offset, __local half *p);
 void __ovld vstorea_half16(float16 data, size_t offset, __local half *p);
 
-void __ovld vstorea_half_rte(float data, size_t offset, __local half *p);
 void __ovld vstorea_half2_rte(float2 data, size_t offset, __local half *p);
 void __ovld vstorea_half3_rte(float3 data, size_t offset, __local half *p);
 void __ovld vstorea_half4_rte(float4 data, size_t offset, __local half *p);
 void __ovld vstorea_half8_rte(float8 data, size_t offset, __local half *p);
 void __ovld vstorea_half16_rte(float16 data, size_t offset, __local half *p);
 
-void __ovld vstorea_half_rtz(float data, size_t offset, __local half *p);
 void __ovld vstorea_half2_rtz(float2 data, size_t offset, __local half *p);
 void __ovld vstorea_half3_rtz(float3 data, size_t offset, __local half *p);
 void __ovld vstorea_half4_rtz(float4 data, size_t offset, __local half *p);
 void __ovld vstorea_half8_rtz(float8 data, size_t offset, __local half *p);
 void __ovld vstorea_half16_rtz(float16 data, size_t offset, __local half *p);
 
-void __ovld vstorea_half_rtp(float data, size_t offset, __local half *p);
 void __ovld vstorea_half2_rtp(float2 data, size_t offset, __local half *p);
 void __ovld vstorea_half3_rtp(float3 data, size_t offset, __local half *p);
 void __ovld vstorea_half4_rtp(float4 data, size_t offset, __local half *p);
 void __ovld vstorea_half8_rtp(float8 data, size_t offset, __local half *p);
 void __ovld vstorea_half16_rtp(float16 data, size_t offset, __local half *p);
 
-void __ovld vstorea_half_rtn(float data, size_t offset, __local half *p);
 void __ovld vstorea_half2_rtn(float2 data, size_t offset, __local half *p);
 void __ovld vstorea_half3_rtn(float3 data, size_t offset, __local half *p);
 void __ovld vstorea_half4_rtn(float4 data, size_t offset, __local half *p);
 void __ovld vstorea_half8_rtn(float8 data, size_t offset, __local half *p);
 void __ovld vstorea_half16_rtn(float16 data, size_t offset, __local half *p);
 
-void __ovld vstorea_half(float data, size_t offset, __private half *p);
 void __ovld vstorea_half2(float2 data, size_t offset, __private half *p);
 void __ovld vstorea_half3(float3 data, size_t offset, __private half *p);
 void __ovld vstorea_half4(float4 data, size_t offset, __private half *p);
 void __ovld vstorea_half8(float8 data, size_t offset, __private half *p);
 void __ovld vstorea_half16(float16 data, size_t offset, __private half *p);
 
-void __ovld vstorea_half_rte(float data, size_t offset, __private half *p);
 void __ovld vstorea_half2_rte(float2 data, size_t offset, __private half *p);
 void __ovld vstorea_half3_rte(float3 data, size_t offset, __private half *p);
 void __ovld vstorea_half4_rte(float4 data, size_t offset, __private half *p);
 void __ovld vstorea_half8_rte(float8 data, size_t offset, __private half *p);
 void __ovld vstorea_half16_rte(float16 data, size_t offset, __private half *p);
 
-void __ovld vstorea_half_rtz(float data, size_t offset, __private half *p);
 void __ovld vstorea_half2_rtz(float2 data, size_t offset, __private half *p);
 void __ovld vstorea_half3_rtz(float3 data, size_t offset, __private half *p);
 void __ovld vstorea_half4_rtz(float4 data, size_t offset, __private half *p);
 void __ovld vstorea_half8_rtz(float8 data, size_t offset, __private half *p);
 void __ovld vstorea_half16_rtz(float16 data, size_t offset, __private half *p);
 
-void __ovld vstorea_half_rtp(float data, size_t offset, __private half *p);
 void __ovld vstorea_half2_rtp(float2 data, size_t offset, __private half *p);
 void __ovld vstorea_half3_rtp(float3 data, size_t offset, __private half *p);
 void __ovld vstorea_half4_rtp(float4 data, size_t offset, __private half *p);
 void __ovld vstorea_half8_rtp(float8 data, size_t offset, __private half *p);
 void __ovld vstorea_half16_rtp(float16 data, size_t offset, __private half *p);
 
-void __ovld vstorea_half_rtn(float data, size_t offset, __private half *p);
 void __ovld vstorea_half2_rtn(float2 data, size_t offset, __private half *p);
 void __ovld vstorea_half3_rtn(float3 data, size_t offset, __private half *p);
 void __ovld vstorea_half4_rtn(float4 data, size_t offset, __private half *p);
 void __ovld vstorea_half8_rtn(float8 data, size_t offset, __private half *p);
 void __ovld vstorea_half16_rtn(float16 data, size_t offset, __private half *p);
 
 #ifdef cl_khr_fp64
-void __ovld vstorea_half(double data, size_t offset, __global half *p);
 void __ovld vstorea_half2(double2 data, size_t offset, __global half *p);
 void __ovld vstorea_half3(double3 data, size_t offset, __global half *p);
 void __ovld vstorea_half4(double4 data, size_t offset, __global half *p);
 void __ovld vstorea_half8(double8 data, size_t offset, __global half *p);
 void __ovld vstorea_half16(double16 data, size_t offset, __global half *p);
 
-void __ovld vstorea_half_rte(double data, size_t offset, __global half *p);
 void __ovld vstorea_half2_rte(double2 data, size_t offset, __global half *p);
 void __ovld vstorea_half3_rte(double3 data, size_t offset, __global half *p);
 void __ovld vstorea_half4_rte(double4 data, size_t offset, __global half *p);
 void __ovld vstorea_half8_rte(double8 data, size_t offset, __global half *p);
 void __ovld vstorea_half16_rte(double16 data, size_t offset, __global half *p);
 
-void __ovld vstorea_half_rtz(double data, size_t offset, __global half *p);
 void __ovld vstorea_half2_rtz(double2 data, size_t offset, __global half *p);
 void __ovld vstorea_half3_rtz(double3 data, size_t offset, __global half *p);
 void __ovld vstorea_half4_rtz(double4 data, size_t offset, __global half *p);
 void __ovld vstorea_half8_rtz(double8 data, size_t offset, __global half *p);
 void __ovld vstorea_half16_rtz(double16 data, size_t offset, __global half *p);
 
-void __ovld vstorea_half_rtp(double data, size_t offset, __global half *p);
 void __ovld vstorea_half2_rtp(double2 data, size_t offset, __global half *p);
 void __ovld vstorea_half3_rtp(double3 data, size_t offset, __global half *p);
 void __ovld vstorea_half4_rtp(double4 data, size_t offset, __global half *p);
 void __ovld vstorea_half8_rtp(double8 data, size_t offset, __global half *p);
 void __ovld vstorea_half16_rtp(double16 data, size_t offset, __global half *p);
 
-void __ovld vstorea_half_rtn(double data, size_t offset, __global half *p);
 void __ovld vstorea_half2_rtn(double2 data, size_t offset, __global half *p);
 void __ovld vstorea_half3_rtn(double3 data, size_t offset, __global half *p);
 void __ovld vstorea_half4_rtn(double4 data, size_t offset, __global half *p);
 void __ovld vstorea_half8_rtn(double8 data, size_t offset, __global half *p);
 void __ovld vstorea_half16_rtn(double16 data, size_t offset, __global half *p);
 
-void __ovld vstorea_half(double data, size_t offset, __local half *p);
 void __ovld vstorea_half2(double2 data, size_t offset, __local half *p);
 void __ovld vstorea_half3(double3 data, size_t offset, __local half *p);
 void __ovld vstorea_half4(double4 data, size_t offset, __local half *p);
 void __ovld vstorea_half8(double8 data, size_t offset, __local half *p);
 void __ovld vstorea_half16(double16 data, size_t offset, __local half *p);
 
-void __ovld vstorea_half_rte(double data, size_t offset, __local half *p);
 void __ovld vstorea_half2_rte(double2 data, size_t offset, __local half *p);
 void __ovld vstorea_half3_rte(double3 data, size_t offset, __local half *p);
 void __ovld vstorea_half4_rte(double4 data, size_t offset, __local half *p);
 void __ovld vstorea_half8_rte(double8 data, size_t offset, __local half *p);
 void __ovld vstorea_half16_rte(double16 data, size_t offset, __local half *p);
 
-void __ovld vstorea_half_rtz(double data, size_t offset, __local half *p);
 void __ovld vstorea_half2_rtz(double2 data, size_t offset, __local half *p);
 void __ovld vstorea_half3_rtz(double3 data, size_t offset, __local half *p);
 void __ovld vstorea_half4_rtz(double4 data, size_t offset, __local half *p);
 void __ovld vstorea_half8_rtz(double8 data, size_t offset, __local half *p);
 void __ovld vstorea_half16_rtz(double16 data, size_t offset, __local half *p);
 
-void __ovld vstorea_half_rtp(double data, size_t offset, __local half *p);
 void __ovld vstorea_half2_rtp(double2 data, size_t offset, __local half *p);
 void __ovld vstorea_half3_rtp(double3 data, size_t offset, __local half *p);
 void __ovld vstorea_half4_rtp(double4 data, size_t offset, __local half *p);
 void __ovld vstorea_half8_rtp(double8 data, size_t offset, __local half *p);
 void __ovld vstorea_half16_rtp(double16 data, size_t offset, __local half *p);
 
-void __ovld vstorea_half_rtn(double data, size_t offset, __local half *p);
 void __ovld vstorea_half2_rtn(double2 data, size_t offset, __local half *p);
 void __ovld vstorea_half3_rtn(double3 data, size_t offset, __local half *p);
 void __ovld vstorea_half4_rtn(double4 data, size_t offset, __local half *p);
 void __ovld vstorea_half8_rtn(double8 data, size_t offset, __local half *p);
 void __ovld vstorea_half16_rtn(double16 data, size_t offset, __local half *p);
 
-void __ovld vstorea_half(double data, size_t offset, __private half *p);
 void __ovld vstorea_half2(double2 data, size_t offset, __private half *p);
 void __ovld vstorea_half3(double3 data, size_t offset, __private half *p);
 void __ovld vstorea_half4(double4 data, size_t offset, __private half *p);
 void __ovld vstorea_half8(double8 data, size_t offset, __private half *p);
 void __ovld vstorea_half16(double16 data, size_t offset, __private half *p);
 
-void __ovld vstorea_half_rte(double data, size_t offset, __private half *p);
 void __ovld vstorea_half2_rte(double2 data, size_t offset, __private half *p);
 void __ovld vstorea_half3_rte(double3 data, size_t offset, __private half *p);
 void __ovld vstorea_half4_rte(double4 data, size_t offset, __private half *p);
 void __ovld vstorea_half8_rte(double8 data, size_t offset, __private half *p);
 void __ovld vstorea_half16_rte(double16 data, size_t offset, __private half *p);
 
-void __ovld vstorea_half_rtz(double data, size_t offset, __private half *p);
 void __ovld vstorea_half2_rtz(double2 data, size_t offset, __private half *p);
 void __ovld vstorea_half3_rtz(double3 data, size_t offset, __private half *p);
 void __ovld vstorea_half4_rtz(double4 data, size_t offset, __private half *p);
 void __ovld vstorea_half8_rtz(double8 data, size_t offset, __private half *p);
 void __ovld vstorea_half16_rtz(double16 data, size_t offset, __private half *p);
 
-void __ovld vstorea_half_rtp(double data, size_t offset, __private half *p);
 void __ovld vstorea_half2_rtp(double2 data, size_t offset, __private half *p);
 void __ovld vstorea_half3_rtp(double3 data, size_t offset, __private half *p);
 void __ovld vstorea_half4_rtp(double4 data, size_t offset, __private half *p);
 void __ovld vstorea_half8_rtp(double8 data, size_t offset, __private half *p);
 void __ovld vstorea_half16_rtp(double16 data, size_t offset, __private half *p);
 
-void __ovld vstorea_half_rtn(double data, size_t offset, __private half *p);
 void __ovld vstorea_half2_rtn(double2 data,size_t offset, __private half *p);
 void __ovld vstorea_half3_rtn(double3 data,size_t offset, __private half *p);
 void __ovld vstorea_half4_rtn(double4 data,size_t offset, __private half *p);
 void __ovld vstorea_half8_rtn(double8 data,size_t offset, __private half *p);
 void __ovld vstorea_half16_rtn(double16 data,size_t offset, __private half *p);
 #endif //cl_khr_fp64
 #endif //defined(__opencl_c_generic_address_space)
 

clang/lib/Sema/OpenCLBuiltins.td

let MinVersion = CL20 in {
   defm : VloadVstore<[GenericAS], 1>;
 }
 // vload with constant address space is available regardless of version.
 defm : VloadVstore<[ConstantAS], 0>;
 
 multiclass VloadVstoreHalf<list<AddressSpace> addrspaces, bit defStores> {
   foreach AS = addrspaces in {
     def : Builtin<"vload_half", [Float, Size, PointerType<ConstType<Half>, AS>]>;
-    def : Builtin<"vloada_half", [Float, Size, PointerType<ConstType<Half>, AS>]>;
     foreach VSize = [2, 3, 4, 8, 16] in {
       foreach name = ["vload_half" # VSize] in {
         def : Builtin<name, [VectorType<Float, VSize>, Size, PointerType<ConstType<Half>, AS>]>;
       }
     }
     if defStores then {
       foreach rnd = ["", "_rte", "_rtz", "_rtp", "_rtn"] in {
-        foreach name = ["vstore_half" # rnd, "vstorea_half" # rnd] in {
+        foreach name = ["vstore_half" # rnd] in {
           def : Builtin<name, [Void, Float, Size, PointerType<Half, AS>]>;
           def : Builtin<name, [Void, Double, Size, PointerType<Half, AS>]>;
         }
         foreach VSize = [2, 3, 4, 8, 16] in {
           foreach name = ["vstore_half" # VSize # rnd] in {
             def : Builtin<name, [Void, VectorType<Float, VSize>, Size, PointerType<Half, AS>]>;
             def : Builtin<name, [Void, VectorType<Double, VSize>, Size, PointerType<Half, AS>]>;
           }