[Libclc-dev] [PATCH 1/1] select: simplify implementation and fix fp16

[Jeroen Ketema via Libclc-dev]

LGTM!

Thanks.

Jeroen

> On 6 Apr 2018, at 20:38, Jan Vesely <jan.vesely at rutgers.edu> wrote:
> 
> Fix half precision implementation
> Vector ?: operator should behave exactly as select
> Passes CTS on carrizo
> 
> Signed-off-by: Jan Vesely <jan.vesely at rutgers.edu>
> ---
> the select test segfaults at the end, but I don't think that's related
> to libclc.
> 
> generic/lib/relational/select.inc | 22 ++++------------------
> 1 file changed, 4 insertions(+), 18 deletions(-)
> 
> diff --git a/generic/lib/relational/select.inc b/generic/lib/relational/select.inc
> index a9790d1..7691af2 100644
> --- a/generic/lib/relational/select.inc
> +++ b/generic/lib/relational/select.inc
> @@ -9,39 +9,25 @@
> #define __CLC_S_GENTYPE __CLC_XCONCAT(int, __CLC_VECSIZE)
> #define __CLC_U_GENTYPE __CLC_XCONCAT(uint, __CLC_VECSIZE)
> #elif __CLC_FPSIZE == 16
> -#define __CLC_S_GENTYPE __CLC_XCONCAT(char, __CLC_VECSIZE)
> -#define __CLC_U_GENTYPE __CLC_XCONCAT(uchar, __CLC_VECSIZE)
> -#endif
> -#ifdef __CLC_FPSIZE
> -#define __CLC_GENSIZE   __CLC_FPSIZE
> +#define __CLC_S_GENTYPE __CLC_XCONCAT(short, __CLC_VECSIZE)
> +#define __CLC_U_GENTYPE __CLC_XCONCAT(ushort, __CLC_VECSIZE)
> #endif
> 
> -#define __CLC_AS_S_GENTYPE __CLC_XCONCAT(as_, __CLC_S_GENTYPE)
> -#define __CLC_AS_GENTYPE __CLC_XCONCAT(as_, __CLC_GENTYPE)
> -
> _CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE select(__CLC_GENTYPE x, __CLC_GENTYPE y, __CLC_S_GENTYPE z)
> {
> -#ifdef __CLC_SCALAR
> 	return z ? y : x;
> -#else
> -	__CLC_S_GENTYPE bitmask = z >> (__CLC_GENSIZE - 1);
> -	return __CLC_AS_GENTYPE(bitselect(__CLC_AS_S_GENTYPE(x), __CLC_AS_S_GENTYPE(y), bitmask));
> -#endif
> }
> 
> _CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE select(__CLC_GENTYPE x, __CLC_GENTYPE y, __CLC_U_GENTYPE z)
> {
> -	return select(x, y, __CLC_AS_S_GENTYPE(z));
> +	return z ? y : x;
> }
> 
> -#undef __CLC_AS_S_GENTYPE
> -#undef __CLC_AS_GENTYPE
> -
> #ifdef __CLC_FPSIZE
> #undef __CLC_S_GENTYPE
> #undef __CLC_U_GENTYPE
> -#undef __CLC_GENSIZE
> #endif
> +
> #ifdef __CLC_SCALAR
> #undef __CLC_VECSIZE
> #endif
> -- 
> 2.14.3
>