Am 01.08.2018 um 04:54 schrieb Marek Olšák:
> From: Marek Olšák <marek.ol...@amd.com>
>
> This fixes the build if __builtin_clz is unsupported.
> ---
> src/util/half_float.c | 12 +++++++++++-
> 1 file changed, 11 insertions(+), 1 deletion(-)
>
> diff --git a/src/util/half_float.c b/src/util/half_float.c
> index 588f050d924..2eff2c84f51 100644
> --- a/src/util/half_float.c
> +++ b/src/util/half_float.c
> @@ -208,21 +208,31 @@ uint8_t _mesa_half_to_unorm8(uint16_t val)
> * Takes a uint16_t, divides by 65536, converts the infinite-precision
> * result to fp16 with round-to-zero. Used by the ASTC decoder.
> */
> uint16_t _mesa_uint16_div_64k_to_half(uint16_t v)
> {
> /* Zero or subnormal. Set the mantissa to (v << 8) and return. */
> if (v < 4)
> return v << 8;
>
> /* Count the leading 0s in the uint16_t */
> - int n = __builtin_clz(v) - (sizeof(unsigned int) - sizeof(uint16_t)) * 8;
> +#ifdef HAVE___BUILTIN_CLZ
> + int n = __builtin_clz(v) - 16;
> +#else
> + int n = 16;
> + for (int i = 15; i >= 0; i--) {
> + if (v & (1 << i)) {
> + n = 15 - i;
> + break;
> + }
> + }
> +#endif
Not sure why you're not just using util_last_bit (I think it's best to
keep such bit scan hackery in separate util files), but whatever fixes
the compile error, so
Reviewed-by: Roland Scheidegger <srol...@vmware.com>
>
> /* Shift the mantissa up so bit 16 is the hidden 1 bit,
> * mask it off, then shift back down to 10 bits
> */
> int m = ( ((uint32_t)v << (n + 1)) & 0xffff ) >> 6;
>
> /* (0{n} 1 X{15-n}) * 2^-16
> * = 1.X * 2^(15-n-16)
> * = 1.X * 2^(14-n - 15)
> * which is the FP16 form with e = 14 - n
>
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
