Emilio G. Cota writes:
> On Thu, Dec 20, 2018 at 11:10:08 +, Alex Bennée wrote:
> (snip)
>> +#if defined(__GLIBC__) && (__GLIBC__ == 2 && __GLIBC_MINOR__ <= 12)
>> +#define QEMU_HARDFLOAT_USE_FMA 0
>> +#else
>> +#define QEMU_HARDFLOAT_USE_FMA 1
>> +#endif
>> +#else
>> +#define QEMU_HARDFLOA
Richard Henderson writes:
> On 12/21/18 11:30 AM, Emilio G. Cota wrote:
>> +ua.s = 0x0021;
>> +ub.s = 0x3ca0;
>> +uc.s = 0x0020;
>> +ur.h = fma(ua.h, ub.h, uc.h);
>> +if (ur.s != 0x0021) {
>
> Forgot your ull's, but otherwise o
On Dec 20, 2018 12:11 PM, "Alex Bennée" wrote:
>
> Some versions of glibc have been reported to have problems with
> fused-multiply-accumulate operations. If the underlying fma
> implementation does a two step operation it will instroduce subtle
> rounding errors. Newer versions of the library see
On 12/21/18 11:30 AM, Emilio G. Cota wrote:
> +ua.s = 0x0021;
> +ub.s = 0x3ca0;
> +uc.s = 0x0020;
> +ur.h = fma(ua.h, ub.h, uc.h);
> +if (ur.s != 0x0021) {
Forgot your ull's, but otherwise ok.
In email to Alex, I did wonder if we
On Thu, Dec 20, 2018 at 11:10:08 +, Alex Bennée wrote:
(snip)
> +#if defined(__GLIBC__) && (__GLIBC__ == 2 && __GLIBC_MINOR__ <= 12)
> +#define QEMU_HARDFLOAT_USE_FMA 0
> +#else
> +#define QEMU_HARDFLOAT_USE_FMA 1
> +#endif
> +#else
> +#define QEMU_HARDFLOAT_USE_FMA 1
> +#endif
> +
> /*
> *
Hi,
On Thu, Dec 20, 2018 at 12:10 PM Alex Bennée wrote:
>
> Some versions of glibc have been reported to have problems with
> fused-multiply-accumulate operations. If the underlying fma
> implementation does a two step operation it will instroduce subtle
> rounding errors. Newer versions of the l
On Dec 20, 2018 12:11 PM, "Alex Bennée" wrote:
>
> Some versions of glibc have been reported to have problems with
> fused-multiply-accumulate operations. If the underlying fma
> implementation does a two step operation it will instroduce subtle
> rounding errors. Newer versions of the library see
Some versions of glibc have been reported to have problems with
fused-multiply-accumulate operations. If the underlying fma
implementation does a two step operation it will instroduce subtle
rounding errors. Newer versions of the library seem to deal with this
better and modern hardware has fused o
