[PATCH 3/5] iee754: prvoide gcc builtins based generic fma functions

[Adhemerval Zanella]

On 29/05/2020 23:00, Vineet Gupta wrote:

Ok with the fixes below.

Reviewed-by: Adhemerval Zanella  <adhemerval.zanella@linaro.org>

> ---
>  sysdeps/generic/math-use-builtins.h         | 4 ++++
>  sysdeps/ieee754/dbl-64/s_fma.c              | 6 ++++++
>  sysdeps/ieee754/dbl-64/s_fmaf.c             | 6 ++++++
>  sysdeps/ieee754/float128/float128_private.h | 2 ++
>  sysdeps/ieee754/ldbl-128/s_fmal.c           | 5 +++++
>  5 files changed, 23 insertions(+)
> 
> diff --git a/sysdeps/generic/math-use-builtins.h b/sysdeps/generic/math-use-builtins.h
> index fc724c824a17..9e96807a3370 100644
> --- a/sysdeps/generic/math-use-builtins.h
> +++ b/sysdeps/generic/math-use-builtins.h
> @@ -63,4 +63,8 @@
>  #define USE_SQRT_BUILTIN 0
>  #define USE_SQRTF_BUILTIN 0
>  
> +#define USE_FMA_BUILTIN 0
> +#define USE_FMAF_BUILTIN 0
> +#define USE_FMAL_BUILTIN 0
> +
>  #endif /* math-use-builtins.h */

For float128 support it should also contain a:

  #define USE_FMAF128_BUILTIN

> diff --git a/sysdeps/ieee754/dbl-64/s_fma.c b/sysdeps/ieee754/dbl-64/s_fma.c
> index 876df6e78bdc..1e4b2da1511d 100644
> --- a/sysdeps/ieee754/dbl-64/s_fma.c
> +++ b/sysdeps/ieee754/dbl-64/s_fma.c
> @@ -25,6 +25,7 @@
>  #include <fenv_private.h>
>  #include <libm-alias-double.h>
>  #include <tininess.h>
> +#include <math-use-builtins.h>
>  
>  /* This implementation uses rounding to odd to avoid problems with
>     double rounding.  See a paper by Boldo and Melquiond:
> @@ -33,6 +34,10 @@
>  double
>  __fma (double x, double y, double z)
>  {
> +#if USE_FMA_BUILTIN
> +  return __builtin_fma (x);

It should be:

  return __builtin_fma (x, y, z);

Same for float and long double variant.

> +#else
> +  /* Use generic implementation.  */
>    union ieee754_double u, v, w;
>    int adjust = 0;
>    u.d = x;
> @@ -292,6 +297,7 @@ __fma (double x, double y, double z)
>        v.ieee.mantissa1 |= j;
>        return v.d * 0x1p-108;
>      }
> +#endif /* ! USE_FMA_BUILTIN  */
>  }
>  #ifndef __fma
>  libm_alias_double (__fma, fma)> diff --git a/sysdeps/ieee754/dbl-64/s_fmaf.c b/sysdeps/ieee754/dbl-64/s_fmaf.c
> index 57329d0a87fe..f15b18262124 100644
> --- a/sysdeps/ieee754/dbl-64/s_fmaf.c
> +++ b/sysdeps/ieee754/dbl-64/s_fmaf.c
> @@ -23,6 +23,7 @@
>  #include <math-barriers.h>
>  #include <fenv_private.h>
>  #include <libm-alias-float.h>
> +#include <math-use-builtins.h>
>  
>  /* This implementation relies on double being more than twice as
>     precise as float and uses rounding to odd in order to avoid problems
> @@ -33,6 +34,10 @@
>  float
>  __fmaf (float x, float y, float z)
>  {
> +#if USE_FMAF_BUILTIN
> +  return __builtin_fmaf (x);
> +#else
> +  /* Use generic implementation.  */
>    fenv_t env;
>  
>    /* Multiplication is always exact.  */
> @@ -60,6 +65,7 @@ __fmaf (float x, float y, float z)
>  
>    /* And finally truncation with round to nearest.  */
>    return (float) u.d;
> +#endif /* ! USE_FMAF_BUILTIN  */
>  }
>  #ifndef __fmaf
>  libm_alias_float (__fma, fma)> diff --git a/sysdeps/ieee754/float128/float128_private.h b/sysdeps/ieee754/float128/float128_private.h
> index f97463d9dc1b..a697a7c29038 100644
> --- a/sysdeps/ieee754/float128/float128_private.h
> +++ b/sysdeps/ieee754/float128/float128_private.h
> @@ -154,6 +154,8 @@
>  #define USE_ROUNDL_BUILTIN USE_ROUNDF128_BUILTIN
>  #undef USE_COPYSIGNL_BUILTIN
>  #define USE_COPYSIGNL_BUILTIN USE_COPYSIGNF128_BUILTIN
> +#undef USE_FMAL_BUILTIN
> +#define USE_FMAL_BUILTIN USE_FMA128_BUILTIN
>  
>  /* IEEE function renames.  */
>  #define __ieee754_acoshl __ieee754_acoshf128

Ok.

> diff --git a/sysdeps/ieee754/ldbl-128/s_fmal.c b/sysdeps/ieee754/ldbl-128/s_fmal.c
> index 7475015bcec6..1403734a5aeb 100644
> --- a/sysdeps/ieee754/ldbl-128/s_fmal.c
> +++ b/sysdeps/ieee754/ldbl-128/s_fmal.c
> @@ -25,6 +25,7 @@
>  #include <math_private.h>
>  #include <libm-alias-ldouble.h>
>  #include <tininess.h>
> +#include <math-use-builtins.h>
>  
>  /* This implementation uses rounding to odd to avoid problems with
>     double rounding.  See a paper by Boldo and Melquiond:
> @@ -33,6 +34,9 @@
>  _Float128
>  __fmal (_Float128 x, _Float128 y, _Float128 z)
>  {
> +#if USE_FMAL_BUILTIN
> +  return __builtin_fmal (x);
> +#else
>    union ieee854_long_double u, v, w;
>    int adjust = 0;
>    u.d = x;
> @@ -296,5 +300,6 @@ __fmal (_Float128 x, _Float128 y, _Float128 z)
>        v.ieee.mantissa3 |= j;
>        return v.d * L(0x1p-228);
>      }
> +#endif /* ! USE_FMAL_BUILTIN  */
>  }
>  libm_alias_ldouble (__fma, fma)
>