powerpc: Avoid putting floating point values in memory
Commit Message
2017-09-22 Tulio Magno Quites Machado Filho <tuliom@linux.vnet.ibm.com>
* sysdeps/powerpc/fpu/math_private.h (math_opt_barrier):
(math_force_eval): Add powerpc version.
---
sysdeps/powerpc/fpu/math_private.h | 7 +++++++
1 file changed, 7 insertions(+)
Comments
Missing [BZ #22189] ChangeLog marker.
On 22/09/2017 15:19, Tulio Magno Quites Machado Filho wrote:
> 2017-09-22 Tulio Magno Quites Machado Filho <tuliom@linux.vnet.ibm.com>
>
> * sysdeps/powerpc/fpu/math_private.h (math_opt_barrier):
> (math_force_eval): Add powerpc version.
> ---
> sysdeps/powerpc/fpu/math_private.h | 7 +++++++
> 1 file changed, 7 insertions(+)
>
> diff --git a/sysdeps/powerpc/fpu/math_private.h b/sysdeps/powerpc/fpu/math_private.h
> index 396fd05..472182d 100644
> --- a/sysdeps/powerpc/fpu/math_private.h
> +++ b/sysdeps/powerpc/fpu/math_private.h
> @@ -23,6 +23,13 @@
> #include <ldsodefs.h>
> #include <dl-procinfo.h>
> #include <fenv_private.h>
> +
> +/* Avoid putting floating point values in memory. */
> +# define math_opt_barrier(x) \
> + ({ __typeof (x) __x = (x); __asm ("" : "+dwa" (__x)); __x; })
> +# define math_force_eval(x) \
> + ({ __typeof (x) __x = (x); __asm __volatile__ ("" : : "dwa" (__x)); })
> +
> #include_next <math_private.h>
>
> #if defined _ARCH_PWR9 && __HAVE_DISTINCT_FLOAT128
>
LGTM, it seems to work on older GCC as well (4.8).
Joseph Myers <joseph@codesourcery.com> writes:
> Missing [BZ #22189] ChangeLog marker.
I wasn't aware of that bug yet.
I'm adding it.
Thanks!
@@ -23,6 +23,13 @@
#include <ldsodefs.h>
#include <dl-procinfo.h>
#include <fenv_private.h>
+
+/* Avoid putting floating point values in memory. */
+# define math_opt_barrier(x) \
+ ({ __typeof (x) __x = (x); __asm ("" : "+dwa" (__x)); __x; })
+# define math_force_eval(x) \
+ ({ __typeof (x) __x = (x); __asm __volatile__ ("" : : "dwa" (__x)); })
+
#include_next <math_private.h>
#if defined _ARCH_PWR9 && __HAVE_DISTINCT_FLOAT128