Commit Message
2017-06-09 Paul A. Clarke <pc@us.ibm.com>
* benchtests/Makefile (bench-math): Add powf.
* benchtests/powf-inputs: Copy from benchtests/pow-inputs.
---
benchtests/Makefile | 2 +-
benchtests/{pow-inputs => powf-inputs} | 4 ++--
2 files changed, 3 insertions(+), 3 deletions(-)
copy benchtests/{pow-inputs => powf-inputs} (99%)
Comments
On Fri, 9 Jun 2017, Paul Clarke wrote:
> diff --git a/benchtests/pow-inputs b/benchtests/powf-inputs
> similarity index 99%
> copy from benchtests/pow-inputs
> copy to benchtests/powf-inputs
A large proportion of the pow inputs are sufficiently large or small that
they would round to 0 or infinity as float values, or the result would
round to 0, infinity or 1 as a float value. I don't think such special
cases should be considered typical and so I don't think copying the pow
inputs is a good way of producing a set of benchmark inputs to powf.
@@ -23,7 +23,7 @@ subdir := benchtests
include ../Makeconfig
bench-math := acos acosh asin asinh atan atanh cos cosh exp exp2 log log2 \
modf pow rint sin sincos sinh sqrt tan tanh fmin fmax fminf \
- fmaxf
+ fmaxf powf
bench-pthread := pthread_once
similarity index 99%
copy from benchtests/pow-inputs
copy to benchtests/powf-inputs
@@ -1,5 +1,5 @@
-## args: double:double
-## ret: double
+## args: float:float
+## ret: float
## includes: math.h
42.0, 42.0
-0x1.086f771a2b37bp696, 0x1.592b453fe869fp-81