diff options
| author | Pierre Blanchard <pierre.blanchard@arm.com> | 2020-11-05 16:26:36 +0000 |
|---|---|---|
| committer | Szabolcs Nagy <szabolcs.nagy@arm.com> | 2020-11-13 14:43:49 +0000 |
| commit | 00e5afdbaf600cb8d56072685e987800d8678248 (patch) | |
| tree | 312326613eb7e581fdf0e17766d48cf01d40da65 /math/test/runulp.sh | |
| parent | fc3fcc91ba7d79b30876d8e7ee68b8442a50da5c (diff) | |
| download | arm-optimized-routines-00e5afdbaf600cb8d56072685e987800d8678248.tar.gz | |
math: add scalar erf
Only tested in round-to-nearest mode. The expected worst case error
is 1.01 ULP near x=1.25. Benchmarked over random x in [-6,6] and
can increase performance by > 2x (> 3.5x for throughput) on big ooo
cores compared to the implementation in glibc 2.28.
Includes data for erfc too, but this patch only adds erf.
Diffstat (limited to 'math/test/runulp.sh')
| -rwxr-xr-x | math/test/runulp.sh | 8 |
1 files changed, 8 insertions, 0 deletions
diff --git a/math/test/runulp.sh b/math/test/runulp.sh index 283145e..492a6f2 100755 --- a/math/test/runulp.sh +++ b/math/test/runulp.sh @@ -72,6 +72,14 @@ t pow 0x1.ffffffffffff0p-1 0x1.0000000000008p0 x 0x1p60 0x1p68 50000 t pow 0x1.ffffffffff000p-1 0x1p0 x 0x1p50 0x1p52 50000 t pow -0x1.ffffffffff000p-1 -0x1p0 x 0x1p50 0x1p52 50000 +L=1.0 +t erf 0 0xffff000000000000 10000 +t erf 0x1p-1022 0x1p-26 40000 +t erf -0x1p-1022 -0x1p-26 40000 +t erf 0x1p-26 0x1p3 40000 +t erf -0x1p-26 -0x1p3 40000 +t erf 0 inf 40000 + L=0.01 t expf 0 0xffff0000 10000 t expf 0x1p-14 0x1p8 50000 |