The port is straight from amd-builtins, analogous to erfc. Just like in sinpi, the 32-bit float variant had optional code relying on subnormals which was removed during porting.
Hope this can be merged.
Regards,
Vedran
libclc-add-erf.patch (16.6 KB)
Trying without attaching...
The port is straight from amd-builtins, analogous to erfc. Just like in
sinpi, the 32-bit float variant had optional code relying on subnormals
which was removed during porting.
diff --git a/generic/include/clc/clc.h b/generic/include/clc/clc.h
index b106923..f53fb7d 100644
--- a/generic/include/clc/clc.h
+++ b/generic/include/clc/clc.h
@@ -47,6 +47,7 @@
#include <clc/math/cos.h>
#include <clc/math/cospi.h>
#include <clc/math/ceil.h>
+#include <clc/math/erf.h>
#include <clc/math/erfc.h>
#include <clc/math/exp.h>
#include <clc/math/exp10.h>
diff --git a/generic/include/clc/math/erf.h b/generic/include/clc/math/erf.h
new file mode 100644
index 0000000..1aaf689
--- /dev/null
+++ b/generic/include/clc/math/erf.h
@@ -0,0 +1,9 @@
+#undef erfc
Trying without attaching...
gmail wrapped the lines in your patch so it fails to apply. if you used
git then git format-patch and git send-email are your friends.
LGTM, passes piglit test on AMD Turks and Kaveri.
do you need somebody to push it for you?
Jan
The scalar float/double function bodies are a direct copy/paste,
aside from the removed (optional) code in float function body that
requires subnormals.
čet, 5. svi 2016. u 00:08 Jan Vesely <jan.vesely@rutgers.edu> napisao je:
gmail wrapped the lines in your patch so it fails to apply. if you used
git then git format-patch and git send-email are your friends.
Tried, worked. Very nice, thanks!
LGTM, passes piglit test on AMD Turks and Kaveri.
Tested here on Kabini using Piglit, works too.
do you need somebody to push it for you?
Jan
Yes, please.
Regards,
Vedran
čet, 5. svi 2016. u 00:08 Jan Vesely <jan.vesely@rutgers.edu> napisao
je:>
> gmail wrapped the lines in your patch so it fails to apply. if you
> used
> git then git format-patch and git send-email are your friends.
>
>
Tried, worked. Very nice, thanks!>
> LGTM, passes piglit test on AMD Turks and Kaveri.
>
>
Tested here on Kabini using Piglit, works too.>
> do you need somebody to push it for you?
>
>
Jan
>
>
>
Yes, please.
pushed as r268766.
Jan
Thanks!
Vedran