aaron/ispc
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

added fast approximate rcp(double) accurate to 15 digits

Browse Source
This commit is contained in:
Evghenii
2014-02-04 15:23:34 +01:00
parent eb1a495a7a
commit fe98fe8cdc
3 changed files with 34 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
builtins/util.m4
View File

@@ -4548,18 +4548,16 @@ define <WIDTH x double> @__rsqrt_varying_double(<WIDTH x double>, <WIDTH x MASK
')
define(`rcp_double', `
define double @__rcp_uniform_double(double, <WIDTH x MASK>) nounwind alwaysinline readnone
declare double @__rcp_safe_uniform_double___und(double, <WIDTH x MASK>)
define double @__rcp_uniform_double(double, <WIDTH x MASK>) nounwind alwaysinline readnone
{
%flt = fptrunc double %0 to float
%res = call float @__rcp_uniform_float(float %flt)
%dres = fpext float %res to double
ret double %dres
%res = call double @__rcp_safe_uniform_double___und(double %0, <WIDTH x MASK> %1)
ret double %res
}
define <WIDTH x double> @__rcp_varying_double(<WIDTH x double>, <WIDTH x MASK>) nounwind alwaysinline readnone
declare <WIDTH x double> @__rcp_safe_varying_double___vyd(<WIDTH x double>, <WIDTH x MASK>)
define <WIDTH x double> @__rcp_varying_double(<WIDTH x double>, <WIDTH x MASK>) nounwind alwaysinline readnone
{
%flt = fptrunc <WIDTH x double> %0 to <WIDTH x float>
%res = call <WIDTH x float> @__rcp_varying_float(<WIDTH x float> %flt)
%dres = fpext <WIDTH x float> %res to <WIDTH x double>
ret <WIDTH x double> %dres
%res = call <WIDTH x double> @__rcp_safe_varying_double___vyd(<WIDTH x double> %0, <WIDTH x MASK> %1)
ret <WIDTH x double> %res
}
')