https://gcc.gnu.org/bugzilla/show_bug.cgi?id=125193
Bug ID: 125193
Summary: z13: vec_splats for BFP constant often cause load
Product: gcc
Version: 15.1.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: jens.seifert at de dot ibm.com
Target Milestone: ---
gcc already uses vgm to generate special floating point values e.g.
vec_splats((double)INFINITY). That is great.
The same approach could be used for a wide range of floating point numbers e.g.
0.5,1.0,1.5,...
combine with VECTOR FP PERFORM SIGN
OPERATION this can be extended to
-0.5,-1.0,-1.5
All floating point numbers which have a signed 16-bit integer equivalent can be
constructed like this:
vector double vm10_2()
{
return vec_double(vec_splats(-10LL));
}
v10_opt():
vrepi %v24,10,3
vcdgb %v24,%v24,0,0
br %r14
=> a wide range of values can be constructed using 1-2 instructions.
This is faster than a load + replicate that gets created today.
larl %r5,.L4
vlrepg %v24,.L5-.L4(%r5)
