https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113458
Bug ID: 113458
Summary: Missed SLP for reduction of multiplication/addition
with promotion
Product: gcc
Version: 14.0
Status: UNCONFIRMED
Keywords: missed-optimization
Severity: enhancement
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: pinskia at gcc dot gnu.org
Target Milestone: ---
Target: aarch64-*-*
Take:
```
int f(short *a, signed char *b)
{
int sum = 0;
sum += a[0]*b[0];
sum += a[1]*b[1];
sum += a[2]*b[2];
sum += a[3]*b[3];
return sum;
}
```
This is not SLPed with GCC.
With `-fno-vect-cost-model` it is but in a very inefficient way.
LLVM produces:
```
ldr s0, [x1]
ldr d1, [x0]
sshll v0.8h, v0.8b, #0 // promote to short
smull v0.4s, v0.4h, v1.4h //multiply 2 shorts to ints
addv s0, v0.4s // do the reduction
fmov w0, s0
```
Which GCC should be to produce this too.
