https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116075
Andrew Pinski <pinskia at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Severity|normal |enhancement
Component|target |tree-optimization
Last reconfirmed| |2024-07-24
Assignee|unassigned at gcc dot gnu.org |pinskia at gcc dot
gnu.org
Status|UNCONFIRMED |ASSIGNED
Target|aarch64 |aarch64 riscv
Ever confirmed|0 |1
--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
> _72 = .VEC_SHL_INSERT ({ 0, ... }, 0);
Confirmed.
This could be optimized at the gimple level to just `{0, ...}` I think.
And it shows up in RISCV too:
```
vsetvli a4,zero,e32,m1,ta,ma
vmv.v.i v2,0
csrr a3,vlenb
li a0,32768
srli a3,a3,2
addi a0,a0,-768
lui a5,%hi(in)
vslide1up.vx v1,v2,zero
```
The internal function is a direct function to the optab vec_shl_insert:
```
@cindex @code{vec_shl_insert_@var{m}} instruction pattern
@item @samp{vec_shl_insert_@var{m}}
Shift the elements in vector input operand 1 left one element (i.e.@:
away from element 0) and fill the vacated element 0 with the scalar
in operand 2. Store the result in vector output operand 0. Operands
0 and 1 have mode @var{m} and operand 2 has the mode appropriate for
one element of @var{m}.
```
So if we have 0 for both then the result is also 0.
So this should fold it:
```
(simplify
(INF_VEC_SHL_INSERT @1 uniform_vector_p@2)
(if (operands_equal (@1, uniform_vector_p(@2))
@2))
```
That is if we are inserting a value that is the same as the uniform vector we
can just simplify it to the uniform vector.
I will take care of this.
