[Bug target/104412] New: [Aarch64] Failure to optimize vector initialization from int64s

[Bug target/104412] New: [Aarch64] Failure to optimize vector initialization from int64s

[gabravier at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104412

            Bug ID: 104412
           Summary: [Aarch64] Failure to optimize vector initialization
                    from int64s
           Product: gcc
           Version: 12.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: gabravier at gmail dot com
  Target Milestone: ---

#include <stdint.h>

typedef int64_t v2i64 __attribute__((vector_size(16)));

v2i64 _mm_set_epi64x(int64_t i1, int64_t i2)
{
    union {
        int64_t data[2];
        v2i64 v;
    } d = {.data = {i2, i1}};
    return d.v;
}

With -O3, AArch64 GCC outputs this:

_mm_set_epi64x(long, long):
  sub sp, sp, #16
  stp x1, x0, [sp]
  ldr q0, [sp]
  add sp, sp, 16
  ret

LLVM instead outputs this:

_mm_set_epi64x(long, long):
  fmov d0, x1
  mov v0.d[1], x0
  ret

[Bug tree-optimization/104412] [Aarch64] Failure to optimize vector initialization from int64s

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104412

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|normal                      |enhancement
          Component|target                      |tree-optimization

[Bug target/104412] union initialization of a vector is not optimized due to vector cost model

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104412

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|tree-optimization           |target
   Last reconfirmed|                            |2022-02-06
            Summary|[Aarch64] Failure to        |union initialization of a
                   |optimize vector             |vector is not optimized due
                   |initialization from int64s  |to vector cost model
     Ever confirmed|0                           |1
             Status|UNCONFIRMED                 |NEW

[Bug target/104412] union initialization of a vector is not optimized due to vector cost model

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104412

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Adding -fno-vect-cost-model allows it to be optimized. So a cost model issue
....

[Bug target/104412] union initialization of a vector is not optimized due to vector cost model

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104412

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
On x86_64 we get

        movq    %rsi, %xmm0
        movq    %rdi, %xmm1
        punpcklqdq      %xmm1, %xmm0
        ret

> ./cc1 -quiet t.c -I include -O2 -fopt-info-vec 
t.c:10:7: optimized: basic block part vectorized using 16 byte vectors

and at -O1

        movq    %rsi, -24(%rsp)
        movq    %rdi, -16(%rsp)
        movdqa  -24(%rsp), %xmm0
        ret

costing is a bit difficult since we get

t.c:10:7: note: Cost model analysis:
i2_4(D) 1 times scalar_store costs 12 in body
i1_6(D) 1 times scalar_store costs 12 in body
i2_4(D) 1 times vector_store costs 12 in body
<unknown> 1 times vec_construct costs 8 in prologue
t.c:10:7: note: Cost model analysis for part in loop 0:
  Vector cost: 20
  Scalar cost: 24

as we do not have an idea how costly the construction is (depends on
calling conventions) or how the return d.v allows us to elide the store
and the load.