[Bug tree-optimization/61301] New: missed optimization of move if vector passed by reference

[vincenzo.innocente at cern dot ch]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61301

            Bug ID: 61301
           Summary: missed optimization of move if vector passed by
                    reference
           Product: gcc
           Version: 4.9.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: vincenzo.innocente at cern dot ch

in the following test shuffle2 generates not optimized moves.
the other two are ok.
the problem occurs in real life when the vector is a data member of a class
and the function is a method, as in "foo"

typedef float __attribute__( ( vector_size( 16 ) ) ) float32x4_t;
typedef int   __attribute__( ( vector_size( 16 ) ) ) int32x4_t;


float32x4_t shuffle1(float32x4_t x) {
   return float32x4_t{x[1],x[0],x[3],x[2]};
}


float32x4_t shuffle2(float32x4_t const & x) {
   return float32x4_t{x[1],x[0],x[3],x[2]};
}

float32x4_t shuffle3(float32x4_t const & x) {
   return __builtin_shuffle(x,int32x4_t{1,0,3,2});
}

struct foo {
  float32x4_t x;
  float32x4_t shuffle2() const;
  float32x4_t shuffle3() const;
};

float32x4_t foo::shuffle2() const {
  return float32x4_t{x[1],x[0],x[3],x[2]};
}
float32x4_t foo::shuffle3() const {
   return __builtin_shuffle(x,int32x4_t{1,0,3,2});
}


compiled with

c++ -std=c++11 -Ofast -march=nehalem  -S shuffle.cc; cat shuffle.s
generates:
__Z8shuffle1U8__vectorf:
LFB0:
    shufps    $177, %xmm0, %xmm0
    ret

__Z8shuffle2RKU8__vectorf:
LFB1:
    movss    12(%rdi), %xmm1
    insertps    $0x10, 8(%rdi), %xmm1
    movss    4(%rdi), %xmm0
    insertps    $0x10, (%rdi), %xmm0
    movlhps    %xmm1, %xmm0
    ret

__Z8shuffle3RKU8__vectorf:
LFB2:
    movaps    (%rdi), %xmm0
    shufps    $177, %xmm0, %xmm0
    ret