https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61292

            Bug ID: 61292
           Summary: auto keyword to vector reference generates wrong
                    alignment move
           Product: gcc
           Version: 4.9.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: vincenzo.innocente at cern dot ch

given
typedef float __attribute__( ( vector_size( 16 ) ) ) float32x4_t;
typedef float __attribute__( ( vector_size( 16 ) , aligned(4) ) )
float32x4a4_t;


void add11(float * x, float y, float32x4_t v) {
   auto & k1 = *(float32x4a4_t*)(x);
   auto & k2 = *(float32x4a4_t*)(x);
   k1 +=v;
   k2 += k1+v;
}

void add98(float * x, float y, float32x4_t v) {
   float32x4a4_t & k1 = *(float32x4a4_t*)(x);
   float32x4a4_t & k2 = *(float32x4a4_t*)(x);
   k1 +=v;
   k2 += k1+v;
}

I get
c++ -std=c++1y -Ofast -march=haswell  -S extNotAlign.cc; cat extNotAlign.s
__Z5add11PffU8__vectorf:
LFB4:
    vaddps    (%rdi), %xmm1, %xmm0
    vaddps    %xmm0, %xmm0, %xmm0
    vaddps    %xmm1, %xmm0, %xmm1
    vmovaps    %xmm1, (%rdi)
    ret
LFE4:
    .section __TEXT,__text_cold,regular,pure_instructions
LCOLDE4:
    .text
LHOTE4:
    .section __TEXT,__text_cold,regular,pure_instructions
LCOLDB5:
    .text
LHOTB5:
    .align 4,0x90
    .globl __Z5add98PffU8__vectorf
__Z5add98PffU8__vectorf:
LFB5:
    vaddps    (%rdi), %xmm1, %xmm0
    vaddps    %xmm0, %xmm0, %xmm0
    vaddps    %xmm1, %xmm0, %xmm1
    vmovups    %xmm1, (%rdi)
    ret

second of course is correct (vmovups not vmovaps)

Reply via email to