如何知道循环是否具有优化潜力

Question

我的代码在以下循环中花费了相当多的时间。我使用gcc-4.8和-O3 -march=native来编译代码。由于我是优化的绝对新手，我怎么知道编译器是否做到了所有这一切？我正在使用AMD FX(tm)-6200

 float* __restrict__ ApsiPtr = Apsi.begin();
 const float* const __restrict__ psiPtr = psi.begin();
 const float* const __restrict__ diagPtr = diag().begin();
 register const label nCells = diag().size();

 for (register label cell=0; cell<nCells; cell++) {
     ApsiPtr[cell] = diagPtr[cell]*psiPtr[cell];
 }

ddd转储以下汇编代码。

     Dump of assembler code from 0x7ffff1d32010 to 0x7ffff1d32110:¬
=>        mov    (%rsp),%rdi¬
          callq  0x7ffff1ba5ca0 <_ZNK4Foam9lduMatrix4diagEv@plt>¬
          mov    0x8(%rax),%edx¬
          test   %edx,%edx¬
          mov    %edx,%esi¬
          mov    %edx,0x34(%rsp)¬
          jle    0x7ffff1d3251f ¬
          mov    0x38(%rsp),%r10¬
          lea    0x10(%rbx),%r8¬
          lea    0x10(%rbp),%rax¬
          lea    0x10(%r10),%rdi¬
          cmp    %rdi,%rbx¬
          setae  %r9b¬
          cmp    %r8,%r10¬
          setae  %r11b¬
          or     %r11d,%r9d¬
          cmp    %rax,%rbx¬
          setae  %dl¬
          cmp    %r8,%rbp¬
          setae  %cl¬
          or     %ecx,%edx¬
          test   %dl,%r9b¬
          je     0x7ffff1d32728 ¬
          cmp    $0xc,%esi¬
          jbe    0x7ffff1d32728 ¬
          shr    %esi¬
          lea    0x40(%r10),%rax¬
          mov    %rbx,%rdx¬
          lea    -0x5(%rsi),%r8d¬
          lea    (%rsi,%rsi,1),%r9d¬
          mov    %esi,0x38(%rsp)¬
          shr    $0x2,%r8d¬
          mov    %r9d,0x4c(%rsp)¬
          mov    %rbp,%rsi¬
          shl    $0x6,%r8¬
          mov    $0x0,%r9d¬
          lea    0x80(%r10,%r8,1),%r11¬
          mov    %r11,%rcx¬
          sub    %rax,%rcx¬
          and    $0x40,%ecx¬
          movupd -0x40(%rax),%xmm0¬
          movupd 0x0(%rbp),%xmm1¬
          prefetcht0 0x1e0(%r10)¬
          prefetcht0 0x1e0(%rbp)¬
          prefetchw 0x1e0(%rbx)¬
          mov    $0x4,%r9d¬
          mulpd  %xmm1,%xmm0¬
          lea    0x40(%rbp),%rsi¬
          lea    0x40(%rbx),%rdx¬
          mov    %r10,0x40(%rsp)¬
          movlpd %xmm0,(%rbx)¬
          movhpd %xmm0,0x8(%rbx)¬
          movupd -0x30(%rax),%xmm2¬
          movupd 0x10(%rbp),%xmm3¬
          mulpd  %xmm3,%xmm2¬
          movlpd %xmm2,0x10(%rbx)¬
          movhpd %xmm2,0x18(%rbx)¬
          movupd -0x20(%rax),%xmm4¬
          movupd 0x20(%rbp),%xmm5¬
End of assembler dump.¬