LI78467: INEFFICIENT VEC_SPLATS WHEN LOADING FROM MEMORY

XL C/C++ for Linux Fix Pack 2 (February 2015 Update) for 13.1

APAR status

• Closed as program error.

Error description

• vec_splats for float, int/unsigned int and short/unsigned short
  are inefficient when loading from memory.
  
  Consider the following testcase:
  
  vector float splatFloat(float *in)
  {
     return vec_splats(*in);
  }
  
  xlC generates:
  
       | 000000                           PDEF
  splatFloat(float *)
       | 000000                           AKA
  splatFloat__FPf
      9|                                  PROC      in,gr3
     11| 000020 lfs      C0030000   1     LFS
  vs0=(float)(gr3,0)
     11| 000024 xscvdpsp F0000424   1     XCVDPSP   vs0=vs0
     11| 000028 xxspltw  F0400291   1     VSPLTW    vs34=vs0,0
     12| 00002C bclr     4E800020   1     BA        lr
  
  
  Optimal output can be generated with:
  vector float splatFloat_opt(float *in)
  {
     double d;
     vector float v;
     asm("lfiwzx %0,0,%1":"=f"(d):"r"(in));
     asm("xxspltw %x0,%1,2":"=v"(v):"f"(d));
     return v;
  }
  
       | 000000                           PDEF
  splatFloat_opt(float *)
       | 000000                           AKA
  splatFloat_opt__FPf
     14|                                  PROC      in,gr3
     18| 000040 lfiwzx   7C001EEE   1     ASMFPLS   vs0='lfiwzx
  %0,0,%1,'f r:,1,(float)",gr3
     19| 000044 xxspltw  F0420291   1     ASMFP     vs34='xxspltw
  %x0,%1,2,'v f:,2,vs0
     21| 000048 bclr     4E800020   1     BA        lr
  
  xlC -O2 -q64 -qarch=pwr7 -qaltivec -qlist -c splats.C
  -qtune=pwr7
  

Local fix

• N/A
  

Problem summary

• INEFFICIENT VEC_SPLATS WHEN LOADING FROM MEMORY
  

Problem conclusion

• The problem had been addressed.
  

Temporary fix

Comments

APAR Information

• APAR is sysrouted FROM one or more of the following:

  IV69753

• APAR is sysrouted TO one or more of the following:

Fix information

• Fixed component name

  XL C/C++ FOR LI

• Fixed component ID

  5725C7300

Applicable component levels

• RD10 PSY

     UP