LI77244: VECTOR4DOUBLE INCORRECTLY PASSED BY VALUE

May 2013 Update for XL C/C++ for Blue Gene/Q, V12.1

APAR status

• Closed as program error.

Error description

• Consider the following testcase:
  
  /***************** test.c **************************/
  #include "stdlib.h"
  #include "stdio.h"
  
  #define DLEN 1024
  
  typedef struct { vector4double c1, c2, c3; } s12_t;
  
  s12_t gs;
  
  double gd[DLEN];
  
  void init_gd() {
    int i;
    double d=1.0;
    for(i=0; i<DLEN; i++) {
      gd[i] = d;
      d += .001;
    }
  }
  
  void show_s12v(char* str, s12_t v) {
    printf("%s: c1[0] ... = %f %f %f %f\n", str, v.c1[0],
  v.c1[1], v.c1[2], v.c1[3]);
    printf("%s: c2[0] ... = %f %f %f %f\n", str, v.c2[0],
  v.c2[1], v.c2[2], v.c2[3]);
    printf("%s: c3[0] ... = %f %f %f %f\n", str, v.c3[0],
  v.c3[1], v.c3[2], v.c3[3]);
  }
  
  int main(void){
    double *p;
  
    init_gd();
    p = gd;
  
    /* Show original values */
    printf("gd[0] ... = %f %f %f %f\n",gd[0], gd[1], gd[2],
  gd[3]);
    printf("gd[4] ... = %f %f %f %f\n",gd[4], gd[5], gd[6],
  gd[7]);
    printf("gd[8] ... = %f %f %f %f\n",gd[8], gd[9], gd[10],
  gd[11]);
  
    /* Assign gs */
    gs.c1 = vec_ld(0L, &gd[0]);
    gs.c2 = vec_ld(0L, &gd[4]);
    gs.c3 = vec_ld(0L, &gd[8]);
  
    /* Show gs */
    p=(double*)&gs;
    printf("gs.c1[0] ... = %f %f %f %f\n",p[0], p[1], p[2], p[3])
    printf("gs.c2[0] ... = %f %f %f %f\n",p[4], p[5], p[6], p[7])
    printf("gs.c3[0] ... = %f %f %f %f\n",p[8], p[9], p[10],
  p[11]);
  
  
    /* Pass gs by value */
    show_s12v("gs passed by val",gs);
  
    return 0;
  }
  /************** end of test.c **************************/
  
  Build the testcase binary using the command:
  
  mpixlc -qstrict -qarch=qp -qtune=qp test.c
  
  
  Compiler displays incorrect values of the elements of structure
  of quad vector literal vector4double's data type passed by
  value:
  
  gd[0] ... = 1.000000 1.001000 1.002000 1.003000
  gd[4] ... = 1.004000 1.005000 1.006000 1.007000
  gd[8] ... = 1.008000 1.009000 1.010000 1.011000
  gs.c1[0] ... = 1.000000 1.001000 1.002000 1.003000
  gs.c2[0] ... = 1.004000 1.005000 1.006000 1.007000
  gs.c3[0] ... = 1.008000 1.009000 1.010000 1.011000
  gs passed by val: c1[0] ... = 1.003000 1.004000 1.005000
  1.011000
  gs passed by val: c2[0] ... = 0.000000 0.000000 0.000000
  1.006000
  gs passed by val: c3[0] ... = 1.007000 1.008000 1.009000
  1.010000
  
  LOCAL FIX:
  N/A
  

Local fix

• N/A
  

Problem summary

• aggregates containing vectors not correctly passed by value
  

Problem conclusion

• There were 2 problems.  Aggregates are passed in gprs. The
  called procedure was incorrect in determining the starting
  register when the extra alignment requirements of vector needed
  to be taken into account. The alignment also needed to be taken
  into account when determining the location on the stack of the
  parameter structure.
  

Temporary fix

Comments

APAR Information

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

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

Fix information

• Fixed component name

  XL C/C++ FOR BG

• Fixed component ID

  5799AG100

Applicable component levels

• RC10 PSY

     UP