[Valgrind-users] [Callgrind] jcnd in char array initialization

[Skarakis, K. <kon...@un...>]

Hello,

I am running callgrind with --collect-jumps=yes. I want to parse the callgrind out-files to generate coverage reports. I am getting many conditional jumps that I don't understand though. For instance this line is reported as a conditional jump:

      char host[USC_STRING_LENGTH]="";

The callgrind out-file section for this line is:

0x406460 46 1
0x40646b 46 1
0x406472 46 1
0x406479 46 1
0x406484 46 1
0x406489 46 1
0x40648c 46 1
0x40648e 46 1
0x406499 46 1
0x4064a0 46 1
0x4064a7 46 1
0x4064ae 46 1
jcnd=1/1 0x4064ae 46
0x4064ae 46
0x4064ae 46 15
jcnd=14/15 0x4064ae 46
0x4064ae 46


In source view in QCachegrind it suppresses this jump (I read it does that for all "same-line" or "next-line" jumps) but in assembly view it confirms this as a jump:

40 6460     48 c7 85 90 fe ff ff    movq  $0x0,-0x170(%rbp)
40 6467     00 00 00 00
40 646B     48 8d 85 98 fe ff ff    lea   -0x168(%rbp),%rax
40 6472     48 89 85 b0 48 ff ff    mov   %rax,-0xb750(%rbp)
40 6479     48 c7 85 a8 48 ff ff    movq  $0x0,-0xb758(%rbp)
40 6480     00 00 00 00
40 6484     b8 78 00 00 00          mov   $0x78,%eax
40 6489     83 f8 08          cmp   $0x8,%eax
40 648C     72 23             jb    <main+0x207>
40 648E     48 c7 85 a0 48 ff ff    movq  $0xf,-0xb760(%rbp)
40 6495     0f 00 00 00
40 6499     48 8b bd b0 48 ff ff    mov   -0xb750(%rbp),%rdi
40 64A0     48 8b 8d a0 48 ff ff    mov   -0xb760(%rbp),%rcx
40 64A7     48 8b 85 a8 48 ff ff    mov   -0xb758(%rbp),%rax
40 64AE     ┌->f3 48 ab       rep   stos %rax,%es:(%rdi)
            └-                      Jump 15 of 16 times to 0x4064AE

(I hope the ASCII comes out readable)

Can anyone explain why this is a conditional jump?

The source line mentioned above is in the "main" function in file UscFtpLinkTest.cc which includes UscConfiguration.h which includes UscCommon.h which defines the constant as:

int USC_STRING_LENGTH = 128;


I am using Benjamin Meyer's tool (http://benjamin-meyer.blogspot.de/2007/12/valgrind-callgrind-tools-part-3-code.html) to parse the callgrind files and generate coverage reports. It does have a serious bug though with jump coverage calculations and I was wondering if this has anything to do with it.


Thanks in advance,
Costas Skarakis