This is the mail archive of the cygwin mailing list for the Cygwin project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: strange bug in gettimeofday function


On Mon, Feb 19, 2007 at 12:52:10PM -0800, Shankar Unni wrote:
> Andrew Makhorin wrote:
> 
> >{       double t0 = get_time(), t1 = get_time();
> 
> [Maybe OT?]
> 
> 1. I can't remember if C guarantees that comma-separated *declarations* 
> are initialized in order or not..  And to think I used to be an ANSI C 
> guru :-(.

Should be fine in this case.

> 2. The reason that the "t0 > t1" fails, but t0 and t1 get dumped to be 
> the same, is that C allows the implementation to use larger-than-64-bit 
> (for 64-bit) intermediate double representations. In the case of X86, 
> the CPU's floating-point registers are 80 bits wide.
> 
> When they get written to stack, the value is rounded (or truncated?) to 
> 64 bits.

I don't understand why they just didn't write:

double t0, t1;

t0 = t1 = get_time();

Not everything *has* to be initialized at declaration time.

> 
> In the optimized code, I'll bet you that the two locals (t0 and t1) are 
> kept entirely in registers, at least until the "&t0" and "&t1" calls. So 
> at the point of comparison, it's comparing two 80-bit values, but when 
> you flush them to memory to dump them as integer values, they get 
> truncated to the (same) 64-bit value.

Possible. Consider SSE ops (64-bit vs 80-bit on x87) and use of
fast-math as well.

-cl

--
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple
Problem reports:       http://cygwin.com/problems.html
Documentation:         http://cygwin.com/docs.html
FAQ:                   http://cygwin.com/faq/


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]